Rebuild: fresh AdonisJS 7 API app on Postgres
The rebuild branch starts over (REBUILD.md): the API starter kit, with Postgres instead of SQLite (its own database, DB_SEARCH_PATH for a shared one in development) and the #lib alias. Phase 1's code is imported from the current app module by module. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Daan MeijerandClaude Opus 5.5 committed
1 parent
699395c51f
commit
96f0ade10e
264 files changed
+777
-24132
No files matched your search
@@ -1,170 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
import FileStore from '#services/FileStore'
|
||||
import IngestService from '#services/IngestService'
|
||||
import type { RarInfo } from '#lib/rar/index'
|
||||
import { fileStem, isGenericName, rarReleaseIdentity } from '#lib/release/index'
|
||||
|
||||
/** Stored per file and release; a listing of thousands of files is not useful. */
|
||||
const MAX_ENTRIES = 500
|
||||
|
||||
export interface ArchiveResult {
|
||||
info: RarInfo
|
||||
releaseId?: number
|
||||
releaseRenamed?: string
|
||||
grouped?: boolean
|
||||
}
|
||||
|
||||
/**
|
||||
* Records what a RAR volume's headers say: the listing on the file, the
|
||||
* release it belongs to (from the shared directory or continued entry, when
|
||||
* the file has none yet), and for first volumes the contents and password
|
||||
* flag on the release. See CONCEPT.md (D19, D21).
|
||||
*/
|
||||
class ArchiveService {
|
||||
async record(file: any, binary: any, info: RarInfo): Promise<ArchiveResult> {
|
||||
const { next: _next, ...listing } = info
|
||||
const stored = { ...listing, entries: info.entries.slice(0, MAX_ENTRIES).map(({ stored: _s, dataOffset: _d, ...entry }) => entry) }
|
||||
await FileStore.setDetails(Number(file.id), { archive: stored })
|
||||
|
||||
let releaseId = file.release_id ?? binary?.release_id ?? null
|
||||
let grouped = false
|
||||
const identity = rarReleaseIdentity(info.entries)
|
||||
if (!releaseId && identity) {
|
||||
const ids = await IngestService.upsertReleases(db.connection().getWriteClient(), [
|
||||
{
|
||||
release: identity,
|
||||
poster: binary?.poster ?? file.poster,
|
||||
weakIdentity: true,
|
||||
groups: binary?.groups ?? file.groups ?? [],
|
||||
postedAt: binary?.posted_at ? new Date(binary.posted_at) : undefined,
|
||||
},
|
||||
])
|
||||
releaseId = ids.get(identity.key)!
|
||||
grouped = true
|
||||
if (binary) await db.from('binaries').where('id', binary.id).update({ release_id: releaseId })
|
||||
await db.from('files').where('id', file.id).update({ release_id: releaseId })
|
||||
}
|
||||
if (!releaseId) return { info }
|
||||
|
||||
const release = await db.from('releases').where('id', releaseId).first()
|
||||
const update: Record<string, unknown> = {}
|
||||
if (info.firstVolume !== false) {
|
||||
update.encrypted = info.headersEncrypted || info.entries.some((e) => e.encrypted)
|
||||
if (info.entries.length > 0) update.contents = JSON.stringify(stored.entries)
|
||||
} else if (info.headersEncrypted) {
|
||||
update.encrypted = true
|
||||
}
|
||||
const better = releaseNameFromContents(info)
|
||||
// A spot's title is kept (CONCEPT.md D29).
|
||||
const renamed =
|
||||
release && better && release.source !== 'spot' && (looksRandom(release.name) || release.source === 'post') ? better : undefined
|
||||
if (renamed) update.name = renamed
|
||||
if (Object.keys(update).length > 0) await db.from('releases').where('id', releaseId).update(update)
|
||||
|
||||
if (renamed) logger.info(`Release ${releaseId} renamed "${release.name}" → "${renamed}" from its archive.`)
|
||||
return { info, releaseId, releaseRenamed: renamed, grouped }
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* A release name from an archive listing: the top-level directory shared by
|
||||
* all entries when it reads like a release name, otherwise the stem of the
|
||||
* largest entry with a readable name (e.g. an archive holding just
|
||||
* `Joker.Folie.a.Deux.2024.2160p….nzb`). Samples are skipped.
|
||||
*/
|
||||
export function releaseNameFromContents(info: Pick<RarInfo, 'entries'>): string | undefined {
|
||||
// Hidden entries name nothing: one poster adds the same 22 GB ".1917-….mkv" decoy to every upload (LOG.md 2026-10-06 (2)).
|
||||
const listed = info.entries.filter((e) => !isHidden(e.name))
|
||||
const entries = listed.filter((e) => !e.directory)
|
||||
const tops = new Set(listed.map((e) => e.name.split(/[\\/]/)[0]))
|
||||
if (tops.size === 1 && listed.some((e) => /[\\/]/.test(e.name))) {
|
||||
const [top] = tops
|
||||
if (/[A-Za-z]{2,}.*[._-].*[A-Za-z0-9]{2,}/.test(top) && !looksRandom(top)) return top
|
||||
}
|
||||
const candidates = entries
|
||||
.map((e) => ({ stem: fileStem(e.name.split(/[\\/]/).pop() ?? e.name), size: e.size ?? 0 }))
|
||||
.filter(({ stem }) => !/(^|[._ -])sample([._ -]|$)/i.test(stem))
|
||||
.filter(({ stem }) => /[A-Za-z]{2,}.*[._ -].*[A-Za-z0-9]{2,}/.test(stem) && !looksRandom(stem))
|
||||
.sort((a, b) => b.size - a.size)
|
||||
return candidates[0]?.stem
|
||||
}
|
||||
|
||||
/** An entry in a hidden directory or a hidden file itself (`.x/…`, `dir/.x`). */
|
||||
export function isHidden(name: string): boolean {
|
||||
return name.split(/[\\/]/).some((part) => part.startsWith('.'))
|
||||
}
|
||||
|
||||
/**
|
||||
* Releases named after a hidden archive entry before those were skipped:
|
||||
* renamed from their stored listing (no downloads). Used by
|
||||
* `release:rename --hidden`.
|
||||
*/
|
||||
export async function renameHiddenNamed(): Promise<number> {
|
||||
const releases = await db
|
||||
.from('releases')
|
||||
.where('name', 'like', '.%')
|
||||
.whereNot('source', 'spot')
|
||||
.whereNotNull('contents')
|
||||
.select('id', 'name', 'contents')
|
||||
let renamed = 0
|
||||
for (const release of releases) {
|
||||
const entries: any[] = Array.isArray(release.contents) ? release.contents : []
|
||||
const fromHidden = entries.some((e) => isHidden(e.name) && fileStem(String(e.name).split(/[\\/]/).pop() ?? '') === release.name)
|
||||
if (!fromHidden) continue
|
||||
const better = releaseNameFromContents({ entries })
|
||||
if (!better || better === release.name) continue
|
||||
await db.from('releases').where('id', release.id).update({ name: better })
|
||||
renamed++
|
||||
}
|
||||
return renamed
|
||||
}
|
||||
|
||||
/**
|
||||
* Renames random-named releases from their stored archive listing (no
|
||||
* downloads). Used by `release:rename`.
|
||||
*/
|
||||
export async function renameFromStoredContents(): Promise<number> {
|
||||
const releases = await db
|
||||
.from('releases')
|
||||
.whereNotNull('contents')
|
||||
.whereRaw("jsonb_array_length(contents) > 0")
|
||||
.select('id', 'name', 'source', 'contents')
|
||||
let renamed = 0
|
||||
for (const release of releases) {
|
||||
if (release.source === 'spot' || (!looksRandom(release.name) && release.source !== 'post')) continue
|
||||
const better = releaseNameFromContents({ entries: release.contents })
|
||||
if (!better || better === release.name) continue
|
||||
await db.from('releases').where('id', release.id).update({ name: better })
|
||||
renamed++
|
||||
}
|
||||
return renamed
|
||||
}
|
||||
|
||||
/**
|
||||
* Obfuscated names: one long token of mixed letters and digits, no
|
||||
* separators, and the other generated shapes seen on Usenet (LOG.md
|
||||
* 2026-10-01 (3)). Kept in step with SQL's is_readable_name().
|
||||
*/
|
||||
export function looksRandom(name: string): boolean {
|
||||
// An unknown short extension (e.g. ".bin") says nothing about the name.
|
||||
name = name.replace(/\.[A-Za-z0-9]{1,4}$/, '')
|
||||
// VIDEO_TS, Sample, CD1: real words, but no name.
|
||||
if (isGenericName(name)) return true
|
||||
if (/^[0-9a-f]{16,}$/i.test(name)) return true
|
||||
// 6327d9fa2d6acf3e9a61fb053359d5e7.1790574045 (hex and a Unix time)
|
||||
if (/^[0-9a-f]{16,}\.\d{9,}$/i.test(name)) return true
|
||||
// UUIDs
|
||||
if (/^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i.test(name)) return true
|
||||
// 1650777167 (yEncBin), "83111763 n" (rclone uploads)
|
||||
if (/^\d{6,}( n)?$/.test(name)) return true
|
||||
// prvjr7hapd0i1ny14pz4, and rclone's long base32 names
|
||||
if (/^[a-z0-9]{12,}$/.test(name) && (name.match(/\d/g)?.length ?? 0) >= 2 && (name.match(/[a-z]/g)?.length ?? 0) >= 4) return true
|
||||
// wHJ7s71sAGE: short, but capitals all over the token, not only at word starts
|
||||
if (/^[A-Za-z0-9]{8,15}$/.test(name) && /\d/.test(name) && /[a-z]/.test(name) && (name.slice(1).match(/[A-Z]/g)?.length ?? 0) >= 3) return true
|
||||
if (!/^[A-Za-z0-9]{16,}$/.test(name) || !/[a-z]/.test(name) || !/[A-Z]/.test(name)) return false
|
||||
// Digits, or many capitals without separators (e.g. RmISakytFwXSCEzjVEpaKk).
|
||||
return /\d/.test(name) || (name.match(/[A-Z]/g) ?? []).length >= 6
|
||||
}
|
||||
|
||||
export default new ArchiveService()
|
||||
@@ -1,160 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import env from '#start/env'
|
||||
import NntpService from '#services/NntpService'
|
||||
import QueueService from '#services/QueueService'
|
||||
import { segmentsOf } from '#services/FileStore'
|
||||
import { looksRandom, releaseNameFromContents } from '#services/ArchiveService'
|
||||
import { rarVolume } from '#lib/content/index'
|
||||
import type { RarEntry, RarInfo } from '#lib/rar/index'
|
||||
import { probeWalk, type WalkProbe, type WalkTask } from '#lib/content/probe'
|
||||
|
||||
/** Stored per release, as for a first-volume listing (ArchiveService). */
|
||||
const MAX_ENTRIES = 500
|
||||
|
||||
/** Listed sizes below this share of the release's size: the first segment listed only part of it. */
|
||||
const COVERED = 0.7
|
||||
|
||||
/** Inspection queues a walk this long after the first volume, when the release has had time to complete. */
|
||||
const WALK_DELAY_MS = 30 * 60 * 1000
|
||||
|
||||
export interface WalkOutcome {
|
||||
releaseId: number
|
||||
/** Nothing downloaded: why. */
|
||||
skipped?: 'missing' | 'covered' | 'volumes'
|
||||
entries?: number
|
||||
before?: number
|
||||
complete?: boolean
|
||||
stopped?: string
|
||||
segments?: number
|
||||
renamed?: string
|
||||
}
|
||||
|
||||
/**
|
||||
* Complete archive listings by walking RAR headers (CONCEPT.md D40). The
|
||||
* first segment of a first volume lists the entries that start in it: for
|
||||
* a set of many small files (an audiobook's MP3s, a music album) that is
|
||||
* the first one or two. The walk reads one segment per further entry and
|
||||
* volume, at low priority: its own queue, one job at a time per collection
|
||||
* worker, and NNTP connections only when inspection isn't waiting for one.
|
||||
*/
|
||||
class ArchiveWalkService {
|
||||
/** Segments one walk may download (~0.7 MB each). */
|
||||
readonly maxSegments = env.get('ARCHIVE_WALK_MAX_SEGMENTS', 400)
|
||||
|
||||
/** A listing that leaves most of the release unaccounted for. */
|
||||
worthWalking(entries: number, listedBytes: number, releaseSize: number): boolean {
|
||||
if (entries === 0 || entries >= MAX_ENTRIES) return false
|
||||
return listedBytes < COVERED * releaseSize
|
||||
}
|
||||
|
||||
/** After inspection listed a first RAR volume whose listing ends inside the set. */
|
||||
async queueAfterInspection(releaseId: number | null, info: RarInfo) {
|
||||
if (!releaseId || info.format === '7z' || info.firstVolume === false) return
|
||||
if (!info.truncated || info.headersEncrypted || info.entries.length === 0) return
|
||||
await QueueService.archiveWalkQueue.add('walk', { releaseId }, { jobId: `walk-${releaseId}`, delay: WALK_DELAY_MS })
|
||||
}
|
||||
|
||||
/**
|
||||
* The database half before a walk (D41): the release's volumes, or the
|
||||
* outcome when there's nothing to walk.
|
||||
*/
|
||||
async prepare(releaseId: number): Promise<{ task: WalkTask } | { outcome: WalkOutcome }> {
|
||||
const release = await db.from('releases').where('id', releaseId).select('id', 'size', 'contents').first()
|
||||
if (!release) return { outcome: { releaseId, skipped: 'missing' } }
|
||||
const before = Array.isArray(release.contents) ? release.contents : []
|
||||
const listed = before.reduce((sum: number, e: any) => sum + Number(e.size ?? 0), 0)
|
||||
if (!this.worthWalking(before.length, listed, Number(release.size))) {
|
||||
return { outcome: await this.done(releaseId, { releaseId, skipped: 'covered' }) }
|
||||
}
|
||||
const volumes = await this.volumes(releaseId)
|
||||
if (!volumes) return { outcome: await this.done(releaseId, { releaseId, skipped: 'volumes' }) }
|
||||
return {
|
||||
task: {
|
||||
kind: 'walk',
|
||||
releaseId,
|
||||
volumes: volumes.map((file) => segmentsOf(file).map(({ number, id }) => ({ number, id }))),
|
||||
maxSegments: this.maxSegments,
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
/** The database half after a walk: the fuller listing, the password flag and a better name. */
|
||||
async apply(releaseId: number, walked: WalkProbe): Promise<WalkOutcome> {
|
||||
const release = await db.from('releases').where('id', releaseId).select('id', 'name', 'source', 'contents', 'encrypted').first()
|
||||
if (!release) return { releaseId, skipped: 'missing' }
|
||||
const before = Array.isArray(release.contents) ? release.contents : []
|
||||
const outcome: WalkOutcome = {
|
||||
releaseId,
|
||||
before: before.length,
|
||||
entries: walked.entries.length,
|
||||
complete: walked.complete,
|
||||
stopped: walked.stopped,
|
||||
segments: walked.segments,
|
||||
}
|
||||
if (walked.entries.length > before.length) {
|
||||
const contents = walked.entries.slice(0, MAX_ENTRIES)
|
||||
const update: Record<string, unknown> = {
|
||||
contents: JSON.stringify(contents),
|
||||
encrypted: Boolean(release.encrypted) || walked.headersEncrypted || walked.entries.some((e) => e.encrypted),
|
||||
}
|
||||
const better = releaseNameFromContents({ entries: walked.entries as RarEntry[] })
|
||||
// The rule of ArchiveService.record: random and post-subject names give way, a spot's title stays.
|
||||
if (better && better !== release.name && release.source !== 'spot' && (looksRandom(release.name) || release.source === 'post')) {
|
||||
update.name = better
|
||||
outcome.renamed = better
|
||||
}
|
||||
await db.from('releases').where('id', releaseId).update(update)
|
||||
}
|
||||
return this.done(releaseId, outcome)
|
||||
}
|
||||
|
||||
/** Walks a release here, on the inspection pool's connections when inspection isn't waiting for one. */
|
||||
async walk(releaseId: number): Promise<WalkOutcome> {
|
||||
const prepared = await this.prepare(releaseId)
|
||||
if ('outcome' in prepared) return prepared.outcome
|
||||
return this.apply(releaseId, await probeWalk(prepared.task, fetchLow))
|
||||
}
|
||||
|
||||
private async done(releaseId: number, outcome: WalkOutcome) {
|
||||
await db.from('releases').where('id', releaseId).update({ contents_walked_at: new Date() })
|
||||
return outcome
|
||||
}
|
||||
|
||||
/**
|
||||
* The release's RAR volumes in order, from volume 1 up to the first gap.
|
||||
* Null without a first volume (e.g. volumes with random names).
|
||||
*/
|
||||
private async volumes(releaseId: number): Promise<any[] | null> {
|
||||
const files = await db.from('files').where('release_id', releaseId).select('id', 'filename', 'real_name', 'segments')
|
||||
const sets = new Map<string, Map<number, any>>()
|
||||
for (const file of files) {
|
||||
const volume = rarVolume(String(file.real_name ?? file.filename))
|
||||
if (!volume || segmentsOf(file).length === 0) continue
|
||||
const set = sets.get(volume.base) ?? new Map()
|
||||
if (!set.has(volume.number)) set.set(volume.number, file)
|
||||
sets.set(volume.base, set)
|
||||
}
|
||||
// A release holds one set, now and then a second (a sample, subs): the biggest.
|
||||
const set = [...sets.values()].sort((a, b) => b.size - a.size)[0]
|
||||
if (!set?.has(1)) return null
|
||||
const ordered: any[] = []
|
||||
for (let n = 1; set.has(n); n++) ordered.push(set.get(n))
|
||||
return ordered
|
||||
}
|
||||
}
|
||||
|
||||
/** One article body on a pooled connection, behind inspection's waiting jobs. */
|
||||
async function fetchLow(messageId: string): Promise<Buffer> {
|
||||
const conn = await NntpService.acquire(true)
|
||||
let failed = false
|
||||
try {
|
||||
return await NntpService.timed(NntpService.fetchBody(conn, messageId), 'BODY')
|
||||
} catch (error) {
|
||||
failed = true
|
||||
throw error
|
||||
} finally {
|
||||
NntpService.release(conn, failed)
|
||||
}
|
||||
}
|
||||
|
||||
export default new ArchiveWalkService()
|
||||
@@ -1,112 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
import indexerConfig from '#config/indexer'
|
||||
import NntpService from '#services/NntpService'
|
||||
import FetchRangeService from '#services/FetchRangeService'
|
||||
import { articleAtDate } from '#lib/nntp/date_search'
|
||||
|
||||
/** Articles XOVERed to find a posting date near an article number. */
|
||||
const PROBE = 200n
|
||||
|
||||
/**
|
||||
* Header backfill (CONCEPT.md D32). `start` gives a group a cursor from
|
||||
* where live indexing began back to the article posted N days ago; `tick`
|
||||
* (every scheduler run) queues low-priority, header-only fetch ranges from
|
||||
* that cursor downwards, newest first, while few are waiting and the
|
||||
* database stays within its size limits.
|
||||
*/
|
||||
class BackfillService {
|
||||
/** Sets the group's backfill to reach back `days` days. Returns its range. */
|
||||
async start(groupName: string, days: number): Promise<{ from: bigint; to: bigint } | null> {
|
||||
const group = await db.from('groups').where('name', groupName).first()
|
||||
if (!group) throw new Error(`Unknown group ${groupName}`)
|
||||
const conn = await NntpService.acquire()
|
||||
let failed = false
|
||||
try {
|
||||
const { first, last } = await NntpService.timed(NntpService.selectGroup(conn, groupName), `GROUP ${groupName}`)
|
||||
const dateOf = async (n: bigint) => {
|
||||
const { overviews } = await NntpService.timed<any>(conn.xover(n, n + PROBE), `XOVER ${groupName} ${n}`)
|
||||
const header = overviews.find(([number, h]: [number | null, any]) => number !== null && h?.date)?.[1]
|
||||
const date = header ? Date.parse(header.date) : Number.NaN
|
||||
return Number.isNaN(date) ? null : date
|
||||
}
|
||||
const until = await articleAtDate(first, last, Date.now() - days * 24 * 60 * 60 * 1000, dateOf)
|
||||
// Live indexing began a little before the group was added: continue from there.
|
||||
const began = await articleAtDate(first, last, new Date(group.created_at).getTime(), dateOf)
|
||||
const next = group.backfill_next_id !== null && BigInt(group.backfill_next_id) < began ? BigInt(group.backfill_next_id) : began
|
||||
await db.from('groups').where('id', group.id).update({ backfill_until_id: until.toString(), backfill_next_id: next.toString() })
|
||||
return next > until ? { from: until, to: next } : null
|
||||
} catch (error) {
|
||||
failed = true
|
||||
throw error
|
||||
} finally {
|
||||
NntpService.release(conn, failed)
|
||||
}
|
||||
}
|
||||
|
||||
/** Stops a group's backfill (queued ranges still run). */
|
||||
async stop(groupName: string) {
|
||||
await db.from('groups').where('name', groupName).update({ backfill_until_id: null, backfill_next_id: null })
|
||||
}
|
||||
|
||||
/** Queues the next backfill ranges. Returns how many. */
|
||||
async tick(): Promise<number> {
|
||||
const { maxQueued, maxDefaultGb, maxHddGb } = indexerConfig.backfill
|
||||
const groups = await db
|
||||
.from('groups')
|
||||
.where('active', true)
|
||||
.whereNotNull('backfill_until_id')
|
||||
.whereRaw('backfill_next_id > backfill_until_id')
|
||||
.orderBy('id')
|
||||
if (groups.length === 0) return 0
|
||||
|
||||
const room = await this.room(maxDefaultGb, maxHddGb)
|
||||
if (room !== true) {
|
||||
logger.warn(`Backfill paused: ${room}.`)
|
||||
return 0
|
||||
}
|
||||
const [{ count }] = await db.from('fetch_ranges').where('backfill', true).where('status', 'queued').count('* as count')
|
||||
let free = maxQueued - Number(count)
|
||||
let queued = 0
|
||||
// Round-robin over the groups, one range each per pass.
|
||||
while (free > 0 && groups.length > 0) {
|
||||
for (const group of [...groups]) {
|
||||
if (free <= 0) break
|
||||
const next = BigInt(group.backfill_next_id)
|
||||
const until = BigInt(group.backfill_until_id)
|
||||
const start = next - indexerConfig.batchSize + 1n > until ? next - indexerConfig.batchSize + 1n : until + 1n
|
||||
await FetchRangeService.queue(group, start, next, { backfill: true })
|
||||
group.backfill_next_id = (start - 1n).toString()
|
||||
await db.from('groups').where('id', group.id).update({ backfill_next_id: group.backfill_next_id })
|
||||
free--
|
||||
queued++
|
||||
if (start - 1n <= until) {
|
||||
groups.splice(groups.indexOf(group), 1)
|
||||
logger.info(`${group.name}: backfill fully queued.`)
|
||||
}
|
||||
}
|
||||
}
|
||||
return queued
|
||||
}
|
||||
|
||||
/** True when both parts of the database are under their limits, else why not. */
|
||||
/**
|
||||
* Whether the database has room for more history: the default tablespace
|
||||
* (NVMe) and tablespace `hdd` against BACKFILL_MAX_NVME_GB/_HDD_GB. Seed
|
||||
* leases (CONCEPT.md D38) check the same limits.
|
||||
*/
|
||||
async room(
|
||||
maxDefaultGb = indexerConfig.backfill.maxDefaultGb,
|
||||
maxHddGb = indexerConfig.backfill.maxHddGb
|
||||
): Promise<true | string> {
|
||||
const { rows } = await db.rawQuery(
|
||||
`SELECT spcname, pg_tablespace_size(oid) AS bytes FROM pg_tablespace WHERE spcname IN ('pg_default', 'hdd')`
|
||||
)
|
||||
const gb = (name: string) => Number(rows.find((r: any) => r.spcname === name)?.bytes ?? 0) / 1024 ** 3
|
||||
if (gb('pg_default') > maxDefaultGb) return `database on the NVMe is ${gb('pg_default').toFixed(1)} GB (limit ${maxDefaultGb})`
|
||||
if (gb('hdd') > maxHddGb) return `files table on the HDD is ${gb('hdd').toFixed(1)} GB (limit ${maxHddGb})`
|
||||
return true
|
||||
}
|
||||
}
|
||||
|
||||
export default new BackfillService()
|
||||
@@ -1,144 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import FileStore from '#services/FileStore'
|
||||
import SfvService from '#services/SfvService'
|
||||
import ContentService from '#services/ContentService'
|
||||
import InspectionQueue from '#services/InspectionQueue'
|
||||
|
||||
export interface BinaryFilesResult {
|
||||
/** New files. */
|
||||
saved: number
|
||||
/** Binaries whose file already existed (same first segment or file_key). */
|
||||
existing: number
|
||||
/** Conflicted, gone or already linked: nothing to do. */
|
||||
skipped: number
|
||||
/** Not all parts stored yet (the job is retried for these). */
|
||||
incomplete: number[]
|
||||
inspections: number
|
||||
}
|
||||
|
||||
/**
|
||||
* Turns completed binaries into files (worker:file), a batch at a time
|
||||
* (LOG.md 2026-10-06 (7)): one query each for the binaries, their parts and
|
||||
* files that exist already, one INSERT for the new files (so the files
|
||||
* statement trigger updates each release once), and one UPDATE and DELETE
|
||||
* for the binaries' links and copied parts. It was one job of ~8 queries per
|
||||
* binary, ~700 a minute.
|
||||
*/
|
||||
class BinaryFileService {
|
||||
async store(binaryIds: number[]): Promise<BinaryFilesResult> {
|
||||
const result: BinaryFilesResult = { saved: 0, existing: 0, skipped: 0, incomplete: [], inspections: 0 }
|
||||
const ids = [...new Set(binaryIds)]
|
||||
const binaries = await db
|
||||
.from('binaries')
|
||||
.whereIn('id', ids)
|
||||
.whereNull('conflicted_at')
|
||||
.whereNull('file_id')
|
||||
.select('id', 'file_key', 'name', 'real_name', 'poster', 'posted_at', 'groups', 'release_id', 'total_parts')
|
||||
result.skipped = ids.length - binaries.length
|
||||
if (binaries.length === 0) return result
|
||||
|
||||
const partsOf = new Map<number, { number: number; id: string; size: number }[]>()
|
||||
const parts = await db
|
||||
.from('parts')
|
||||
.whereIn('binary_id', binaries.map((b) => b.id))
|
||||
.orderBy([{ column: 'binary_id' }, { column: 'number' }])
|
||||
.select('binary_id', 'number', 'message_id', 'bytes')
|
||||
for (const part of parts) {
|
||||
const list = partsOf.get(Number(part.binary_id)) ?? []
|
||||
list.push({ number: Number(part.number), id: part.message_id, size: Number(part.bytes) })
|
||||
partsOf.set(Number(part.binary_id), list)
|
||||
}
|
||||
const complete = binaries.filter((b) => {
|
||||
const ok = (partsOf.get(Number(b.id))?.length ?? 0) >= b.total_parts
|
||||
if (!ok) result.incomplete.push(Number(b.id))
|
||||
return ok
|
||||
})
|
||||
if (complete.length === 0) return result
|
||||
|
||||
// Already known as a file, e.g. added from a posted NZB before its own articles completed (id only: the row holds the segment list).
|
||||
const firsts = complete.map((b) => partsOf.get(Number(b.id))![0].id)
|
||||
const byFirst = new Map<string, number>()
|
||||
for (const row of await db.from('files').whereIn('first_message_id', firsts).select('id', 'first_message_id')) {
|
||||
byFirst.set(row.first_message_id, Number(row.id))
|
||||
}
|
||||
const fileIdOf = new Map<number, number>()
|
||||
const fresh = complete.filter((b) => {
|
||||
const existing = byFirst.get(partsOf.get(Number(b.id))![0].id)
|
||||
if (existing) fileIdOf.set(Number(b.id), existing)
|
||||
return !existing
|
||||
})
|
||||
|
||||
const now = new Date()
|
||||
// file_key is unique: a second completion of the same upload inserts nothing.
|
||||
const inserted = await FileStore.insert(
|
||||
fresh.map((b) => {
|
||||
const segments = partsOf.get(Number(b.id))!
|
||||
return {
|
||||
file_key: b.file_key,
|
||||
filename: b.name,
|
||||
poster: b.poster,
|
||||
date: b.posted_at ? new Date(b.posted_at).getTime() : now.getTime(),
|
||||
parts: segments.length,
|
||||
segments,
|
||||
groups: JSON.stringify(b.groups),
|
||||
release_id: b.release_id,
|
||||
real_name: b.real_name,
|
||||
size: segments.reduce((sum, s) => sum + s.size, 0),
|
||||
created_at: now,
|
||||
updated_at: now,
|
||||
}
|
||||
})
|
||||
)
|
||||
const insertedByKey = new Map(inserted.map((row) => [row.file_key, row.id]))
|
||||
const missing = fresh.filter((b) => !insertedByKey.has(b.file_key)).map((b) => b.file_key)
|
||||
const byKey = new Map<string, number>()
|
||||
if (missing.length > 0) {
|
||||
for (const row of await db.from('files').whereIn('file_key', missing).select('id', 'file_key')) byKey.set(row.file_key, Number(row.id))
|
||||
}
|
||||
for (const b of fresh) {
|
||||
const id = insertedByKey.get(b.file_key) ?? byKey.get(b.file_key)
|
||||
if (id) fileIdOf.set(Number(b.id), id)
|
||||
}
|
||||
const fresher = new Map(fresh.map((b) => [b.file_key, b]))
|
||||
result.saved = inserted.length
|
||||
result.existing = fileIdOf.size - inserted.length
|
||||
|
||||
// Links, locked in file_key order like ingest's upsert, so the two can't deadlock.
|
||||
const linked = complete.filter((b) => fileIdOf.has(Number(b.id))).sort((a, b) => a.file_key.localeCompare(b.file_key))
|
||||
if (linked.length > 0) {
|
||||
await db.transaction(async (trx) => {
|
||||
await trx.rawQuery(
|
||||
`SELECT id FROM binaries WHERE id IN (SELECT unnest(?::bigint[])) ORDER BY file_key FOR UPDATE`,
|
||||
[linked.map((b) => b.id)]
|
||||
)
|
||||
// Files queued for inspection below keep their binary until it has its fingerprint (par2 naming, D16).
|
||||
const queued = new Set(inserted.filter((row) => row.file_key && ContentService.isCandidate(String(fresher.get(row.file_key)?.name ?? ''))).map((row) => row.id))
|
||||
await trx.rawQuery(
|
||||
`UPDATE binaries b SET file_id = v.file_id, awaiting_inspection = CASE WHEN v.queued THEN true ELSE b.awaiting_inspection END
|
||||
FROM unnest(?::bigint[], ?::bigint[], ?::boolean[]) AS v(id, file_id, queued) WHERE b.id = v.id`,
|
||||
[linked.map((b) => b.id), linked.map((b) => fileIdOf.get(Number(b.id))!), linked.map((b) => queued.has(fileIdOf.get(Number(b.id))!))]
|
||||
)
|
||||
})
|
||||
// Header-derived parts are in the file's segment list now; body-derived ones (with a CRC) stay for split checks (D20).
|
||||
await db.rawQuery(`DELETE FROM parts WHERE binary_id = ANY(?::bigint[]) AND pcrc32 IS NULL`, [linked.map((b) => b.id)])
|
||||
}
|
||||
|
||||
// As for one file before: an SFV posted earlier may list them (D34), then inspection.
|
||||
const added = inserted.flatMap((row) => {
|
||||
const b = row.file_key ? fresher.get(row.file_key) : undefined
|
||||
return b
|
||||
? [{ id: row.id, name: b.real_name ?? b.name, posted: String(b.name), poster: b.poster, date: b.posted_at ? new Date(b.posted_at).getTime() : null, release_id: b.release_id }]
|
||||
: []
|
||||
})
|
||||
if (added.length > 0) await SfvService.claimAll(added.map(({ posted: _p, ...file }) => file))
|
||||
// By the posted name, as before: inspection skips later volumes and readable media names anyway.
|
||||
const jobs = added
|
||||
.filter((file) => ContentService.isCandidate(file.posted))
|
||||
.map((file) => file.id)
|
||||
if (jobs.length > 0) await InspectionQueue.add(jobs)
|
||||
result.inspections = jobs.length
|
||||
return result
|
||||
}
|
||||
}
|
||||
|
||||
export default new BinaryFileService()
|
||||
@@ -1,321 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
import NntpService from '#services/NntpService'
|
||||
import Par2Service, { type Par2Result } from '#services/Par2Service'
|
||||
import ArchiveService, { looksRandom, type ArchiveResult } from '#services/ArchiveService'
|
||||
import NameService from '#services/NameService'
|
||||
import { decodeSegment } from '#lib/yenc/index'
|
||||
import NzbImportService, { type NzbImportResult } from '#services/NzbImportService'
|
||||
import {
|
||||
isArchiveStart,
|
||||
isLaterVolume,
|
||||
NICOVIDEO_FILE,
|
||||
NICOVIDEO_RELEASE,
|
||||
nicovideoReleaseName,
|
||||
type ContentType,
|
||||
type NicovideoInfo,
|
||||
} from '#lib/content/index'
|
||||
import { chunkMd5, probeInspection, type InspectProbe, type InspectTask, type TaskSegment } from '#lib/content/probe'
|
||||
import type { Segment } from '#lib/nzb/index'
|
||||
import { fileStem, releaseIdentity } from '#lib/release/index'
|
||||
import IngestService from '#services/IngestService'
|
||||
import { looksLikeFilename } from '#lib/subject/index'
|
||||
import { sevenZipVolume } from '#lib/sevenzip/index'
|
||||
import SevenZipService from '#services/SevenZipService'
|
||||
import SfvService from '#services/SfvService'
|
||||
import ArchiveWalkService from '#services/ArchiveWalkService'
|
||||
import FileStore, { segmentsOf } from '#services/FileStore'
|
||||
import GarbageFilter from '#services/GarbageFilter'
|
||||
|
||||
/**
|
||||
* MD5 of the first 16 KiB of a file, as par2 records it, from the body of
|
||||
* the file's first part. Undefined when that part is shorter than 16 KiB but
|
||||
* the file is not (the chunk would span two parts).
|
||||
*/
|
||||
export function firstChunkMd5(body: Buffer, fileSize: number): string | undefined {
|
||||
return chunkMd5(decodeSegment(body).data, fileSize)
|
||||
}
|
||||
|
||||
export interface ContentResult {
|
||||
par2?: Par2Result
|
||||
archive?: ArchiveResult
|
||||
named: number
|
||||
contentType: ContentType | null
|
||||
nzb?: NzbImportResult
|
||||
/** NZBs read from inside an archive. */
|
||||
archiveNzbs?: NzbImportResult[]
|
||||
/** Files an SFV's listing moved into its release. */
|
||||
sfvMoved?: number
|
||||
/** A niconico capture's release name from its API response (D34). */
|
||||
nicovideo?: string
|
||||
/** The real name from the yEnc header, for a file posted under a random subject (D36). */
|
||||
yencName?: string
|
||||
/** Why the file's release was deleted as garbage after this probe (D42). */
|
||||
garbage?: string
|
||||
}
|
||||
|
||||
/**
|
||||
* Takes names and groupings from file contents (CONCEPT.md D21). For a
|
||||
* completed file with a random name, one article (the first) is enough:
|
||||
* - its fingerprint (size + first-16-KiB MD5) matches it to a par2 entry,
|
||||
* - par2 files are recognised by content and name every file of their set,
|
||||
* - RAR headers give the release grouping, name and contents.
|
||||
* Inspection is prepare (database) → probe (downloads and parsing,
|
||||
* lib/content/probe; here or on a seeding node, D41) → apply (database).
|
||||
*/
|
||||
class ContentService {
|
||||
/** Random or missing names, and named first RAR volumes (for contents and password flag). */
|
||||
isCandidate(name: string) {
|
||||
// Even under a random stem: the first volume and the index .par2 are inspected instead.
|
||||
if (isLaterVolume(name)) return false
|
||||
if (!looksLikeFilename(name) || looksRandom(fileStem(name))) return true
|
||||
if (isArchiveStart(name)) return true
|
||||
if (/\.nzb$/i.test(name)) return true
|
||||
// SFVs name the files of their release; niconico captures carry their title (D34).
|
||||
if (/\.sfv$/i.test(name) || NICOVIDEO_FILE.test(name)) return true
|
||||
// Every 7z volume: the set is listed when its last volume is in (see inspect).
|
||||
if (sevenZipVolume(name) !== null) return true
|
||||
return /\.part0*1\.rar$/i.test(name) || (/\.rar$/i.test(name) && !/\.part\d+\.rar$/i.test(name))
|
||||
}
|
||||
|
||||
/**
|
||||
* Uploads that randomise the subject per article often keep the real file
|
||||
* name in the yEnc header inside the body (`Deep.Strike.S01E16.1080p….part01.rar`;
|
||||
* LOG.md 2026-10-02 (2)). A file with a random name takes that name, and
|
||||
* leaves its weak one-file release for the release its name gives, where
|
||||
* the other files of the upload gather. Returns the name, or undefined.
|
||||
*/
|
||||
private async nameFromYenc(file: any, binary: any, yencName: string): Promise<string | undefined> {
|
||||
const current = String(file.real_name ?? file.filename)
|
||||
if (file.real_name || !looksRandom(fileStem(current))) return undefined
|
||||
if (!looksLikeFilename(yencName) || looksRandom(fileStem(yencName)) || yencName === current) return undefined
|
||||
|
||||
const release = file.release_id ? await db.from('releases').where('id', file.release_id).first() : null
|
||||
const weak = !release || (['stem', 'weak-stem'].includes(release.source) && Number(release.file_count) <= 1)
|
||||
let releaseId = file.release_id
|
||||
if (weak) {
|
||||
const identity = releaseIdentity({ poster: file.poster, subject: '', name: yencName, weak: true })
|
||||
if (identity) {
|
||||
const ids = await IngestService.upsertReleases(db.connection().getWriteClient(), [
|
||||
{ release: identity, poster: file.poster, groups: file.groups ?? [], postedAt: file.date ? new Date(Number(file.date)) : undefined, weakIdentity: true },
|
||||
])
|
||||
releaseId = ids.get(identity.key) ?? releaseId
|
||||
}
|
||||
}
|
||||
await db.from('files').where('id', file.id).update({ real_name: yencName, release_id: releaseId })
|
||||
if (binary) await db.from('binaries').where('id', binary.id).update({ real_name: yencName, release_id: releaseId })
|
||||
return yencName
|
||||
}
|
||||
|
||||
/**
|
||||
* Names a niconico capture's release after the video, while it still has
|
||||
* the archiver's `<time>_nicovideo_jp_watch_sm<id>` name; its tags join
|
||||
* the search text.
|
||||
*/
|
||||
private async nameNicovideo(file: any, info: NicovideoInfo, id: string): Promise<string | undefined> {
|
||||
if (!file.release_id) return undefined
|
||||
const release = await db.from('releases').where('id', file.release_id).first()
|
||||
if (!release || !NICOVIDEO_RELEASE.test(release.name)) return undefined
|
||||
const name = nicovideoReleaseName(info, id)
|
||||
await db
|
||||
.from('releases')
|
||||
.where('id', release.id)
|
||||
.update({ name, search_files: info.tags.join(' ').slice(0, 4000) || null })
|
||||
return name
|
||||
}
|
||||
|
||||
/**
|
||||
* The database half before a probe (CONCEPT.md D41): what to inspect for
|
||||
* a file, or null when there's nothing to do. A later 7z volume becomes its
|
||||
* archive's first volume, once all volumes are in.
|
||||
*/
|
||||
async prepare(fileId: number): Promise<InspectTask | null> {
|
||||
const file = await db.from('files').where('id', fileId).first()
|
||||
if (!file || !this.isCandidate(file.filename)) return null
|
||||
// Stored before its signature was found to be garbage (D42): not worth an article.
|
||||
if (await GarbageFilter.skipProbe(file)) return null
|
||||
if ((sevenZipVolume(String(file.real_name ?? file.filename)) ?? 0) > 1) {
|
||||
const volumes = await SevenZipService.volumes(file)
|
||||
if (!volumes || (await FileStore.details(Number(volumes[0].id))).archive) return null
|
||||
return this.prepare(Number(volumes[0].id))
|
||||
}
|
||||
const segments = segmentsOf(file)
|
||||
if (!segments[0]?.id) return null
|
||||
// A 7z listing reads the last volume's tail (sniffing decides whether it's 7z: random names too).
|
||||
const volumes = await SevenZipService.volumes(file)
|
||||
const last = volumes && volumes.length > 1 ? segmentsOf(volumes[volumes.length - 1]) : segments
|
||||
return {
|
||||
kind: 'inspect',
|
||||
fileId: Number(file.id),
|
||||
name: String(file.real_name ?? file.filename),
|
||||
segments: taskSegments(segments),
|
||||
sevenZip: volumes ? { volumes: volumes.length, last: taskSegments(last, true) } : null,
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* The database half after a probe: names, groupings, listings and flags from
|
||||
* what it found. Then the release is deleted if it turned out to be garbage (D42).
|
||||
*/
|
||||
async apply(task: InspectTask, probe: InspectProbe): Promise<ContentResult> {
|
||||
let result: ContentResult | undefined
|
||||
// Another probe deleted a release this one writes to as garbage meanwhile (LOG.md 2026-10-08 (7)):
|
||||
// the file's own release (nothing left to do), or one its par2 set joins. Again once that
|
||||
// deletion has committed: the set then finds the release's files gone and makes its own.
|
||||
for (const wait of [0, 1000, 3000]) {
|
||||
if (wait) await new Promise((resolve) => setTimeout(resolve, wait))
|
||||
try {
|
||||
result = await this.applyProbe(task, probe)
|
||||
break
|
||||
} catch (error: any) {
|
||||
if (error.code !== '23503' || wait === 3000) throw error
|
||||
if (!(await db.from('files').where('id', task.fileId).select('id').first())) {
|
||||
return { named: 0, contentType: probe.contentType, garbage: 'deleted meanwhile' }
|
||||
}
|
||||
}
|
||||
}
|
||||
if (!result) throw new Error(`Applying the probe of file ${task.fileId} gave no result`)
|
||||
result.garbage = (await GarbageFilter.afterProbe(task.fileId, probe.contentType)) ?? undefined
|
||||
return result
|
||||
}
|
||||
|
||||
private async applyProbe(task: InspectTask, probe: InspectProbe): Promise<ContentResult> {
|
||||
const result: ContentResult = { named: 0, contentType: probe.contentType }
|
||||
const file = await db.from('files').where('id', task.fileId).first()
|
||||
if (!file) return result
|
||||
const binary = await db.from('binaries').where('file_id', task.fileId).first()
|
||||
const { contentType } = probe
|
||||
await FileStore.setDetails(Number(file.id), { content_type: contentType })
|
||||
if (contentType === 'missing') {
|
||||
// Nothing to read; the binary needn't wait for inspection any more.
|
||||
if (binary) await db.from('binaries').where('id', binary.id).update({ awaiting_inspection: null })
|
||||
return result
|
||||
}
|
||||
|
||||
// Fingerprint, so par2 sets (now or later) can name this file exactly.
|
||||
if (binary) {
|
||||
await db
|
||||
.from('binaries')
|
||||
.where('id', binary.id)
|
||||
.update({
|
||||
size: db.raw('coalesce(size, ?)', [probe.size]),
|
||||
md5_16k: db.raw('coalesce(md5_16k, ?)', [probe.md5_16k]),
|
||||
// Inspected: the GC may delete it again by its usual rules (LOG.md 2026-10-06 (11)).
|
||||
awaiting_inspection: null,
|
||||
})
|
||||
}
|
||||
|
||||
result.yencName = await this.nameFromYenc(file, binary, probe.yencName ?? '')
|
||||
if (contentType === 'nzbe' || contentType === 'rclone' || contentType === 'yencbin') {
|
||||
// An encrypted NZB, rclone crypt (names and data encrypted) or yEncBin's container: nothing
|
||||
// to read, and useless without the key.
|
||||
const releaseId = file.release_id ?? binary?.release_id
|
||||
if (releaseId) {
|
||||
await db.from('releases').where('id', releaseId).update({ container: contentType, encrypted: true })
|
||||
}
|
||||
return result
|
||||
}
|
||||
const nicovideo = NICOVIDEO_FILE.exec(task.name)
|
||||
if (nicovideo) {
|
||||
if (probe.nicovideo) result.nicovideo = await this.nameNicovideo(file, probe.nicovideo, nicovideo[1])
|
||||
return result
|
||||
}
|
||||
if (/\.sfv$/i.test(task.name)) {
|
||||
if (typeof probe.text === 'string') result.sfvMoved = (await SfvService.attach(file, probe.text)) ?? undefined
|
||||
return result
|
||||
}
|
||||
if (contentType === 'nzb') {
|
||||
if (typeof probe.text === 'string') result.nzb = (await NzbImportService.import(file, probe.text)) ?? undefined
|
||||
return result
|
||||
}
|
||||
if (probe.par2 && binary) {
|
||||
result.par2 = await Par2Service.attach(binary, probe.par2)
|
||||
} else if (probe.archive && probe.archive.format === '7z') {
|
||||
result.archive = await ArchiveService.record(file, binary, probe.archive)
|
||||
} else if (probe.archive) {
|
||||
result.archive = await ArchiveService.record(file, binary, probe.archive)
|
||||
// The rest of the listing, later and at low priority (D40).
|
||||
await ArchiveWalkService.queueAfterInspection(result.archive.releaseId ?? null, probe.archive)
|
||||
if (probe.archiveNzbs?.length) {
|
||||
result.archiveNzbs = []
|
||||
for (const { entryName, xml } of probe.archiveNzbs) {
|
||||
const imported = await NzbImportService.import(file, xml, { entryName, moveContainer: (probe.archiveNzbEntries ?? 0) === 1 })
|
||||
if (imported) result.archiveNzbs.push(imported)
|
||||
}
|
||||
}
|
||||
}
|
||||
if (binary) result.named = await NameService.resolve([binary.id])
|
||||
return result
|
||||
}
|
||||
|
||||
/** Inspects a file here: prepare, probe on the inspection pool's connections, apply. */
|
||||
async inspect(fileId: number): Promise<ContentResult | null> {
|
||||
const task = await this.prepare(fileId)
|
||||
if (!task) return null
|
||||
const probe = await probeInspection(task, fetchPooled, (message) => logger.warn(message))
|
||||
return this.apply(task, probe)
|
||||
}
|
||||
}
|
||||
|
||||
/** A task's segment list: all of a small file's; of a big one, the first and the last few (a 7z tail). */
|
||||
function taskSegments(segments: Segment[], tailOnly = false): TaskSegment[] {
|
||||
const list = segments.map(({ number, id, size }) => ({ number, id, size }))
|
||||
if (list.length <= TASK_SEGMENTS) return list
|
||||
return tailOnly ? list.slice(-TASK_SEGMENTS) : [list[0], ...list.slice(-TASK_SEGMENTS)]
|
||||
}
|
||||
|
||||
/** More than whole files (40 parts) and par2 files (20) are read; a 7z tail of 16 MB fits easily. */
|
||||
const TASK_SEGMENTS = 40
|
||||
|
||||
/** One article body on a pooled connection. */
|
||||
async function fetchPooled(messageId: string): Promise<Buffer> {
|
||||
const conn = await NntpService.acquire()
|
||||
let failed = false
|
||||
try {
|
||||
return await NntpService.timed(NntpService.fetchBody(conn, messageId), `BODY ${messageId}`)
|
||||
} catch (error: any) {
|
||||
// A missing article (430) leaves the connection fine.
|
||||
failed = !/^4\d\d/.test(String(error.message))
|
||||
throw error
|
||||
} finally {
|
||||
NntpService.release(conn, failed)
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* One line per finding, as worker:collection logs them (and the seed API
|
||||
* for nodes' results).
|
||||
*/
|
||||
export function describeContent(filename: string, content: ContentResult | null): string[] {
|
||||
const lines: string[] = []
|
||||
if (content?.par2) {
|
||||
const { par2 } = content
|
||||
lines.push(`Par2 "${filename}": release "${par2.releaseName}", ${par2.listed} files listed.`)
|
||||
} else if (content?.archive) {
|
||||
const { info, releaseRenamed, grouped } = content.archive
|
||||
const flags = [info.headersEncrypted && 'encrypted headers', info.entries.some((e) => e.encrypted) && 'encrypted files']
|
||||
.filter(Boolean)
|
||||
.join(', ')
|
||||
lines.push(
|
||||
`Archive "${filename}": ${info.format}, ${info.entries.length} entries${flags ? ` (${flags})` : ''}` +
|
||||
(grouped ? ', grouped into a release' : '') +
|
||||
(releaseRenamed ? `, release renamed to "${releaseRenamed}"` : '')
|
||||
)
|
||||
}
|
||||
if (content?.named) lines.push(`"${filename}" got its real name from par2.`)
|
||||
if (content?.contentType === 'missing') lines.push(`"${filename}": the server no longer has its first article.`)
|
||||
if (content?.contentType === 'nzbe') lines.push(`"${filename}" is an encrypted NZB container.`)
|
||||
if (content?.contentType === 'rclone') lines.push(`"${filename}" is rclone-encrypted.`)
|
||||
if (content?.yencName) lines.push(`"${filename}" is "${content.yencName}" (yEnc name).`)
|
||||
if (content?.garbage) lines.push(`"${filename}": release deleted as garbage (${content.garbage === 'deleted meanwhile' ? 'by another probe' : `${content.garbage}, no name or password`}).`)
|
||||
if (content?.nicovideo) lines.push(`"${filename}" → "${content.nicovideo}".`)
|
||||
if (content?.sfvMoved) lines.push(`SFV "${filename}" moved ${content.sfvMoved} files into its release.`)
|
||||
if (content?.contentType === 'yencbin') lines.push(`"${filename}" is a yEncBin encrypted container.`)
|
||||
if (content?.nzb) {
|
||||
const n = content.nzb
|
||||
lines.push(`NZB "${filename}": ${n.listed} files (${n.known} known, ${n.added} added) → "${n.releaseName}".`)
|
||||
}
|
||||
return lines
|
||||
}
|
||||
|
||||
export default new ContentService()
|
||||
@@ -1,264 +0,0 @@
|
||||
import { createHash } from 'node:crypto'
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import indexerConfig from '#config/indexer'
|
||||
import NntpService from '#services/NntpService'
|
||||
import QueueService from '#services/QueueService'
|
||||
import { type OverviewHeader } from '#services/IngestService'
|
||||
import { decodeSegment, inferTotalParts, parseYencMeta } from '#lib/yenc/index'
|
||||
import { sniffContentType } from '#lib/content/index'
|
||||
import { parseRarHeaders } from '#lib/rar/index'
|
||||
|
||||
/** Expanding re-reads the file's article range with XOVER; beyond this it is too costly. */
|
||||
const EXPAND_MAX_RANGE = 200_000
|
||||
/** A claimed row whose worker died is taken again after this long. */
|
||||
const RECLAIM_AFTER = '1 hour'
|
||||
|
||||
export interface DeferredArticle {
|
||||
articleNumber: number
|
||||
header: OverviewHeader
|
||||
}
|
||||
|
||||
/** File name from a deferred uploader's subject: `[i/M] "name" yEnc`, no part counter. */
|
||||
export function deferredName(subject: string | undefined): string | null {
|
||||
return /"([^"]+)"/.exec(subject ?? '')?.[1]?.trim() || null
|
||||
}
|
||||
|
||||
export function isDeferred(header: OverviewHeader): boolean {
|
||||
const id = header['message-id'] ?? ''
|
||||
return indexerConfig.deferBodyMessageIds.some((re) => re.test(id))
|
||||
}
|
||||
|
||||
/** A recorded range never spans more articles than this (one XOVER to expand it). */
|
||||
const RANGE_MAX_SPAN = 50_000
|
||||
|
||||
export function bodyJob(header: OverviewHeader, group: string, articleNumber: number | bigint) {
|
||||
return {
|
||||
name: 'process-body',
|
||||
data: { header, group, articleNumber: Number(articleNumber) },
|
||||
// By message-id: an article crossposted to several indexed groups is downloaded once.
|
||||
opts: { jobId: `body-${createHash('sha1').update(header['message-id'] ?? '').digest('hex')}` },
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Deferred uploads (CONCEPT.md D28). Their articles skip the body queue and
|
||||
* are recorded here per file, from the headers alone. `worker:deferred`
|
||||
* later downloads one body per file for its size, part count and content
|
||||
* type, and sends small files (NZBs in RARs, par2 index files) to the body
|
||||
* queue completely.
|
||||
*/
|
||||
class DeferredService {
|
||||
/**
|
||||
* Records one group's deferred articles: per file when the subject names
|
||||
* the file, otherwise only as article ranges. Returns the number of files.
|
||||
*/
|
||||
async record(groupName: string, articles: DeferredArticle[]): Promise<number> {
|
||||
const named = articles.filter((a) => deferredName(a.header.subject) !== null)
|
||||
await this.recordRanges(groupName, articles.filter((a) => deferredName(a.header.subject) === null))
|
||||
return this.recordFiles(groupName, named)
|
||||
}
|
||||
|
||||
private async recordRanges(groupName: string, articles: DeferredArticle[]) {
|
||||
if (articles.length === 0) return
|
||||
const sorted = [...articles].sort((a, b) => Number(a.articleNumber) - Number(b.articleNumber))
|
||||
const now = new Date()
|
||||
const rows: Record<string, unknown>[] = []
|
||||
for (const article of sorted) {
|
||||
const n = Number(article.articleNumber)
|
||||
const bytes = Number.parseInt(article.header[':bytes'] ?? '0', 10) || 0
|
||||
const last = rows.at(-1)
|
||||
if (last && n - (last.first_article as number) <= RANGE_MAX_SPAN) {
|
||||
last.last_article = n
|
||||
;(last.article_count as number)++
|
||||
last.bytes = (last.bytes as number) + bytes
|
||||
} else {
|
||||
rows.push({ group_name: groupName, first_article: n, last_article: n, article_count: 1, bytes, created_at: now })
|
||||
}
|
||||
}
|
||||
await db.table('deferred_ranges').multiInsert(rows)
|
||||
}
|
||||
|
||||
private async recordFiles(groupName: string, articles: DeferredArticle[]): Promise<number> {
|
||||
const files = new Map<string, {
|
||||
name: string; subject: string; poster: string; first: number; firstId: string; last: number
|
||||
count: number; bytes: number; postedAt: Date | null
|
||||
}>()
|
||||
for (const { articleNumber, header } of articles) {
|
||||
const name = deferredName(header.subject)!
|
||||
const n = Number(articleNumber)
|
||||
const bytes = Number.parseInt(header[':bytes'] ?? '0', 10) || 0
|
||||
const date = header.date ? new Date(header.date) : null
|
||||
const postedAt = date && !Number.isNaN(date.getTime()) ? date : null
|
||||
const file = files.get(name)
|
||||
if (!file) {
|
||||
files.set(name, {
|
||||
name, subject: header.subject ?? '', poster: header.from ?? '', first: n,
|
||||
firstId: (header['message-id'] ?? '').replace(/^<|>$/g, ''), last: n, count: 1, bytes, postedAt,
|
||||
})
|
||||
continue
|
||||
}
|
||||
if (n < file.first) {
|
||||
file.first = n
|
||||
file.firstId = (header['message-id'] ?? '').replace(/^<|>$/g, '')
|
||||
}
|
||||
file.last = Math.max(file.last, n)
|
||||
file.count++
|
||||
file.bytes += bytes
|
||||
if (postedAt && (!file.postedAt || postedAt < file.postedAt)) file.postedAt = postedAt
|
||||
}
|
||||
if (files.size === 0) return 0
|
||||
|
||||
const now = new Date()
|
||||
const rows = [...files.values()].map((f) => ({
|
||||
file_key: createHash('sha1').update(`deferred\n${f.name}`).digest('hex'),
|
||||
name: f.name,
|
||||
subject: f.subject,
|
||||
poster: f.poster,
|
||||
group_name: groupName,
|
||||
first_article: f.first,
|
||||
first_message_id: f.firstId,
|
||||
last_article: f.last,
|
||||
article_count: f.count,
|
||||
bytes: f.bytes,
|
||||
posted_at: f.postedAt,
|
||||
first_seen_at: now,
|
||||
last_seen_at: now,
|
||||
}))
|
||||
// Counts and the article range only add up within the group the file was first seen in.
|
||||
await db.rawQuery(
|
||||
`INSERT INTO deferred_files AS d (${Object.keys(rows[0]).join(', ')})
|
||||
SELECT * FROM jsonb_to_recordset(?::jsonb) AS r(
|
||||
file_key varchar, name text, subject text, poster text, group_name text, first_article bigint,
|
||||
first_message_id text, last_article bigint, article_count int, bytes bigint, posted_at timestamptz,
|
||||
first_seen_at timestamptz, last_seen_at timestamptz)
|
||||
ON CONFLICT (file_key) DO UPDATE SET
|
||||
last_seen_at = EXCLUDED.last_seen_at,
|
||||
first_message_id = CASE WHEN d.group_name = EXCLUDED.group_name AND EXCLUDED.first_article < d.first_article
|
||||
THEN EXCLUDED.first_message_id ELSE d.first_message_id END,
|
||||
first_article = CASE WHEN d.group_name = EXCLUDED.group_name THEN LEAST(d.first_article, EXCLUDED.first_article) ELSE d.first_article END,
|
||||
last_article = CASE WHEN d.group_name = EXCLUDED.group_name THEN GREATEST(d.last_article, EXCLUDED.last_article) ELSE d.last_article END,
|
||||
article_count = d.article_count + CASE WHEN d.group_name = EXCLUDED.group_name THEN EXCLUDED.article_count ELSE 0 END,
|
||||
bytes = d.bytes + CASE WHEN d.group_name = EXCLUDED.group_name THEN EXCLUDED.bytes ELSE 0 END,
|
||||
posted_at = LEAST(d.posted_at, EXCLUDED.posted_at)`,
|
||||
[JSON.stringify(rows)]
|
||||
)
|
||||
return files.size
|
||||
}
|
||||
|
||||
/** Takes up to `limit` files that stopped growing; they are marked `processing`. */
|
||||
async claim(limit: number): Promise<any[]> {
|
||||
const settled = new Date(Date.now() - indexerConfig.deferredSettleMs)
|
||||
const { rows } = await db.rawQuery(
|
||||
`UPDATE deferred_files SET status = 'processing', processed_at = now()
|
||||
WHERE id IN (
|
||||
SELECT id FROM deferred_files
|
||||
WHERE (status = 'pending' AND last_seen_at < ?)
|
||||
OR (status = 'processing' AND processed_at < now() - interval '${RECLAIM_AFTER}')
|
||||
ORDER BY id LIMIT ? FOR UPDATE SKIP LOCKED)
|
||||
RETURNING *`,
|
||||
[settled, limit]
|
||||
)
|
||||
return rows
|
||||
}
|
||||
|
||||
/**
|
||||
* Queues the bodies of up to `limit` recorded ranges, oldest first: the
|
||||
* articles are found again with XOVER. Returns ranges and articles queued.
|
||||
*/
|
||||
async expandRanges(limit: number): Promise<{ ranges: number; articles: number }> {
|
||||
const ranges = await db.from('deferred_ranges').whereNull('expanded_at').orderBy('id').limit(limit)
|
||||
let articles = 0
|
||||
for (const range of ranges) {
|
||||
const conn = await NntpService.acquire()
|
||||
let failed = false
|
||||
let overviews: [number, OverviewHeader][]
|
||||
try {
|
||||
await conn.group(range.group_name)
|
||||
overviews = (await conn.xover(String(range.first_article), String(range.last_article))).overviews
|
||||
} catch (error) {
|
||||
failed = true
|
||||
throw error
|
||||
} finally {
|
||||
NntpService.release(conn, failed)
|
||||
}
|
||||
const jobs = overviews
|
||||
.filter(([, header]) => header && isDeferred(header) && deferredName(header.subject) === null)
|
||||
.map(([articleNumber, header]) => bodyJob(header, range.group_name, articleNumber))
|
||||
for (let i = 0; i < jobs.length; i += 5000) await QueueService.bodyQueue.addBulk(jobs.slice(i, i + 5000))
|
||||
await db.from('deferred_ranges').where('id', range.id).update({ expanded_at: new Date() })
|
||||
articles += jobs.length
|
||||
}
|
||||
return { ranges: ranges.length, articles }
|
||||
}
|
||||
|
||||
/** Reads one body of the file; expands small files into the body queue. Returns the new status. */
|
||||
async process(file: any): Promise<string> {
|
||||
const update: Record<string, unknown> = { processed_at: new Date(), error: null }
|
||||
try {
|
||||
const conn = await NntpService.acquire()
|
||||
let failed = false
|
||||
let body: Buffer
|
||||
try {
|
||||
body = await NntpService.fetchBody(conn, file.first_message_id)
|
||||
} catch (error) {
|
||||
failed = true
|
||||
throw error
|
||||
} finally {
|
||||
NntpService.release(conn, failed)
|
||||
}
|
||||
|
||||
const meta = parseYencMeta(body)
|
||||
const totalParts = inferTotalParts(meta)
|
||||
Object.assign(update, {
|
||||
yenc_name: meta.begin.name,
|
||||
size: meta.begin.size,
|
||||
total_parts: totalParts,
|
||||
sampled_part: meta.begin.part ?? 1,
|
||||
})
|
||||
if ((meta.begin.part ?? 1) === 1) {
|
||||
const data = decodeSegment(body, { lenientSize: true, verifyCrc: false }).data
|
||||
update.content_type = sniffContentType(data)
|
||||
const rar = parseRarHeaders(data)
|
||||
if (rar) update.encrypted = rar.headersEncrypted || rar.entries.some((e) => e.encrypted)
|
||||
}
|
||||
|
||||
update.status = 'sampled'
|
||||
if (totalParts <= indexerConfig.deferredExpandMaxParts && update.encrypted !== true) {
|
||||
const queued = await this.expand(file)
|
||||
if (queued > 0) update.status = 'expanded'
|
||||
}
|
||||
} catch (error: any) {
|
||||
update.status = 'failed'
|
||||
update.error = String(error.message).slice(0, 500)
|
||||
}
|
||||
await db.from('deferred_files').where('id', file.id).update(update)
|
||||
return update.status as string
|
||||
}
|
||||
|
||||
/** Finds the file's articles again (XOVER over its range) and queues their bodies. */
|
||||
private async expand(file: any): Promise<number> {
|
||||
const first = Number(file.first_article)
|
||||
const last = Number(file.last_article)
|
||||
if (last - first > EXPAND_MAX_RANGE) return 0
|
||||
|
||||
const conn = await NntpService.acquire()
|
||||
let failed = false
|
||||
let overviews: [number, OverviewHeader][]
|
||||
try {
|
||||
await conn.group(file.group_name)
|
||||
overviews = (await conn.xover(String(first), String(last))).overviews
|
||||
} catch (error) {
|
||||
failed = true
|
||||
throw error
|
||||
} finally {
|
||||
NntpService.release(conn, failed)
|
||||
}
|
||||
const jobs = overviews
|
||||
.filter(([, header]) => header && deferredName(header.subject) === file.name)
|
||||
.map(([articleNumber, header]) => bodyJob(header, file.group_name, articleNumber))
|
||||
if (jobs.length > 0) await QueueService.bodyQueue.addBulk(jobs)
|
||||
return jobs.length
|
||||
}
|
||||
}
|
||||
|
||||
export default new DeferredService()
|
||||
@@ -1,118 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
import QueueService from '#services/QueueService'
|
||||
import indexerConfig from '#config/indexer'
|
||||
|
||||
/** A failed range is retried after this long... */
|
||||
const REQUEUE_AFTER_MS = 5 * 60 * 1000
|
||||
/** ...at most this many times; after that it stays visible as a gap. */
|
||||
const MAX_REQUEUES = 5
|
||||
/** A queued range without a job in Redis (e.g. Redis was flushed) is re-queued after this long. */
|
||||
const LOST_AFTER_MS = 60 * 60 * 1000
|
||||
|
||||
interface Group {
|
||||
id: number
|
||||
name: string
|
||||
}
|
||||
|
||||
/**
|
||||
* Tracks the article ranges given to worker:fetch. A range only counts as
|
||||
* indexed once the worker reports it done, so failed or lost fetch jobs are
|
||||
* retried instead of silently skipped. See CONCEPT.md (D18).
|
||||
*/
|
||||
class FetchRangeService {
|
||||
/** Queues a range. A backfill range is header-only and waits behind live ones (CONCEPT.md D32). */
|
||||
async queue(group: Group, startId: bigint, endId: bigint, { backfill = false } = {}) {
|
||||
const now = new Date()
|
||||
const [row] = await db
|
||||
.table('fetch_ranges')
|
||||
.insert({ group_id: group.id, start_id: startId.toString(), end_id: endId.toString(), backfill, created_at: now, updated_at: now })
|
||||
.returning(['id', 'requeues'])
|
||||
await this.addJob(group.name, Number(row.id), 0, startId, endId, backfill)
|
||||
}
|
||||
|
||||
private async addJob(groupName: string, rangeId: number, requeues: number, startId: bigint, endId: bigint, backfill: boolean) {
|
||||
await QueueService.nntpFetchQueue.add(
|
||||
'fetch-headers',
|
||||
{ groupName, startId: startId.toString(), endId: endId.toString(), rangeId, ...(backfill ? { backfill: true } : {}) },
|
||||
// The job id changes per attempt: BullMQ ignores an add() for an id it still knows.
|
||||
{ jobId: this.jobId(rangeId, requeues), ...(backfill ? { priority: indexerConfig.backfill.priority } : {}) }
|
||||
)
|
||||
}
|
||||
|
||||
private jobId(rangeId: number, requeues: number) {
|
||||
return `range-${rangeId}-${requeues}`
|
||||
}
|
||||
|
||||
async markDone(rangeId: number) {
|
||||
await db.from('fetch_ranges').where('id', rangeId).update({ status: 'done', error: null, updated_at: new Date() })
|
||||
}
|
||||
|
||||
async markFailed(rangeId: number, error: string) {
|
||||
await db.from('fetch_ranges').where('id', rangeId).update({ status: 'failed', error, updated_at: new Date() })
|
||||
}
|
||||
|
||||
/** Re-queues failed ranges, and queued ranges whose job no longer exists. */
|
||||
async requeue(group: Group): Promise<number> {
|
||||
const now = Date.now()
|
||||
const candidates = await db
|
||||
.from('fetch_ranges')
|
||||
.where('group_id', group.id)
|
||||
.where('requeues', '<', MAX_REQUEUES)
|
||||
.where((q) =>
|
||||
q
|
||||
.where((f) => f.where('status', 'failed').where('updated_at', '<', new Date(now - REQUEUE_AFTER_MS)))
|
||||
.orWhere((l) => l.where('status', 'queued').where('updated_at', '<', new Date(now - LOST_AFTER_MS)))
|
||||
)
|
||||
.orderBy('start_id')
|
||||
|
||||
let requeued = 0
|
||||
for (const range of candidates) {
|
||||
if (range.status === 'queued') {
|
||||
const job = await QueueService.nntpFetchQueue.getJob(this.jobId(Number(range.id), range.requeues))
|
||||
const state = job ? await job.getState() : 'missing'
|
||||
if (state !== 'missing' && state !== 'failed' && state !== 'unknown') continue
|
||||
}
|
||||
await db
|
||||
.from('fetch_ranges')
|
||||
.where('id', range.id)
|
||||
.update({ status: 'queued', requeues: range.requeues + 1, updated_at: new Date() })
|
||||
await this.addJob(group.name, Number(range.id), range.requeues + 1, BigInt(range.start_id), BigInt(range.end_id), Boolean(range.backfill))
|
||||
requeued++
|
||||
logger.warn(
|
||||
`Re-queued ${group.name} ${range.start_id}-${range.end_id} (attempt ${range.requeues + 2}): ${range.error ?? 'job lost'}`
|
||||
)
|
||||
}
|
||||
return requeued
|
||||
}
|
||||
|
||||
/**
|
||||
* Moves `groups.last_indexed_id` up to just below the first live range
|
||||
* that is not done yet (backfill ranges lie below it and don't count), and
|
||||
* forgets done ranges. Returns the new watermark.
|
||||
*/
|
||||
async advanceWatermark(group: Group): Promise<bigint> {
|
||||
const { rows } = await db.rawQuery(
|
||||
`UPDATE groups g SET last_indexed_id = greatest(g.last_indexed_id, coalesce(
|
||||
(SELECT min(start_id) - 1 FROM fetch_ranges WHERE group_id = g.id AND status <> 'done' AND NOT backfill),
|
||||
g.last_queued_id))
|
||||
WHERE g.id = ? RETURNING last_indexed_id`,
|
||||
[group.id]
|
||||
)
|
||||
await db.from('fetch_ranges').where('group_id', group.id).where('status', 'done').delete()
|
||||
return BigInt(rows[0].last_indexed_id)
|
||||
}
|
||||
|
||||
/** Ranges that failed for good (for monitoring). */
|
||||
async abandoned(group: Group) {
|
||||
return db
|
||||
.from('fetch_ranges')
|
||||
.where('group_id', group.id)
|
||||
.where('status', 'failed')
|
||||
.where('requeues', '>=', MAX_REQUEUES)
|
||||
.count('* as count')
|
||||
.first()
|
||||
}
|
||||
}
|
||||
|
||||
export default new FetchRangeService()
|
||||
@@ -1,60 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
|
||||
/** files ids per slice: a few seconds of reads from the HDD. */
|
||||
const SLICE = 100_000
|
||||
|
||||
/**
|
||||
* The file_keys filter (LOG.md 2026-10-06 (7)): a 64-bit hash of every
|
||||
* file_key. The insert trigger adds new keys; `fill` copies the keys of
|
||||
* files from before, in id order and a slice at a time, and marks the
|
||||
* filter complete when it reaches the files that existed when it started
|
||||
* (everything after that has been added by the trigger, which was live
|
||||
* first). FileStore uses the filter only once it is complete.
|
||||
*/
|
||||
class FileKeys {
|
||||
private completeSince = 0
|
||||
|
||||
/** Whether every file's key is in file_keys; cached once true. */
|
||||
async complete(): Promise<boolean> {
|
||||
if (this.completeSince) return true
|
||||
const row = await db.from('file_keys_fill').where('id', 1).first()
|
||||
if (row?.complete) this.completeSince = Date.now()
|
||||
return Boolean(row?.complete)
|
||||
}
|
||||
|
||||
/** Forgets the cached state (tests). */
|
||||
forget() {
|
||||
this.completeSince = 0
|
||||
}
|
||||
|
||||
/** Copies keys until `budgetMs` has passed or the fill is done; returns the files read and whether it's complete. */
|
||||
async fill(budgetMs: number): Promise<{ read: number; complete: boolean; nextId: number; untilId: number }> {
|
||||
const started = Date.now()
|
||||
let state = await db.from('file_keys_fill').where('id', 1).first()
|
||||
if (!state) return { read: 0, complete: false, nextId: 0, untilId: 0 }
|
||||
if (state.until_id === null) {
|
||||
// Everything inserted from here on goes through the trigger (live since the migration).
|
||||
const [{ max }] = await db.from('files').max('id as max')
|
||||
await db.from('file_keys_fill').where('id', 1).update({ until_id: Number(max ?? 0), updated_at: new Date() })
|
||||
state = await db.from('file_keys_fill').where('id', 1).first()
|
||||
}
|
||||
let next = Number(state.next_id)
|
||||
const until = Number(state.until_id)
|
||||
let read = 0
|
||||
while (next <= until && Date.now() - started < budgetMs) {
|
||||
const to = Math.min(until, next + SLICE - 1)
|
||||
const result = await db.rawQuery(
|
||||
`INSERT INTO file_keys (key_hash)
|
||||
SELECT DISTINCT hashtextextended(file_key, 0) FROM files WHERE id BETWEEN ? AND ? AND file_key IS NOT NULL
|
||||
ON CONFLICT DO NOTHING`,
|
||||
[next, to]
|
||||
)
|
||||
read += Number(result.rowCount ?? 0)
|
||||
next = to + 1
|
||||
await db.from('file_keys_fill').where('id', 1).update({ next_id: next, complete: next > until, updated_at: new Date() })
|
||||
}
|
||||
return { read, complete: next > until, nextId: next, untilId: until }
|
||||
}
|
||||
}
|
||||
|
||||
export default new FileKeys()
|
||||
@@ -1,50 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
|
||||
/** Ids per partition (as the migration that partitioned files: …150). */
|
||||
export const FILE_PARTITION_SIZE = 4_000_000
|
||||
/**
|
||||
* Ids kept covered beyond the newest file: four partitions (16M ids, about a day of
|
||||
* seeding), so a few sweeps that can't create one are harmless.
|
||||
*/
|
||||
const AHEAD = 4 * FILE_PARTITION_SIZE
|
||||
|
||||
/**
|
||||
* `files` is partitioned by id (CONCEPT.md D39). An insert with an id no
|
||||
* partition covers fails, so gc:scheduler calls ensure() every sweep to keep
|
||||
* four partitions ready ahead of the newest id. Each new partition gets its
|
||||
* own unique index on file_key (FileStore.insert checks the older ones),
|
||||
* in the parent's tablespace (the HDD on homelab).
|
||||
*/
|
||||
class FilePartitions {
|
||||
async ensure(): Promise<string[]> {
|
||||
const created: string[] = []
|
||||
while (true) {
|
||||
// The sequence, not max(id): cheaper, and it counts ids handed out but not committed yet.
|
||||
const { rows } = await db.rawQuery(
|
||||
`SELECT (SELECT last_value FROM files_id_seq) AS max_id,
|
||||
-- ''{0,1}, not ''?: knex reads ? as a binding.
|
||||
max((regexp_match(pg_get_expr(c.relpartbound, c.oid), 'TO \\(''{0,1}(\\d+)''{0,1}\\)'))[1]::bigint) AS upper
|
||||
FROM pg_inherits i JOIN pg_class c ON c.oid = i.inhrelid
|
||||
WHERE i.inhparent = 'files'::regclass`
|
||||
)
|
||||
const maxId = Number(rows[0].max_id)
|
||||
const upper = Number(rows[0].upper)
|
||||
if (upper - maxId >= AHEAD) return created
|
||||
const name = `files_p${upper}`
|
||||
// CREATE … PARTITION OF locks files and its partitions one by one, which deadlocked with
|
||||
// file inserts (11 failed sweeps on 2026-10-07). Taking the parent's lock first, in one
|
||||
// statement, can only wait; after 10 s it gives up until the next sweep.
|
||||
await db.transaction(async (trx) => {
|
||||
await trx.rawQuery(`SET LOCAL lock_timeout = '10s'`)
|
||||
await trx.rawQuery(`LOCK TABLE files IN ACCESS EXCLUSIVE MODE`)
|
||||
await trx.rawQuery(`CREATE TABLE ${name} PARTITION OF files FOR VALUES FROM (${upper}) TO (${upper + FILE_PARTITION_SIZE})`)
|
||||
await trx.rawQuery(`ALTER TABLE ${name} ADD CONSTRAINT ${name}_file_key_unique UNIQUE (file_key)`)
|
||||
})
|
||||
logger.info(`files: partition ${name} created (ids ${upper}–${upper + FILE_PARTITION_SIZE - 1}).`)
|
||||
created.push(name)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
export default new FilePartitions()
|
||||
@@ -1,138 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import FileKeys from '#services/FileKeys'
|
||||
import type { QueryClientContract } from '@adonisjs/lucid/types/database'
|
||||
import { packSegments, unpackSegments, type Segment } from '#lib/nzb/segments'
|
||||
|
||||
/**
|
||||
* The only code that knows how `files` keeps its segment lists (CONCEPT.md
|
||||
* D31): compressed in `files.segments`, with the first segment's message-id
|
||||
* in the indexed `first_message_id` for lookups. Read them with
|
||||
* `segmentsOf(file)`; write new files with `FileStore.insert`.
|
||||
*/
|
||||
|
||||
/** A new `files` row, with its segments as a list. */
|
||||
export type NewFile = Record<string, unknown> & { segments: Segment[] }
|
||||
|
||||
const unpacked = new WeakMap<object, Segment[]>()
|
||||
|
||||
/** The segments of a `files` row (a row from `db.from('files')`), in order. */
|
||||
export function segmentsOf(file: { segments?: Buffer | Segment[] | null }): Segment[] {
|
||||
const { segments } = file
|
||||
if (!segments) return []
|
||||
if (Array.isArray(segments)) return segments
|
||||
let list = unpacked.get(file)
|
||||
if (!list) {
|
||||
list = unpackSegments(segments)
|
||||
unpacked.set(file, list)
|
||||
}
|
||||
return list
|
||||
}
|
||||
|
||||
/** Stored form of a segment list, for the File model's column. */
|
||||
export function storedSegments(segments: Segment[]) {
|
||||
return packSegments(segments)
|
||||
}
|
||||
|
||||
export function loadedSegments(stored: Buffer | null): Segment[] {
|
||||
return stored ? unpackSegments(stored) : []
|
||||
}
|
||||
|
||||
/** What inspection learns about a file, kept beside `files` (LOG.md 2026-10-08 (14)). */
|
||||
export interface FileDetails {
|
||||
content_type?: string | null
|
||||
archive?: unknown
|
||||
}
|
||||
|
||||
const DETAIL_COLUMNS = ['content_type', 'archive'] as const
|
||||
|
||||
/** Rows per INSERT, well below Postgres' 65,535 bind parameters (tests lower it). */
|
||||
export const insertLimits = { rows: 2000 }
|
||||
|
||||
class FileStore {
|
||||
/** Sets some of a file's details, leaving the others as they are. */
|
||||
async setDetails(fileId: number, details: FileDetails, client: Pick<QueryClientContract, 'rawQuery'> = db) {
|
||||
const columns = DETAIL_COLUMNS.filter((c) => details[c] !== undefined)
|
||||
if (columns.length === 0) return
|
||||
const values = columns.map((c) => (c === 'archive' && details[c] !== null ? JSON.stringify(details[c]) : details[c]))
|
||||
await client.rawQuery(
|
||||
`INSERT INTO file_details (file_id, ${columns.join(', ')}) VALUES (?, ${columns.map(() => '?').join(', ')})
|
||||
ON CONFLICT (file_id) DO UPDATE SET ${columns.map((c) => `${c} = excluded.${c}`).join(', ')}`,
|
||||
[fileId, ...(values as any[])]
|
||||
)
|
||||
}
|
||||
|
||||
/** A file's details (empty when it has none). */
|
||||
async details(fileId: number): Promise<FileDetails> {
|
||||
return (await db.from('file_details').where('file_id', fileId).select('content_type', 'archive').first()) ?? {}
|
||||
}
|
||||
|
||||
/**
|
||||
* Inserts files, skipping those whose `file_key` exists. Returns the rows
|
||||
* that were inserted (id, filename, file_key). `files` is partitioned by id
|
||||
* (CONCEPT.md D39) and each partition has its own unique file_key index:
|
||||
* once file_keys (a 64-bit hash of every key) is complete, it says which keys
|
||||
* may exist, and only those are looked up in the partitions; two
|
||||
* writers adding the same new key at once meet in the newest partition's index.
|
||||
*/
|
||||
async insert(
|
||||
files: NewFile[],
|
||||
client: Pick<QueryClientContract, 'knexQuery' | 'rawQuery'> = db
|
||||
): Promise<{ id: number; filename: string; file_key: string | null }[]> {
|
||||
if (files.length === 0) return []
|
||||
const keys = [...new Set(files.map((file) => file.file_key).filter((key): key is string => typeof key === 'string'))]
|
||||
const known = new Set<string>()
|
||||
const filtered = await FileKeys.complete()
|
||||
for (let i = 0; i < keys.length; i += 5000) {
|
||||
const chunk = keys.slice(i, i + 5000)
|
||||
if (!filtered) {
|
||||
const found = await client.knexQuery().from('files').whereIn('file_key', chunk).select('file_key')
|
||||
for (const row of found) known.add(row.file_key)
|
||||
continue
|
||||
}
|
||||
const { rows } = await client.rawQuery(
|
||||
`SELECT k.key FROM unnest(?::text[]) AS k(key)
|
||||
WHERE EXISTS (SELECT 1 FROM file_keys f WHERE f.key_hash = hashtextextended(k.key, 0))`,
|
||||
[chunk]
|
||||
)
|
||||
const maybe = rows.map((row: { key: string }) => row.key)
|
||||
if (maybe.length === 0) continue
|
||||
const found = await client.knexQuery().from('files').whereIn('file_key', maybe).select('file_key')
|
||||
for (const row of found) known.add(row.file_key)
|
||||
}
|
||||
const seen = new Set<string>()
|
||||
const rows = files
|
||||
.filter((file) => {
|
||||
const key = file.file_key as string | null | undefined
|
||||
if (!key) return true
|
||||
if (known.has(key) || seen.has(key)) return false
|
||||
seen.add(key)
|
||||
return true
|
||||
})
|
||||
.map(({ segments, ...file }) => {
|
||||
const sorted = [...segments].sort((a, b) => a.number - b.number)
|
||||
return { ...file, segments: packSegments(sorted), first_message_id: sorted[0]?.id ?? null }
|
||||
})
|
||||
if (rows.length === 0) return []
|
||||
// Details given with a new file (e.g. a known content type) go to file_details.
|
||||
const details = rows.map((row: Record<string, unknown>) => {
|
||||
const d: FileDetails = {}
|
||||
for (const c of DETAIL_COLUMNS) if (c in row) (d as any)[c] = row[c]
|
||||
for (const c of DETAIL_COLUMNS) delete row[c]
|
||||
return d
|
||||
})
|
||||
// A statement takes at most 65,535 parameters: ~5,400 rows of 12 columns.
|
||||
const inserted: { id: number; filename: string; file_key: string | null }[] = []
|
||||
for (let i = 0; i < rows.length; i += insertLimits.rows) {
|
||||
const chunk = await client.knexQuery().table('files').insert(rows.slice(i, i + insertLimits.rows)).onConflict().ignore().returning(['id', 'filename', 'file_key'])
|
||||
for (const row of chunk) inserted.push({ id: Number(row.id), filename: row.filename, file_key: row.file_key ?? null })
|
||||
}
|
||||
const detailsByKey = new Map(rows.map((row: Record<string, unknown>, i: number) => [row.file_key ?? row.filename, details[i]]))
|
||||
for (const row of inserted) {
|
||||
const d = detailsByKey.get(row.file_key ?? row.filename)
|
||||
if (d && Object.keys(d).length > 0) await this.setDetails(row.id, d, client)
|
||||
}
|
||||
return inserted
|
||||
}
|
||||
}
|
||||
|
||||
export default new FileStore()
|
||||
@@ -1,217 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
import indexerConfig from '#config/indexer'
|
||||
import { looksRandom } from '#services/ArchiveService'
|
||||
import { fileStem } from '#lib/release/index'
|
||||
import { explored, signatureGroup, signatureKey, uploadTool } from '#lib/garbage/index'
|
||||
|
||||
/** What the filter needs to know about an upload before it is stored. */
|
||||
export interface Upload {
|
||||
/** The posted file name. */
|
||||
name: string
|
||||
/** Any of its articles' Message-IDs (one tool posts all of a file). */
|
||||
messageId: string | null | undefined
|
||||
groups: unknown
|
||||
/** Its release's key (the file key when it has none): kept or dropped as a whole. */
|
||||
releaseKey: string
|
||||
}
|
||||
|
||||
/** Contents that can't name a random-named release without a key (LOG.md 2026-10-08 (5)). */
|
||||
const CONTAINERS = ['rclone', 'nzbe', 'yencbin']
|
||||
|
||||
/** How long the dropped signatures are cached per process. */
|
||||
const CACHE_MS = 5 * 60 * 1000
|
||||
|
||||
export function isRandomName(name: string) {
|
||||
return looksRandom(fileStem(name))
|
||||
}
|
||||
|
||||
/**
|
||||
* Garbage: random-named uploads that never get a real name and have no
|
||||
* known password (CONCEPT.md D42). Two filters:
|
||||
* - before storing (seeding, ingest): random-named uploads whose signature
|
||||
* (posting tool, first group) is dropped, or whose release was dropped
|
||||
* after its probe, aren't stored;
|
||||
* - after a probe: a random-named release whose probes are all done, that
|
||||
* is still without a real name, password, names from its files or spot,
|
||||
* and is encrypted, a container or missing its first article, is deleted.
|
||||
* The dropped signatures are learned from a sample of probed releases
|
||||
* (learn(), gc:scheduler).
|
||||
*/
|
||||
class GarbageFilter {
|
||||
private dropped: Set<string> | null = null
|
||||
private loadedAt = 0
|
||||
|
||||
/** The dropped signatures, cached for CACHE_MS. */
|
||||
async droppedSignatures(): Promise<Set<string>> {
|
||||
if (this.dropped && Date.now() - this.loadedAt < CACHE_MS) return this.dropped
|
||||
const rows = await db.from('upload_signatures').where('dropped', true).select('tool', 'group_name')
|
||||
this.dropped = new Set(rows.map((row) => signatureKey(row.tool, row.group_name)))
|
||||
this.loadedAt = Date.now()
|
||||
return this.dropped
|
||||
}
|
||||
|
||||
/** Forgets the cached signatures (after learn(), and in tests). */
|
||||
reset() {
|
||||
this.dropped = null
|
||||
}
|
||||
|
||||
/** Whether a random-named upload's signature is dropped and it isn't in the explored share. */
|
||||
private droppedBySignature(upload: Upload, signatures: Set<string>) {
|
||||
if (signatures.size === 0) return false
|
||||
if (!signatures.has(signatureKey(uploadTool(upload.messageId), signatureGroup(upload.groups)))) return false
|
||||
return !explored(upload.releaseKey, indexerConfig.garbage.explorePercent)
|
||||
}
|
||||
|
||||
/**
|
||||
* The items worth storing: drops random-named uploads of dropped signatures
|
||||
* (but explorePercent of their releases) and of dropped releases.
|
||||
*/
|
||||
async keep<T>(items: T[], upload: (item: T) => Upload): Promise<{ kept: T[]; dropped: number }> {
|
||||
if (!indexerConfig.garbage.enabled || items.length === 0) return { kept: items, dropped: 0 }
|
||||
const signatures = await this.droppedSignatures()
|
||||
// Per release, kept as a whole when any of its files has a readable name. A file is
|
||||
// looked at once: an ingest batch has many articles of each.
|
||||
const verdicts = new Map<string, 'keep' | 'drop' | 'check'>()
|
||||
const seen = new Set<string>()
|
||||
for (const item of items) {
|
||||
const u = upload(item)
|
||||
if (verdicts.get(u.releaseKey) === 'keep' || seen.has(`${u.releaseKey}\n${u.name}`)) continue
|
||||
seen.add(`${u.releaseKey}\n${u.name}`)
|
||||
if (!isRandomName(u.name)) verdicts.set(u.releaseKey, 'keep')
|
||||
else if (!verdicts.has(u.releaseKey)) verdicts.set(u.releaseKey, this.droppedBySignature(u, signatures) ? 'drop' : 'check')
|
||||
}
|
||||
const check = [...verdicts].filter(([, v]) => v === 'check').map(([key]) => key)
|
||||
for (let i = 0; i < check.length; i += 5000) {
|
||||
const rows = await db.from('dropped_releases').whereIn('release_key', check.slice(i, i + 5000)).select('release_key')
|
||||
for (const row of rows) verdicts.set(row.release_key, 'drop')
|
||||
}
|
||||
const kept = items.filter((item) => verdicts.get(upload(item).releaseKey) !== 'drop')
|
||||
return { kept, dropped: items.length - kept.length }
|
||||
}
|
||||
|
||||
/**
|
||||
* Whether a stored file's probe can be skipped: a random-named file of a
|
||||
* dropped signature, outside the explored share (files stored before its
|
||||
* signature was dropped). Its binary stops waiting for inspection.
|
||||
*/
|
||||
async skipProbe(file: { id: number | string; filename: string; real_name?: string | null; first_message_id?: string | null; groups: unknown; release_id?: number | string | null }) {
|
||||
if (!indexerConfig.garbage.enabled || file.real_name || !isRandomName(file.filename)) return false
|
||||
const signatures = await this.droppedSignatures()
|
||||
if (!signatures.has(signatureKey(uploadTool(file.first_message_id), signatureGroup(file.groups)))) return false
|
||||
const release = file.release_id ? await db.from('releases').where('id', file.release_id).select('release_key').first() : null
|
||||
if (explored(release?.release_key ?? String(file.id), indexerConfig.garbage.explorePercent)) return false
|
||||
await db.from('binaries').where('file_id', file.id).where('awaiting_inspection', true).update({ awaiting_inspection: null })
|
||||
return true
|
||||
}
|
||||
|
||||
/**
|
||||
* After a file's probe has been applied: notes its release for learning
|
||||
* (a sample) and deletes the release when it is garbage. Never throws:
|
||||
* the probe's result is stored already. Returns why the release was
|
||||
* deleted, or null.
|
||||
*/
|
||||
async afterProbe(fileId: number, contentType: string | null): Promise<string | null> {
|
||||
try {
|
||||
const file = await db.from('files').where('id', fileId).select('id', 'filename', 'real_name', 'first_message_id', 'groups', 'release_id').first()
|
||||
if (!file?.release_id || !isRandomName(String(file.filename))) return null
|
||||
const releaseId = Number(file.release_id)
|
||||
if (releaseId % indexerConfig.garbage.probeSample === 0) {
|
||||
await db.rawQuery(`INSERT INTO signature_probes (release_id, tool, group_name) VALUES (?, ?, ?) ON CONFLICT DO NOTHING`, [
|
||||
releaseId,
|
||||
uploadTool(file.first_message_id),
|
||||
signatureGroup(file.groups),
|
||||
])
|
||||
}
|
||||
if (!indexerConfig.garbage.enabled) return null
|
||||
return await this.dropIfGarbage(releaseId, contentType)
|
||||
} catch (error: any) {
|
||||
logger.warn(`Garbage check after probing file ${fileId}: ${error.message}`)
|
||||
return null
|
||||
}
|
||||
}
|
||||
|
||||
private async dropIfGarbage(releaseId: number, contentType: string | null): Promise<string | null> {
|
||||
const { rows } = await db.rawQuery(
|
||||
`SELECT r.release_key, r.container, r.encrypted,
|
||||
is_real_release_name(r.name) OR r.password IS NOT NULL OR coalesce(r.search_files, '') <> ''
|
||||
OR EXISTS (SELECT 1 FROM spots s WHERE s.release_id = r.id) AS named,
|
||||
EXISTS (SELECT 1 FROM binaries b WHERE b.release_id = r.id AND b.awaiting_inspection) AS waiting
|
||||
FROM releases r WHERE r.id = ?`,
|
||||
[releaseId]
|
||||
)
|
||||
const release = rows[0]
|
||||
if (!release || release.named || release.waiting) return null
|
||||
const reason = CONTAINERS.includes(release.container) ? release.container : release.encrypted === true ? 'encrypted' : contentType === 'missing' ? 'missing' : null
|
||||
if (!reason) return null
|
||||
await db.transaction(async (trx) => {
|
||||
await trx.rawQuery(`INSERT INTO dropped_releases (release_key, reason) VALUES (?, ?) ON CONFLICT DO NOTHING`, [release.release_key, reason])
|
||||
await trx.rawQuery(`DELETE FROM binaries WHERE release_id = ?`, [releaseId])
|
||||
await trx.rawQuery(`DELETE FROM files WHERE release_id = ?`, [releaseId])
|
||||
await trx.rawQuery(`DELETE FROM releases WHERE id = ?`, [releaseId])
|
||||
})
|
||||
return reason
|
||||
}
|
||||
|
||||
/**
|
||||
* Learns the signatures from the probed releases of the last windowDays:
|
||||
* dropped when at least minProbes are known and at most maxNamedRate got a
|
||||
* real name, password, names from their files or a spot. A signature with
|
||||
* fewer probes keeps its verdict (a dropped one only gets explored probes).
|
||||
* Releases deleted as garbage count as unnamed. Returns the signatures
|
||||
* whose verdict changed.
|
||||
*/
|
||||
async learn(): Promise<{ signatures: number; dropped: number; changed: string[] }> {
|
||||
const { minProbes, maxNamedRate, windowDays } = indexerConfig.garbage
|
||||
await db.rawQuery(`DELETE FROM signature_probes WHERE probed_at < now() - make_interval(days => ?)`, [windowDays])
|
||||
const before = new Map((await db.from('upload_signatures').select('tool', 'group_name', 'dropped')).map((row) => [signatureKey(row.tool, row.group_name), row.dropped]))
|
||||
await db.rawQuery(
|
||||
`INSERT INTO upload_signatures (tool, group_name, probed, named, dropped, updated_at)
|
||||
SELECT tool, group_name, probed, named, probed >= ?::int AND named <= probed * ?::float8, now()
|
||||
FROM (
|
||||
SELECT p.tool, p.group_name, count(*)::int AS probed,
|
||||
(count(*) FILTER (WHERE is_real_release_name(r.name) OR r.password IS NOT NULL OR coalesce(r.search_files, '') <> ''
|
||||
OR EXISTS (SELECT 1 FROM spots s WHERE s.release_id = r.id)))::int AS named
|
||||
FROM signature_probes p LEFT JOIN releases r ON r.id = p.release_id
|
||||
GROUP BY p.tool, p.group_name
|
||||
) s
|
||||
ON CONFLICT (tool, group_name) DO UPDATE SET
|
||||
probed = excluded.probed, named = excluded.named, updated_at = now(),
|
||||
dropped = CASE WHEN excluded.probed >= ?::int THEN excluded.dropped ELSE upload_signatures.dropped END`,
|
||||
[minProbes, maxNamedRate, minProbes]
|
||||
)
|
||||
const after = await db.from('upload_signatures').select('tool', 'group_name', 'dropped', 'probed', 'named')
|
||||
const changed = after
|
||||
.filter((row) => (before.get(signatureKey(row.tool, row.group_name)) ?? false) !== row.dropped)
|
||||
.map((row) => `${row.tool} ${row.group_name || '(no group)'}: ${row.dropped ? 'dropped' : 'kept'} (${row.named}/${row.probed} named)`)
|
||||
this.reset()
|
||||
return { signatures: after.length, dropped: after.filter((row) => row.dropped).length, changed }
|
||||
}
|
||||
|
||||
/**
|
||||
* Fills signature_probes from a sample of files probed before it existed
|
||||
* (garbage:learn --seed): random-named files with a content type, from
|
||||
* TABLESAMPLE SYSTEM (percent) of files. Returns the releases added.
|
||||
*/
|
||||
async seed(percent: number): Promise<number> {
|
||||
const { rows } = await db.rawQuery(
|
||||
`SELECT DISTINCT ON (f.release_id) f.release_id, f.first_message_id, f.groups
|
||||
FROM files f TABLESAMPLE SYSTEM (?) JOIN file_details d ON d.file_id = f.id
|
||||
WHERE d.content_type IS NOT NULL AND f.release_id IS NOT NULL AND NOT is_readable_name(f.filename)`,
|
||||
[percent]
|
||||
)
|
||||
let added = 0
|
||||
for (let i = 0; i < rows.length; i += 2000) {
|
||||
const chunk = rows.slice(i, i + 2000)
|
||||
const result = await db.rawQuery(
|
||||
`INSERT INTO signature_probes (release_id, tool, group_name)
|
||||
SELECT * FROM unnest(?::bigint[], ?::text[], ?::text[]) ON CONFLICT DO NOTHING`,
|
||||
[chunk.map((row: any) => row.release_id), chunk.map((row: any) => uploadTool(row.first_message_id)), chunk.map((row: any) => signatureGroup(row.groups))]
|
||||
)
|
||||
added += result.rowCount ?? 0
|
||||
}
|
||||
return added
|
||||
}
|
||||
}
|
||||
|
||||
export default new GarbageFilter()
|
||||
@@ -1,85 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
|
||||
/** Only indexes at least this big are worth checking. */
|
||||
const MIN_BYTES = 50 * 1024 * 1024
|
||||
/** Leaf pages less full than this (percent) mean the index is mostly empty space. */
|
||||
const MAX_LEAF_DENSITY = 50
|
||||
|
||||
export interface Rebuilt {
|
||||
index: string
|
||||
density: number
|
||||
before: number
|
||||
after: number
|
||||
}
|
||||
|
||||
/**
|
||||
* GC deletes millions of rows (single-article binaries, copied parts, empty
|
||||
* releases). Vacuum makes their index entries reusable but doesn't compact
|
||||
* the index, so B-trees on the churning tables end up 10–20% full (LOG.md
|
||||
* 2026-10-01). Rebuilding them online brings them back to size (CONCEPT.md
|
||||
* D33). Needs the pgstattuple extension.
|
||||
*/
|
||||
class IndexMaintenance {
|
||||
async rebuildBloated({ dryRun = false } = {}): Promise<Rebuilt[]> {
|
||||
const { rows: extension } = await db.rawQuery(`SELECT 1 FROM pg_extension WHERE extname = 'pgstattuple'`)
|
||||
if (extension.length === 0) {
|
||||
logger.warn('Index maintenance skipped: the pgstattuple extension is not installed.')
|
||||
return []
|
||||
}
|
||||
if (!dryRun) await this.dropLeftovers()
|
||||
|
||||
const { rows: candidates } = await db.rawQuery(
|
||||
`SELECT c.relname AS index, pg_relation_size(c.oid) AS bytes
|
||||
FROM pg_index i JOIN pg_class c ON c.oid = i.indexrelid JOIN pg_am am ON am.oid = c.relam
|
||||
JOIN pg_namespace n ON n.oid = c.relnamespace
|
||||
WHERE am.amname = 'btree' AND n.nspname = 'public' AND i.indisvalid AND pg_relation_size(c.oid) >= ?
|
||||
ORDER BY 2 DESC`,
|
||||
[MIN_BYTES]
|
||||
)
|
||||
const rebuilt: Rebuilt[] = []
|
||||
for (const { index, bytes } of candidates) {
|
||||
const { rows } = await db.rawQuery(`SELECT avg_leaf_density FROM pgstatindex(?)`, [index])
|
||||
const density = Number(rows[0].avg_leaf_density)
|
||||
if (density >= MAX_LEAF_DENSITY) continue
|
||||
if (!dryRun && !(await this.reindex(index))) continue
|
||||
const { rows: after } = await db.rawQuery(`SELECT pg_relation_size(?::regclass) AS bytes`, [index])
|
||||
const result = { index, density, before: Number(bytes), after: Number(after[0].bytes) }
|
||||
rebuilt.push(result)
|
||||
logger.info(`Index ${index}: ${density.toFixed(0)}% full, ${mb(result.before)} → ${mb(result.after)}${dryRun ? ' (dry run)' : ''}.`)
|
||||
}
|
||||
return rebuilt
|
||||
}
|
||||
|
||||
/**
|
||||
* REINDEX CONCURRENTLY can lose a deadlock against busy writers (seen on
|
||||
* binaries during a backfill). Its half-built copy is dropped and it is
|
||||
* tried once more; a second failure skips the index until the next run.
|
||||
*/
|
||||
/** Invalid copies an interrupted REINDEX CONCURRENTLY left behind (`<index>_ccnew`, `_ccnew1`, …). */
|
||||
private async dropLeftovers(index = '') {
|
||||
const { rows } = await db.rawQuery(
|
||||
`SELECT c.relname FROM pg_index i JOIN pg_class c ON c.oid = i.indexrelid
|
||||
WHERE NOT i.indisvalid AND c.relname LIKE ? || '%\\_ccnew%'`,
|
||||
[index]
|
||||
)
|
||||
for (const { relname } of rows) await db.rawQuery(`DROP INDEX CONCURRENTLY IF EXISTS "${relname}"`)
|
||||
}
|
||||
|
||||
private async reindex(index: string): Promise<boolean> {
|
||||
for (let attempt = 1; attempt <= 2; attempt++) {
|
||||
try {
|
||||
await db.rawQuery(`REINDEX INDEX CONCURRENTLY "${index}"`)
|
||||
return true
|
||||
} catch (error: any) {
|
||||
logger.warn(`Rebuilding index ${index} failed (attempt ${attempt}): ${error.message}`)
|
||||
await this.dropLeftovers(index)
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
}
|
||||
|
||||
const mb = (bytes: number) => `${Math.round(bytes / 1024 / 1024)} MB`
|
||||
|
||||
export default new IndexMaintenance()
|
||||
@@ -1,424 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
import type { QueryClientContract } from '@adonisjs/lucid/types/database'
|
||||
import type { Knex } from 'knex'
|
||||
import QueueService from '#services/QueueService'
|
||||
import NameService from '#services/NameService'
|
||||
import GarbageFilter from '#services/GarbageFilter'
|
||||
import type { IngestSegment } from '#lib/ingest/segment'
|
||||
|
||||
// Header parsing is pure and lives in lib, so seeding nodes need no database (CONCEPT.md D37).
|
||||
export { segmentFromHeader, segmentFromYenc, type IngestSegment, type OverviewHeader } from '#lib/ingest/segment'
|
||||
|
||||
export interface CompletedBinary {
|
||||
id: number
|
||||
fileKey: string
|
||||
}
|
||||
|
||||
const BINARY_CHUNK = 1000
|
||||
/** last_seen_at is only rewritten when it is older than this (it drives GC at 48 h). */
|
||||
const TOUCH_MS = 10 * 60 * 1000
|
||||
|
||||
type Merge = 'least' | 'coalesce' | 'touch'
|
||||
|
||||
/**
|
||||
* Inserts rows keyed by `key` and returns their ids. Existing rows are not
|
||||
* locked unless something in them changes (`least`: an earlier value,
|
||||
* `coalesce`: a value where there was none, `touch`: a timestamp more than
|
||||
* TOUCH_MS newer). Parallel fetch jobs see mostly the same rows, and
|
||||
* `INSERT … ON CONFLICT DO UPDATE` made them wait on each other for every one.
|
||||
*/
|
||||
async function upsertKeyed(
|
||||
knex: Knex,
|
||||
table: 'binaries' | 'releases',
|
||||
key: string,
|
||||
rows: Record<string, unknown>[],
|
||||
merge: Record<string, [Merge, string]>
|
||||
): Promise<Map<string, number>> {
|
||||
const ids = new Map<string, number>()
|
||||
if (rows.length === 0) return ids
|
||||
const inserted: Record<string, any>[] = await knex(table).insert(rows).onConflict(key).ignore().returning(['id', key])
|
||||
for (const row of inserted) ids.set(row[key], Number(row.id))
|
||||
|
||||
const incoming = new Map(rows.filter((r) => !ids.has(r[key] as string)).map((r) => [r[key] as string, r]))
|
||||
if (incoming.size === 0) return ids
|
||||
const columns = Object.keys(merge)
|
||||
const existing: Record<string, any>[] = await knex(table).whereIn(key, [...incoming.keys()]).select('id', key, ...columns)
|
||||
|
||||
const changes: Record<string, unknown>[] = []
|
||||
for (const row of existing) {
|
||||
ids.set(row[key], Number(row.id))
|
||||
const next = incoming.get(row[key])!
|
||||
const changed = columns.some((column) => {
|
||||
const [kind] = merge[column]
|
||||
const current = row[column]
|
||||
const value = next[column]
|
||||
if (value === null || value === undefined) return false
|
||||
if (kind === 'coalesce') return current === null
|
||||
if (kind === 'least') return current === null || value < current
|
||||
return current === null || +(value as Date) - +current > TOUCH_MS
|
||||
})
|
||||
if (changed) changes.push({ id: Number(row.id), ...Object.fromEntries(columns.map((c) => [c, next[c] ?? null])) })
|
||||
}
|
||||
if (changes.length === 0) return ids
|
||||
|
||||
const set = columns
|
||||
.map((c) => {
|
||||
const [kind] = merge[c]
|
||||
const fn = kind === 'least' ? 'least' : kind === 'touch' ? 'greatest' : 'coalesce'
|
||||
return `${c} = ${fn}(t.${c}, v.${c})`
|
||||
})
|
||||
.join(', ')
|
||||
const types = columns.map((c) => `${c} ${merge[c][1]}`).join(', ')
|
||||
changes.sort((a, b) => (a.id as number) - (b.id as number))
|
||||
for (let attempt = 1; ; attempt++) {
|
||||
try {
|
||||
await knex.raw(
|
||||
`UPDATE ${table} t SET ${set}
|
||||
FROM jsonb_to_recordset(?::jsonb) AS v(id bigint, ${types})
|
||||
WHERE t.id = v.id`,
|
||||
[JSON.stringify(changes)]
|
||||
)
|
||||
return ids
|
||||
} catch (error: any) {
|
||||
// Rare now, but two jobs can still update overlapping rows in a different order.
|
||||
if (error.code !== '40P01' || attempt >= 3) throw error
|
||||
}
|
||||
}
|
||||
}
|
||||
/** Completed binaries per worker:file job. */
|
||||
const FILE_BATCH = 100
|
||||
|
||||
/** Size-only matching below this is too ambiguous (nfo, sfv, small volumes). */
|
||||
export const PAR2_SIZE_ONLY_MIN = 1024 * 1024
|
||||
/** A par2 set and its files are posted together; don't match across days. */
|
||||
export const PAR2_MATCH_WINDOW_SECONDS = 12 * 60 * 60
|
||||
const PART_CHUNK = 2000
|
||||
|
||||
function chunks<T>(items: T[], size: number): T[][] {
|
||||
const result: T[][] = []
|
||||
for (let i = 0; i < items.length; i += size) result.push(items.slice(i, i + size))
|
||||
return result
|
||||
}
|
||||
|
||||
function validDate(date?: Date) {
|
||||
return date && !Number.isNaN(date.getTime()) ? date : null
|
||||
}
|
||||
|
||||
/**
|
||||
* Stores segments in `binaries`/`parts` and queues every binary that became
|
||||
* complete exactly once. Idempotent: ingesting the same segments again is a
|
||||
* no-op. See CONCEPT.md (D12).
|
||||
*/
|
||||
class IngestService {
|
||||
async ingest(batch: IngestSegment[]): Promise<CompletedBinary[]> {
|
||||
// Garbage isn't stored at all: no release, binary or parts (CONCEPT.md D42).
|
||||
const { kept: segments } = await GarbageFilter.keep(batch, (s) => ({
|
||||
name: s.name,
|
||||
messageId: s.messageId,
|
||||
groups: s.groups,
|
||||
releaseKey: s.release?.key ?? s.fileKey,
|
||||
}))
|
||||
if (segments.length === 0) return []
|
||||
|
||||
// Short statements, no enclosing transaction: every step is idempotent, and
|
||||
// row locks held for a whole batch made parallel fetch jobs wait on each other.
|
||||
const binaryIds = await (async () => {
|
||||
const knex = db.connection().getWriteClient()
|
||||
const now = new Date()
|
||||
const releaseIds = await this.upsertReleases(knex, segments, now)
|
||||
|
||||
// One row per file_key (ON CONFLICT cannot touch a row twice per statement),
|
||||
// sorted so concurrent workers lock rows in the same order and cannot deadlock.
|
||||
const binaries = new Map<string, Record<string, unknown>>()
|
||||
for (const segment of segments) {
|
||||
const postedAt = validDate(segment.postedAt)
|
||||
const existing = binaries.get(segment.fileKey)
|
||||
if (existing) {
|
||||
if (postedAt && (!existing.posted_at || postedAt < (existing.posted_at as Date))) {
|
||||
existing.posted_at = postedAt
|
||||
}
|
||||
continue
|
||||
}
|
||||
binaries.set(segment.fileKey, {
|
||||
file_key: segment.fileKey,
|
||||
poster: segment.poster,
|
||||
subject: segment.subject,
|
||||
name: segment.name,
|
||||
total_parts: segment.totalParts,
|
||||
groups: JSON.stringify(segment.groups),
|
||||
posted_at: postedAt,
|
||||
first_seen_at: now,
|
||||
last_seen_at: now,
|
||||
release_id: segment.release ? releaseIds.get(segment.release.key) : null,
|
||||
size: segment.size ?? null,
|
||||
md5_16k: segment.md5_16k ?? null,
|
||||
})
|
||||
}
|
||||
const binaryRows = [...binaries.values()].sort((a, b) =>
|
||||
(a.file_key as string).localeCompare(b.file_key as string)
|
||||
)
|
||||
|
||||
const idsByKey = new Map<string, number>()
|
||||
for (const chunk of chunks(binaryRows, BINARY_CHUNK)) {
|
||||
const ids = await upsertKeyed(knex, 'binaries', 'file_key', chunk, {
|
||||
last_seen_at: ['touch', 'timestamptz'],
|
||||
posted_at: ['least', 'timestamptz'],
|
||||
release_id: ['coalesce', 'bigint'],
|
||||
md5_16k: ['coalesce', 'varchar'],
|
||||
})
|
||||
for (const [key, id] of ids) idsByKey.set(key, id)
|
||||
}
|
||||
|
||||
// A binary that is a file already has its segment list in `files`; its
|
||||
// header parts were dropped then. Crossposts in other groups, reposts
|
||||
// and backfill ranges keep bringing the same articles, and inserting
|
||||
// them again grew `parts` to 56M rows nobody reads (LOG.md 2026-10-04).
|
||||
const done = new Set<number>()
|
||||
const ids = [...idsByKey.values()]
|
||||
for (const chunk of chunks(ids, BINARY_CHUNK)) {
|
||||
const { rows } = await knex.raw('SELECT id FROM binaries WHERE id = ANY(?) AND file_id IS NOT NULL', [chunk])
|
||||
for (const row of rows) done.add(Number(row.id))
|
||||
}
|
||||
|
||||
const partRows = segments
|
||||
.map((segment) => ({
|
||||
binary_id: idsByKey.get(segment.fileKey)!,
|
||||
number: segment.number,
|
||||
message_id: segment.messageId,
|
||||
bytes: segment.bytes,
|
||||
article_number: segment.articleNumber?.toString() ?? null,
|
||||
pcrc32: segment.pcrc32 ?? null,
|
||||
part_size: segment.partSize ?? null,
|
||||
file_crc32: segment.fileCrc32 ?? null,
|
||||
md5_16k: segment.md5_16k ?? null,
|
||||
}))
|
||||
const weakRows = partRows.filter((row, i) => segments[i].weakIdentity && !(done.has(row.binary_id) && row.pcrc32 === null))
|
||||
// Body-derived parts (with a per-part CRC) are kept, as worker:file does.
|
||||
const newParts = partRows.filter((row) => !(done.has(row.binary_id) && row.pcrc32 === null))
|
||||
|
||||
newParts.sort((a, b) => a.binary_id - b.binary_id || a.number - b.number)
|
||||
// Parts and the binaries' stored-part counts in one statement, so the count can't miss a
|
||||
// part (LOG.md 2026-10-06 (7)); claimCompleted compares it instead of counting parts. NULL
|
||||
// (binaries from before part_count) is counted once: the statement's own rows aren't in
|
||||
// its snapshot, so they're added on top. Concurrent batches can lock binaries in a
|
||||
// different order: a deadlock rolls the whole statement back, and it is retried.
|
||||
for (const chunk of chunks(newParts, PART_CHUNK)) {
|
||||
const insert = knex('parts').insert(chunk).onConflict(['binary_id', 'number']).ignore().returning('binary_id').toSQL()
|
||||
for (let attempt = 1; ; attempt++) {
|
||||
try {
|
||||
await knex.raw(
|
||||
`WITH added AS (${insert.sql}),
|
||||
counted AS (SELECT binary_id, count(*)::int AS n FROM added GROUP BY binary_id)
|
||||
UPDATE binaries b SET part_count = CASE
|
||||
WHEN b.part_count IS NULL THEN (SELECT count(*) FROM parts p WHERE p.binary_id = b.id) + c.n
|
||||
ELSE b.part_count + c.n END
|
||||
FROM counted c WHERE b.id = c.binary_id`,
|
||||
insert.bindings as any[]
|
||||
)
|
||||
break
|
||||
} catch (error: any) {
|
||||
if (error.code !== '40P01' || attempt >= 5) throw error
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (weakRows.length > 0) await this.flagConflicts(db.connection(), weakRows)
|
||||
|
||||
return [...idsByKey.values()]
|
||||
})()
|
||||
|
||||
// par2 naming needs a size, which only body-derived segments have; for
|
||||
// header batches these queries were most of Postgres' work.
|
||||
const sized = [...new Set(segments.filter((s) => s.size != null).map((s) => s.fileKey))]
|
||||
if (sized.length > 0) {
|
||||
const sizedIds = (await db.from('binaries').whereIn('file_key', sized).select('id')).map((r) => Number(r.id))
|
||||
await this.matchPar2Releases(sizedIds)
|
||||
await NameService.resolve(sizedIds)
|
||||
}
|
||||
|
||||
// After commit, so the last writer of a binary always sees all of its parts.
|
||||
const completed = await this.claimCompleted(binaryIds)
|
||||
await this.enqueue(completed)
|
||||
return completed
|
||||
}
|
||||
|
||||
/**
|
||||
* Creates or touches the releases referenced by `segments` and returns
|
||||
* their ids by release key. Also used by `release:rebuild`.
|
||||
*/
|
||||
async upsertReleases(
|
||||
knex: Knex,
|
||||
segments: Pick<IngestSegment, 'release' | 'poster' | 'groups' | 'postedAt' | 'weakIdentity'>[],
|
||||
now = new Date()
|
||||
): Promise<Map<string, number>> {
|
||||
const releases = new Map<string, Record<string, unknown>>()
|
||||
for (const segment of segments) {
|
||||
const release = segment.release
|
||||
if (!release) continue
|
||||
const postedAt = validDate(segment.postedAt)
|
||||
const existing = releases.get(release.key)
|
||||
if (existing) {
|
||||
if (postedAt && (!existing.posted_at || postedAt < (existing.posted_at as Date))) {
|
||||
existing.posted_at = postedAt
|
||||
}
|
||||
// Not every file carries the [i/M] counter (e.g. a par2 index): take it from any that does.
|
||||
if (existing.file_total === null && release.fileTotal) existing.file_total = release.fileTotal
|
||||
continue
|
||||
}
|
||||
releases.set(release.key, {
|
||||
release_key: release.key,
|
||||
name: release.name,
|
||||
// Obfuscating posters randomise the poster, so it says nothing about the release.
|
||||
poster: segment.weakIdentity ? null : segment.poster,
|
||||
file_total: release.fileTotal ?? null,
|
||||
source: release.source,
|
||||
groups: JSON.stringify(segment.groups),
|
||||
posted_at: postedAt,
|
||||
first_seen_at: now,
|
||||
last_seen_at: now,
|
||||
})
|
||||
}
|
||||
|
||||
const ids = new Map<string, number>()
|
||||
const rows = [...releases.values()].sort((a, b) =>
|
||||
(a.release_key as string).localeCompare(b.release_key as string)
|
||||
)
|
||||
for (const chunk of chunks(rows, BINARY_CHUNK)) {
|
||||
const result = await upsertKeyed(knex, 'releases', 'release_key', chunk, {
|
||||
last_seen_at: ['touch', 'timestamptz'],
|
||||
posted_at: ['least', 'timestamptz'],
|
||||
file_total: ['coalesce', 'integer'],
|
||||
source: ['coalesce', 'varchar'],
|
||||
})
|
||||
for (const [key, id] of result) ids.set(key, id)
|
||||
}
|
||||
return ids
|
||||
}
|
||||
|
||||
/**
|
||||
* Assigns body-derived binaries without a release and without a
|
||||
* fingerprint to a known par2 set by size: only when that size is listed
|
||||
* once in the set, the file is large, and exactly one set nearby lists it.
|
||||
* Binaries with a fingerprint are matched exactly by NameService instead.
|
||||
*/
|
||||
async matchPar2Releases(binaryIds: number[]) {
|
||||
if (binaryIds.length === 0) return
|
||||
// Listed-once first (index-only on release_files (release_id, size): most candidate
|
||||
// sizes are a set's equal-sized volumes), the release's date only for what's left;
|
||||
// OFFSET 0 keeps that order. Was: every size match's date and count, 18 s per batch.
|
||||
await db.rawQuery(
|
||||
`UPDATE binaries b SET release_id = m.release_id
|
||||
FROM (
|
||||
SELECT c.id, min(c.release_id) AS release_id
|
||||
FROM (
|
||||
SELECT b2.id, b2.posted_at, rf.release_id
|
||||
FROM binaries b2
|
||||
JOIN release_files rf ON rf.size = b2.size
|
||||
WHERE b2.id = ANY(?) AND b2.release_id IS NULL AND b2.md5_16k IS NULL AND b2.size >= ?
|
||||
AND NOT EXISTS (
|
||||
SELECT 1 FROM release_files x
|
||||
WHERE x.release_id = rf.release_id AND x.size = rf.size AND x.par2_file_id <> rf.par2_file_id)
|
||||
OFFSET 0
|
||||
) c
|
||||
JOIN releases r ON r.id = c.release_id
|
||||
WHERE abs(extract(epoch FROM (r.posted_at - c.posted_at))) < ?
|
||||
GROUP BY c.id
|
||||
HAVING count(DISTINCT c.release_id) = 1
|
||||
) m
|
||||
WHERE b.id = m.id`,
|
||||
[binaryIds, PAR2_SIZE_ONLY_MIN, PAR2_MATCH_WINDOW_SECONDS]
|
||||
)
|
||||
}
|
||||
|
||||
/**
|
||||
* Flags binaries where a stored part has a different message-id and a
|
||||
* different part CRC than the incoming one: two files were merged under
|
||||
* one identity. Identical reposts (same CRC) are ignored.
|
||||
*/
|
||||
private async flagConflicts(
|
||||
trx: QueryClientContract,
|
||||
rows: {
|
||||
binary_id: number
|
||||
number: number
|
||||
message_id: string
|
||||
bytes: number
|
||||
article_number: string | null
|
||||
pcrc32: string | null
|
||||
part_size: number | null
|
||||
file_crc32: string | null
|
||||
md5_16k: string | null
|
||||
}[]
|
||||
) {
|
||||
const bindings = [
|
||||
rows.map((r) => r.binary_id),
|
||||
rows.map((r) => r.number),
|
||||
rows.map((r) => r.message_id),
|
||||
rows.map((r) => r.bytes),
|
||||
rows.map((r) => r.article_number),
|
||||
rows.map((r) => r.pcrc32),
|
||||
rows.map((r) => r.part_size),
|
||||
rows.map((r) => r.file_crc32),
|
||||
rows.map((r) => r.md5_16k),
|
||||
]
|
||||
// Keep the extra articles: SplitService separates the volumes from them.
|
||||
const conflicts = await trx.rawQuery(
|
||||
`INSERT INTO part_conflicts (binary_id, number, message_id, bytes, article_number, pcrc32, part_size, file_crc32, md5_16k)
|
||||
SELECT v.binary_id, v.number, v.message_id, v.bytes, v.article_number, v.pcrc32, v.part_size, v.file_crc32, v.md5_16k
|
||||
FROM unnest(?::bigint[], ?::int[], ?::text[], ?::int[], ?::bigint[], ?::text[], ?::int[], ?::text[], ?::text[])
|
||||
AS v(binary_id, number, message_id, bytes, article_number, pcrc32, part_size, file_crc32, md5_16k)
|
||||
JOIN parts p ON p.binary_id = v.binary_id AND p.number = v.number
|
||||
WHERE p.message_id <> v.message_id
|
||||
-- Same part CRC: the same data posted again (a repost), not another file.
|
||||
AND (p.pcrc32 IS NULL OR v.pcrc32 IS NULL OR p.pcrc32 <> v.pcrc32)
|
||||
ON CONFLICT DO NOTHING
|
||||
RETURNING binary_id`,
|
||||
bindings as any
|
||||
)
|
||||
const binaryIds: number[] = [
|
||||
...new Set<number>(conflicts.rows.map((r: { binary_id: string }) => Number(r.binary_id))),
|
||||
]
|
||||
if (binaryIds.length === 0) return
|
||||
const result = await trx.rawQuery(
|
||||
`UPDATE binaries SET conflicted_at = now()
|
||||
WHERE conflicted_at IS NULL AND id = ANY(?) RETURNING name`,
|
||||
[binaryIds]
|
||||
)
|
||||
for (const row of result.rows) {
|
||||
logger.warn(`Binary "${row.name}" has conflicting parts; it will be split into volumes.`)
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Marks complete binaries among `binaryIds` as claimed. The single UPDATE
|
||||
* locks each row, so concurrent callers never claim the same binary twice.
|
||||
*/
|
||||
async claimCompleted(binaryIds: number[]): Promise<CompletedBinary[]> {
|
||||
if (binaryIds.length === 0) return []
|
||||
const result = await db.rawQuery(
|
||||
`UPDATE binaries b SET completed_at = now()
|
||||
WHERE b.id = ANY(?) AND b.completed_at IS NULL AND b.conflicted_at IS NULL
|
||||
AND coalesce(b.part_count, (SELECT count(*) FROM parts p WHERE p.binary_id = b.id)) >= b.total_parts
|
||||
RETURNING b.id, b.file_key`,
|
||||
[binaryIds]
|
||||
)
|
||||
return result.rows.map((row: { id: string; file_key: string }) => ({
|
||||
id: Number(row.id),
|
||||
fileKey: row.file_key,
|
||||
}))
|
||||
}
|
||||
|
||||
async enqueue(completed: CompletedBinary[]) {
|
||||
if (completed.length === 0) return
|
||||
// A job per FILE_BATCH binaries: worker:file stores them with a few set-based queries (LOG.md 2026-10-06 (7)).
|
||||
// jobId from the first file_key: BullMQ ignores an add() for an id that already exists.
|
||||
const jobs = []
|
||||
for (let i = 0; i < completed.length; i += FILE_BATCH) {
|
||||
const batch = completed.slice(i, i + FILE_BATCH)
|
||||
jobs.push({ name: 'process-files', data: { binaryIds: batch.map((b) => b.id) }, opts: { jobId: `${batch[0].fileKey}-${batch.length}` } })
|
||||
}
|
||||
await QueueService.fileQueue.addBulk(jobs)
|
||||
}
|
||||
}
|
||||
|
||||
export default new IngestService()
|
||||
@@ -1,66 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import env from '#start/env'
|
||||
import QueueService from '#services/QueueService'
|
||||
|
||||
/** Files per transaction when moving from the table to BullMQ. */
|
||||
const FEED_BATCH = 5000
|
||||
|
||||
/** Priorities: lower goes first. */
|
||||
export const INSPECTION_PRIORITY = { targeted: -1, normal: 0, low: 10 } as const
|
||||
|
||||
/**
|
||||
* The content inspection backlog (CONCEPT.md D21, D41; LOG.md 2026-10-07 (8)).
|
||||
* Producers add file ids to the inspection_queue table; feed() moves them
|
||||
* into BullMQ's collection-queue while fewer than INSPECTION_FEED_MAX jobs
|
||||
* wait there, where worker:collection and inspecting nodes take them as
|
||||
* before. Redis holds a working set instead of a backlog of days.
|
||||
*/
|
||||
class InspectionQueue {
|
||||
readonly feedMax = env.get('INSPECTION_FEED_MAX', 50_000)
|
||||
|
||||
/** Queues files for inspection; a file already queued keeps the higher (lower number) priority. */
|
||||
async add(fileIds: number[], priority: number = INSPECTION_PRIORITY.normal) {
|
||||
const ids = [...new Set(fileIds)]
|
||||
for (let i = 0; i < ids.length; i += 10_000) {
|
||||
await db.rawQuery(
|
||||
`INSERT INTO inspection_queue (file_id, priority)
|
||||
SELECT id, ?::smallint FROM unnest(?::bigint[]) AS id
|
||||
ON CONFLICT (file_id) DO UPDATE SET priority = least(inspection_queue.priority, excluded.priority)`,
|
||||
[priority, ids.slice(i, i + 10_000)]
|
||||
)
|
||||
}
|
||||
}
|
||||
|
||||
/** Tops BullMQ up to feedMax waiting jobs; returns how many were moved. */
|
||||
async feed(): Promise<number> {
|
||||
const queue = QueueService.collectionQueue
|
||||
const counts = await queue.getJobCounts('wait', 'prioritized')
|
||||
let room = this.feedMax - (counts.wait ?? 0) - (counts.prioritized ?? 0)
|
||||
let moved = 0
|
||||
while (room > 0) {
|
||||
const n = await db.transaction(async (trx) => {
|
||||
const { rows } = await trx.rawQuery(
|
||||
`SELECT file_id FROM inspection_queue ORDER BY priority, queued_at LIMIT ? FOR UPDATE SKIP LOCKED`,
|
||||
[Math.min(room, FEED_BATCH)]
|
||||
)
|
||||
if (rows.length === 0) return 0
|
||||
const ids = rows.map((row: { file_id: string }) => Number(row.file_id))
|
||||
// Added before the rows go: a failure leaves them queued (a job that went in anyway is deduplicated by its id).
|
||||
await queue.addBulk(ids.map((id: number) => ({ name: 'process-collection', data: { fileId: id }, opts: { jobId: `file-${id}` } })))
|
||||
await trx.rawQuery(`DELETE FROM inspection_queue WHERE file_id = ANY(?::bigint[])`, [ids])
|
||||
return ids.length
|
||||
})
|
||||
if (n === 0) break
|
||||
moved += n
|
||||
room -= n
|
||||
}
|
||||
return moved
|
||||
}
|
||||
|
||||
async size(): Promise<number> {
|
||||
const { rows } = await db.rawQuery(`SELECT count(*) AS n FROM inspection_queue`)
|
||||
return Number(rows[0].n)
|
||||
}
|
||||
}
|
||||
|
||||
export default new InspectionQueue()
|
||||
@@ -1,128 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import { looksRandom } from '#services/ArchiveService'
|
||||
|
||||
/**
|
||||
* A release that is nothing more than one randomly named file (its own
|
||||
* filename stem). Such a file moves into the par2 set that names it exactly;
|
||||
* real groupings (par2, RAR, NZB, spot, subject, post id) are left alone.
|
||||
*/
|
||||
export const WEAK_RELEASE_SQL = (column: string) =>
|
||||
`EXISTS (SELECT 1 FROM releases wr WHERE wr.id = ${column} AND wr.source IN ('stem', 'weak-stem')
|
||||
AND wr.file_count <= 1 AND (wr.name !~ '[._ -]' OR NOT is_readable_name(wr.name)))`
|
||||
|
||||
/**
|
||||
* Whether `candidate` (e.g. from a par2 listing) should replace `current`:
|
||||
* the current name is random, or a single token without separators
|
||||
* (mOzyyxCyYUj, F8lv3w5agMDI3; too short for looksRandom), while the
|
||||
* candidate reads like a release name.
|
||||
*/
|
||||
export function isBetterName(current: string, candidate: string): boolean {
|
||||
if (looksRandom(candidate) || !/[._ -]/.test(candidate)) return false
|
||||
return looksRandom(current) || !/[._ -]/.test(current)
|
||||
}
|
||||
|
||||
/** Smaller files (nfo, sfv) collide too easily on size alone. */
|
||||
const MIN_SIZE_ONLY = 64 * 1024
|
||||
|
||||
/**
|
||||
* Gives binaries (and their files) the real filename that a par2 set lists
|
||||
* for them. See CONCEPT.md (D21).
|
||||
* - Exact: size and MD5 of the first 16 KiB both match a par2 entry. This
|
||||
* also puts the binary in that par2 set's release.
|
||||
* - Size only, when the fingerprint is unknown: within the binary's own
|
||||
* release, only when exactly one listed file has that size.
|
||||
*/
|
||||
class NameService {
|
||||
async resolve(binaryIds: number[]): Promise<number> {
|
||||
if (binaryIds.length === 0) return 0
|
||||
const exact = await db.rawQuery(
|
||||
`UPDATE binaries b SET real_name = m.name,
|
||||
release_id = CASE WHEN b.release_id IS NULL OR ${WEAK_RELEASE_SQL('b.release_id')} THEN m.release_id ELSE b.release_id END
|
||||
FROM (
|
||||
SELECT DISTINCT ON (b2.id) b2.id, rf.name, rf.release_id
|
||||
FROM binaries b2
|
||||
JOIN release_files rf ON rf.size = b2.size AND rf.md5_16k = b2.md5_16k
|
||||
WHERE b2.id = ANY(?)
|
||||
ORDER BY b2.id, rf.release_id DESC
|
||||
) m
|
||||
WHERE b.id = m.id AND (b.real_name IS DISTINCT FROM m.name OR b.release_id IS NULL
|
||||
OR (b.release_id <> m.release_id AND ${WEAK_RELEASE_SQL('b.release_id')}))`,
|
||||
[binaryIds]
|
||||
)
|
||||
const bySize = await db.rawQuery(
|
||||
`UPDATE binaries b SET real_name = m.name
|
||||
FROM (
|
||||
SELECT b2.id, min(rf.name) AS name
|
||||
FROM binaries b2
|
||||
JOIN release_files rf ON rf.release_id = b2.release_id AND rf.size = b2.size
|
||||
WHERE b2.id = ANY(?) AND b2.real_name IS NULL AND b2.md5_16k IS NULL AND b2.size >= ?
|
||||
GROUP BY b2.id
|
||||
HAVING count(DISTINCT rf.name) = 1
|
||||
) m
|
||||
WHERE b.id = m.id`,
|
||||
[binaryIds, MIN_SIZE_ONLY]
|
||||
)
|
||||
await db.rawQuery(
|
||||
`UPDATE files f SET real_name = b.real_name,
|
||||
release_id = CASE WHEN f.release_id IS NULL OR ${WEAK_RELEASE_SQL('f.release_id')} THEN coalesce(b.release_id, f.release_id) ELSE f.release_id END
|
||||
FROM binaries b
|
||||
WHERE b.id = ANY(?) AND b.file_id = f.id
|
||||
AND (f.real_name IS DISTINCT FROM b.real_name OR (f.release_id IS NULL AND b.release_id IS NOT NULL)
|
||||
OR (f.release_id <> b.release_id AND ${WEAK_RELEASE_SQL('f.release_id')}))`,
|
||||
[binaryIds]
|
||||
)
|
||||
return (exact.rowCount ?? 0) + (bySize.rowCount ?? 0)
|
||||
}
|
||||
|
||||
/**
|
||||
* When a par2 set arrives: the binaries it can name, i.e. those matching
|
||||
* one of its entries exactly (size + first-16-KiB MD5) and those in its
|
||||
* release. Selecting every binary of the listed sizes instead picked up
|
||||
* hundreds of thousands of equal-sized volumes per set (LOG.md 2026-09-28 (18)).
|
||||
*/
|
||||
async resolveForRelease(releaseId: number): Promise<number> {
|
||||
const { rows } = await db.rawQuery(
|
||||
`SELECT b.id FROM release_files rf
|
||||
JOIN binaries b ON b.md5_16k = rf.md5_16k AND b.size = rf.size
|
||||
WHERE rf.release_id = ?
|
||||
UNION
|
||||
SELECT id FROM binaries WHERE release_id = ? AND size IS NOT NULL`,
|
||||
[releaseId, releaseId]
|
||||
)
|
||||
return this.resolve(rows.map((r: { id: string }) => Number(r.id)))
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Repair for data from before files moved into the par2 set that names them
|
||||
* (used by `release:rename`): regroups exact matches out of random-named
|
||||
* single-file releases, then renames par2 releases that still carry a
|
||||
* random name after their listing. No downloads.
|
||||
*/
|
||||
export async function regroupAndRenameFromPar2(): Promise<{ regrouped: number; renamed: number }> {
|
||||
const { rows } = await db.rawQuery(
|
||||
`SELECT DISTINCT b.id FROM binaries b
|
||||
JOIN release_files rf ON rf.md5_16k = b.md5_16k AND rf.size = b.size
|
||||
WHERE b.release_id IS DISTINCT FROM rf.release_id
|
||||
AND (b.release_id IS NULL OR ${WEAK_RELEASE_SQL('b.release_id')})`
|
||||
)
|
||||
const ids = rows.map((r: { id: string }) => Number(r.id))
|
||||
for (let i = 0; i < ids.length; i += 1000) await nameService.resolve(ids.slice(i, i + 1000))
|
||||
|
||||
const { par2ReleaseName } = await import('#lib/release/index')
|
||||
const listings = await db.rawQuery(
|
||||
`SELECT r.id, r.name, array_agg(rf.name) AS names FROM releases r
|
||||
JOIN release_files rf ON rf.release_id = r.id WHERE r.source <> 'spot' GROUP BY r.id, r.name`
|
||||
)
|
||||
let renamed = 0
|
||||
for (const release of listings.rows) {
|
||||
const better = par2ReleaseName(release.names)
|
||||
if (!better || better === release.name || !isBetterName(release.name, better)) continue
|
||||
await db.from('releases').where('id', release.id).update({ name: better })
|
||||
renamed++
|
||||
}
|
||||
return { regrouped: ids.length, renamed }
|
||||
}
|
||||
|
||||
const nameService = new NameService()
|
||||
export default nameService
|
||||
@@ -1,141 +0,0 @@
|
||||
import nntpConfig from '#config/nntp'
|
||||
import { NNTP } from "#lib/nntp/nntp_js";
|
||||
import { parseGroupResponse, type GroupInfo } from '#lib/nntp/group'
|
||||
|
||||
class NntpService {
|
||||
private readonly poolSize: number;
|
||||
private allConnections: Set<any>;
|
||||
private idleConnections: any[];
|
||||
private waiters: ((conn: any) => void)[];
|
||||
/** Background work (archive walks): served only when no one else waits. */
|
||||
private lowWaiters: ((conn: any) => void)[] = [];
|
||||
private createdCount: number;
|
||||
|
||||
constructor(poolSize = 10) {
|
||||
this.poolSize = poolSize;
|
||||
this.allConnections = new Set();
|
||||
this.idleConnections = [];
|
||||
this.waiters = [];
|
||||
this.createdCount = 0;
|
||||
}
|
||||
|
||||
private async _createConnection() {
|
||||
const conn = new NNTP(nntpConfig.host, nntpConfig.port);
|
||||
try {
|
||||
await conn.connect();
|
||||
if (nntpConfig.user) {
|
||||
await conn.login(nntpConfig.user, nntpConfig.password?.release());
|
||||
}
|
||||
} catch (error) {
|
||||
// Free the slot, otherwise a failed connect shrinks the pool for good.
|
||||
this.createdCount--;
|
||||
this._destroy(conn);
|
||||
throw error;
|
||||
}
|
||||
this.allConnections.add(conn);
|
||||
return conn;
|
||||
}
|
||||
|
||||
/** A connection from the pool; with `low`, after every waiting normal caller. */
|
||||
public async acquire(low = false) {
|
||||
if (this.idleConnections.length > 0) {
|
||||
return this.idleConnections.pop();
|
||||
}
|
||||
|
||||
if (this.createdCount < this.poolSize) {
|
||||
this.createdCount++;
|
||||
console.log(`Creating new connection (${this.createdCount}/${this.poolSize}).`);
|
||||
return this._createConnection();
|
||||
}
|
||||
|
||||
return new Promise(resolve => (low ? this.lowWaiters : this.waiters).push(resolve));
|
||||
}
|
||||
|
||||
/**
|
||||
* Returns a connection to the pool. Pass `failed` when a command on it
|
||||
* threw: the connection may be dead (e.g. closed by the server while idle),
|
||||
* so it is discarded instead of being handed to the next job.
|
||||
*/
|
||||
public release(conn: any, failed = false) {
|
||||
if (failed) {
|
||||
this.discard(conn);
|
||||
return;
|
||||
}
|
||||
const resolve = this.nextWaiter();
|
||||
if (resolve) {
|
||||
resolve(conn);
|
||||
} else {
|
||||
this.idleConnections.push(conn);
|
||||
}
|
||||
}
|
||||
|
||||
/** Closes a broken connection and frees its slot for a new one. */
|
||||
public discard(conn: any) {
|
||||
this.allConnections.delete(conn);
|
||||
this.createdCount--;
|
||||
this._destroy(conn);
|
||||
|
||||
// Someone is waiting for a slot: open a fresh connection for them.
|
||||
const resolve = this.nextWaiter();
|
||||
if (resolve) {
|
||||
this.createdCount++;
|
||||
this._createConnection().then(resolve, () => this.waiters.unshift(resolve));
|
||||
}
|
||||
}
|
||||
|
||||
private nextWaiter() {
|
||||
return this.waiters.shift() ?? this.lowWaiters.shift();
|
||||
}
|
||||
|
||||
private _destroy(conn: any) {
|
||||
try {
|
||||
conn.sock?.destroy();
|
||||
} catch {
|
||||
// already closed
|
||||
}
|
||||
}
|
||||
|
||||
/** Selects a group and returns its correctly parsed article range. */
|
||||
public async selectGroup(conn: any, name: string): Promise<GroupInfo> {
|
||||
const { resp } = await conn.group(name)
|
||||
return parseGroupResponse(resp)
|
||||
}
|
||||
|
||||
/**
|
||||
* Rejects when `promise` hasn't settled within `ms`. nntp-js has no
|
||||
* timeouts: a connection the server stops answering (still ESTABLISHED,
|
||||
* nothing in flight) leaves its command waiting forever. The caller must
|
||||
* then release the connection as failed, so it's destroyed.
|
||||
*/
|
||||
public timed<T>(promise: Promise<T>, what: string, ms = nntpConfig.commandTimeoutMs): Promise<T> {
|
||||
let timer: NodeJS.Timeout
|
||||
const timeout = new Promise<never>((_, reject) => {
|
||||
timer = setTimeout(() => reject(new Error(`NNTP ${what}: no answer within ${ms / 1000} s`)), ms)
|
||||
})
|
||||
return Promise.race([promise, timeout]).finally(() => clearTimeout(timer))
|
||||
}
|
||||
|
||||
/** Fetches a raw (still dot-stuffed) article body as a Buffer. */
|
||||
public async fetchBody(conn: any, messageId: string): Promise<Buffer> {
|
||||
const id = messageId.startsWith('<') ? messageId : `<${messageId}>`
|
||||
const { data } = await conn.body(id)
|
||||
return Buffer.from(data.buffer, data.byteOffset, data.byteLength)
|
||||
}
|
||||
|
||||
public async shutdown() {
|
||||
console.log('Shutting down all connections in the pool.');
|
||||
// A connection the server already closed may never answer QUIT.
|
||||
const quit = (conn: any) =>
|
||||
Promise.race([conn.quit(), new Promise((resolve) => setTimeout(resolve, 2000).unref())])
|
||||
.catch(() => {})
|
||||
.finally(() => this._destroy(conn));
|
||||
await Promise.all([...this.allConnections].map(quit));
|
||||
this.allConnections.clear();
|
||||
this.idleConnections.length = 0;
|
||||
this.waiters.length = 0;
|
||||
this.lowWaiters.length = 0;
|
||||
this.createdCount = 0;
|
||||
}
|
||||
}
|
||||
|
||||
export default new NntpService(nntpConfig.poolSize);
|
||||
@@ -1,229 +0,0 @@
|
||||
import { createHash } from 'node:crypto'
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
import IngestService from '#services/IngestService'
|
||||
import { type ReleaseIdentity } from '#lib/release/index'
|
||||
import { parseNzb, type ParsedNzb } from '#lib/nzb/index'
|
||||
import { parseSubject } from '#lib/subject/index'
|
||||
import { fileStem } from '#lib/release/index'
|
||||
import { looksRandom } from '#services/ArchiveService'
|
||||
import FileStore from '#services/FileStore'
|
||||
import { isArchiveStart } from '#lib/content/index'
|
||||
import InspectionQueue from '#services/InspectionQueue'
|
||||
|
||||
export interface NzbImportResult {
|
||||
releaseId: number
|
||||
releaseName: string
|
||||
listed: number
|
||||
known: number
|
||||
added: number
|
||||
password: boolean
|
||||
}
|
||||
|
||||
/** Filename of an NZB entry, from its subject (tolerating odd quoting). */
|
||||
export function nzbFileName(subject: string): string {
|
||||
const parsed = parseSubject(subject)
|
||||
const quoted = parsed?.name ?? /"([^"]+)"/.exec(subject)?.[1]
|
||||
if (quoted) return quoted.trim().replace(/^"+|"+$/g, '').trim() || subject.trim()
|
||||
return unquotedFileName(subject)
|
||||
}
|
||||
|
||||
/**
|
||||
* A filename from a subject without quotes, as some spot NZBs have:
|
||||
* `Adventure.part0055.rar - [056/175] (1/1)`, `x.vol007+008.par2 (1/0) (1/0)`,
|
||||
* `06 - Song.flac Artist - Album (2020) (01/35)`. Counters, "yEnc" and a
|
||||
* trailing size are dropped; then the name ends at the first extension
|
||||
* followed by a space.
|
||||
*/
|
||||
export function unquotedFileName(subject: string): string {
|
||||
const cleaned = subject
|
||||
.replace(/\(\d+\/\d+\)|\[\d+\/\d+\]|\byEnc\b/gi, ' ')
|
||||
.replace(/\s\d+\s*$/, ' ')
|
||||
.replace(/(\s+-)+\s*$/, '')
|
||||
.replace(/^\s*-\s+/, '')
|
||||
.replace(/\s+/g, ' ')
|
||||
.trim()
|
||||
const end = /\.(?:[A-Za-z][A-Za-z0-9]{1,4}|\d{3})(?=\s)/.exec(cleaned)
|
||||
const name = end ? cleaned.slice(0, end.index + end[0].length) : cleaned
|
||||
return name.replace(/\s+-$/, '').trim() || subject.trim()
|
||||
}
|
||||
|
||||
export interface NzbImportOptions {
|
||||
/**
|
||||
* For NZBs inside an archive: the entry's name. It keeps several NZBs of
|
||||
* one archive apart and names their release when the NZB has no name.
|
||||
*/
|
||||
entryName?: string
|
||||
/** Move the posted file into the release the NZB describes (default true). */
|
||||
moveContainer?: boolean
|
||||
/** A name given with the NZB (e.g. a Spotnet title); wins over the NZB's own. */
|
||||
name?: string
|
||||
/** Release source for new releases (default `nzb`). */
|
||||
source?: ReleaseIdentity['source']
|
||||
/**
|
||||
* The NZB is authoritative for which files form the release and what it is
|
||||
* called (Spotnet spots): its files get the NZB's own release, named
|
||||
* `name`, also when name rules had put them in another release. Files
|
||||
* another trusted NZB claimed first stay where they are.
|
||||
*/
|
||||
trusted?: boolean
|
||||
}
|
||||
|
||||
/**
|
||||
* Reads an NZB that was posted to Usenet into the index (CONCEPT.md D26).
|
||||
* Files it lists that are already known (by first segment) keep their
|
||||
* release, which the NZB then joins; the others are added as files, so posts
|
||||
* spread over groups we do not index still become complete.
|
||||
*/
|
||||
class NzbImportService {
|
||||
async import(nzbFile: any, xml: string, options: NzbImportOptions = {}): Promise<NzbImportResult | null> {
|
||||
const nzb = parseNzb(xml)
|
||||
if (!nzb || nzb.files.length === 0) return null
|
||||
|
||||
const entries = nzb.files.map((f) => ({
|
||||
file: f,
|
||||
name: nzbFileName(f.subject),
|
||||
firstId: f.segments[0].id,
|
||||
}))
|
||||
|
||||
const added: { id: number; filename: string }[] = []
|
||||
const result = await db.transaction(async (trx) => {
|
||||
const known = await trx
|
||||
.from('files')
|
||||
.whereIn('first_message_id', entries.map((e) => e.firstId))
|
||||
.select('id', 'release_id', 'first_message_id')
|
||||
const knownIds = new Set(known.map((f) => f.first_message_id))
|
||||
|
||||
// Join the release the listed files are already in, else the posted NZB's own, else a new one.
|
||||
const counts = new Map<number, number>()
|
||||
for (const f of known) if (f.release_id) counts.set(Number(f.release_id), (counts.get(Number(f.release_id)) ?? 0) + 1)
|
||||
let releaseId: number | undefined = [...counts].sort((a, b) => b[1] - a[1])[0]?.[0]
|
||||
const title = options.name?.trim() || this.title(nzb, nzbFile, entries.map((e) => e.name), options.entryName)
|
||||
const origin = `nzb\n${nzbFile.first_message_id ?? nzbFile.id}`
|
||||
const key = createHash('sha1').update(options.entryName ? `${origin}\n${options.entryName}` : origin).digest('hex')
|
||||
|
||||
// Releases of other trusted NZBs: their files are not taken over.
|
||||
let claimed = new Set<number>()
|
||||
if (options.trusted) {
|
||||
// Only while that NZB still stands: a disposed spot's files go to its correction.
|
||||
const holders = await trx
|
||||
.from('releases as r')
|
||||
.whereIn('r.id', [...counts.keys()])
|
||||
.where('r.source', 'spot')
|
||||
.whereNot('r.release_key', key)
|
||||
.whereNotExists((q) => q.from('spots as s').whereRaw('s.release_id = r.id').whereIn('s.status', ['removed', 'dummy']))
|
||||
.select('r.id')
|
||||
claimed = new Set(holders.map((r) => Number(r.id)))
|
||||
// Join another spot's release when it holds most of these files (a repost); else use our own.
|
||||
if (releaseId && !claimed.has(releaseId)) releaseId = undefined
|
||||
if (releaseId) claimed.delete(releaseId)
|
||||
}
|
||||
if (!releaseId) {
|
||||
const ids = await IngestService.upsertReleases(trx.knexClient, [
|
||||
{
|
||||
release: { key, name: title, source: options.source ?? 'nzb' },
|
||||
poster: nzbFile.poster,
|
||||
groups: nzbFile.groups ?? [],
|
||||
postedAt: nzbFile.date ? new Date(Number(nzbFile.date)) : undefined,
|
||||
},
|
||||
])
|
||||
releaseId = ids.get(key)!
|
||||
}
|
||||
|
||||
const now = new Date()
|
||||
const missing = entries.filter((e) => !knownIds.has(e.firstId))
|
||||
for (let i = 0; i < missing.length; i += 500) {
|
||||
added.push(
|
||||
...(await FileStore.insert(
|
||||
missing.slice(i, i + 500).map(({ file, name, firstId }) => ({
|
||||
file_key: createHash('sha1').update(`nzb-file\n${firstId}`).digest('hex'),
|
||||
filename: name,
|
||||
poster: file.poster || nzbFile.poster,
|
||||
date: (file.date || Math.floor(Number(nzbFile.date) / 1000)) * 1000,
|
||||
parts: file.segments.length,
|
||||
size: file.segments.reduce((sum, s) => sum + s.bytes, 0),
|
||||
segments: file.segments.map((s) => ({ number: s.number, id: s.id, size: s.bytes })),
|
||||
groups: JSON.stringify(file.groups),
|
||||
release_id: releaseId,
|
||||
created_at: now,
|
||||
updated_at: now,
|
||||
})),
|
||||
trx
|
||||
))
|
||||
)
|
||||
}
|
||||
// The poster's NZB says which files form the post: known files join this
|
||||
// release, also when name rules had split them over several.
|
||||
const moving = known.filter((f) => Number(f.release_id) !== releaseId && !claimed.has(Number(f.release_id)))
|
||||
const previous = [...new Set(moving.map((f) => f.release_id).filter(Boolean))]
|
||||
await trx
|
||||
.from('files')
|
||||
.whereIn('id', moving.map((f) => f.id))
|
||||
.update({ release_id: releaseId })
|
||||
if (previous.length) {
|
||||
await trx.rawQuery(
|
||||
`UPDATE binaries b SET release_id = ? FROM files f WHERE b.file_id = f.id AND f.release_id = ? AND b.release_id IS DISTINCT FROM ?`,
|
||||
[releaseId, releaseId, releaseId]
|
||||
)
|
||||
await trx.from('releases').whereIn('id', previous).where('file_count', 0).delete()
|
||||
}
|
||||
// The NZB file joins the release it describes, unless it is part of a real release already.
|
||||
const own = nzbFile.release_id ? Number(nzbFile.release_id) : null
|
||||
if (own !== releaseId && options.moveContainer !== false) {
|
||||
const ownRelease = own ? await trx.from('releases').where('id', own).first() : null
|
||||
if (!ownRelease || ownRelease.file_count <= 1) {
|
||||
await trx.from('files').where('id', nzbFile.id).update({ release_id: releaseId })
|
||||
if (own) await trx.from('releases').where('id', own).where('file_count', 0).delete()
|
||||
}
|
||||
}
|
||||
|
||||
const release = await trx.from('releases').where('id', releaseId).first()
|
||||
const update: Record<string, unknown> = {}
|
||||
if (nzb.meta.password) Object.assign(update, { password: nzb.meta.password, password_source: 'nzb' })
|
||||
if (!release.file_total) update.file_total = entries.length
|
||||
if (options.trusted) {
|
||||
// Only its own release: a repost joining another spot's release keeps that one's title.
|
||||
if (release.release_key === key && release.name !== title) update.name = title
|
||||
if (release.release_key === key && options.source && release.source !== options.source) update.source = options.source
|
||||
} else if ((looksRandom(release.name) || release.source === 'post') && !looksRandom(title)) {
|
||||
update.name = title
|
||||
}
|
||||
if (Object.keys(update).length) await trx.from('releases').where('id', releaseId).update(update)
|
||||
|
||||
const result = {
|
||||
releaseId,
|
||||
releaseName: (update.name as string) ?? release.name,
|
||||
listed: entries.length,
|
||||
known: known.length,
|
||||
added: missing.length,
|
||||
password: Boolean(nzb.meta.password),
|
||||
}
|
||||
logger.info(
|
||||
`NZB "${options.entryName ? `${nzbFile.filename}:${options.entryName}` : nzbFile.filename}" → release ${releaseId} "${result.releaseName}": ${result.listed} files listed, ` +
|
||||
`${result.known} known, ${result.added} added${result.password ? ', with password' : ''}.`
|
||||
)
|
||||
return result
|
||||
})
|
||||
// Their first segment tells what the archives hold (contents, password); nothing else reads it for files from an NZB.
|
||||
const archives = added.filter((f) => isArchiveStart(f.filename))
|
||||
if (archives.length) {
|
||||
await InspectionQueue.add(archives.map((f) => Number(f.id)))
|
||||
}
|
||||
return result
|
||||
}
|
||||
|
||||
/** Release name: the NZB's meta name, else the NZB's (entry or file) name, else its largest file's stem. */
|
||||
private title(nzb: ParsedNzb, nzbFile: any, names: string[], entryName?: string): string {
|
||||
if (nzb.meta.name) return nzb.meta.name
|
||||
const entry = entryName ? fileStem(entryName.split(/[\\/]/).pop() ?? entryName) : ''
|
||||
if (entry && !looksRandom(entry)) return entry
|
||||
const own = fileStem(String(nzbFile.real_name ?? nzbFile.filename ?? ''))
|
||||
if (own && !looksRandom(own)) return own
|
||||
const largest = nzb.files
|
||||
.map((f, i) => ({ name: names[i], bytes: f.segments.reduce((s, x) => s + x.bytes, 0) }))
|
||||
.sort((a, b) => b.bytes - a.bytes)[0]
|
||||
return fileStem(largest?.name ?? own)
|
||||
}
|
||||
}
|
||||
|
||||
export default new NzbImportService()
|
||||
@@ -1,133 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
import IngestService, { PAR2_MATCH_WINDOW_SECONDS } from '#services/IngestService'
|
||||
import type { Par2Set } from '#lib/par2/index'
|
||||
import { par2ReleaseIdentity, par2ReleaseName } from '#lib/release/index'
|
||||
import NameService, { WEAK_RELEASE_SQL, isBetterName } from '#services/NameService'
|
||||
|
||||
/** Size-only matching below this is too ambiguous. */
|
||||
const SIZE_ONLY_MIN = 1024 * 1024
|
||||
|
||||
function countBy<T>(values: T[]): Map<T, number> {
|
||||
const counts = new Map<T, number>()
|
||||
for (const value of values) counts.set(value, (counts.get(value) ?? 0) + 1)
|
||||
return counts
|
||||
}
|
||||
|
||||
export interface Par2Result {
|
||||
releaseId: number
|
||||
releaseName: string
|
||||
listed: number
|
||||
matched: number
|
||||
}
|
||||
|
||||
/**
|
||||
* Recognises releases of obfuscated posts from their par2 files: the par2
|
||||
* lists the real name and exact size of every file in the set, and
|
||||
* body-derived binaries carry their exact size. See CONCEPT.md (D16).
|
||||
*/
|
||||
class Par2Service {
|
||||
async attach(binary: any, set: Par2Set): Promise<Par2Result> {
|
||||
const result = await db.transaction(async (trx) => {
|
||||
// Exact: size and first-16-KiB MD5 of a listed file. Only these may pull
|
||||
// the set into a release the files already belong to; size alone is
|
||||
// ambiguous across posts (standard volume sizes recur).
|
||||
const exact = await trx
|
||||
.from('binaries')
|
||||
.whereIn(['size', 'md5_16k'] as any, set.files.map((f) => [f.length, f.md5_16k]) as any)
|
||||
.select('id', 'release_id')
|
||||
|
||||
// Size only, for files without a fingerprint: the size must be listed
|
||||
// once in this set, be large, and belong to exactly one binary nearby.
|
||||
const listedOnce = [...countBy(set.files.map((f) => f.length))]
|
||||
.filter(([size, n]) => n === 1 && size >= SIZE_ONLY_MIN)
|
||||
.map(([size]) => size)
|
||||
const bySize = listedOnce.length
|
||||
? await trx
|
||||
.from('binaries')
|
||||
.whereIn('size', listedOnce)
|
||||
.whereNull('md5_16k')
|
||||
.whereRaw('abs(extract(epoch FROM (posted_at - ?::timestamptz))) < ?', [
|
||||
binary.posted_at,
|
||||
PAR2_MATCH_WINDOW_SECONDS,
|
||||
])
|
||||
.select('id', 'release_id', 'size')
|
||||
: []
|
||||
const sizeCounts = countBy(bySize.map((b) => Number(b.size)))
|
||||
const fuzzy = bySize.filter((b) => sizeCounts.get(Number(b.size)) === 1)
|
||||
const candidates = [...exact, ...fuzzy]
|
||||
|
||||
// Prefer a release the exactly matched files are already in (e.g.
|
||||
// obfuscated volumes grouped by their stem); otherwise the set's own.
|
||||
const counts = countBy(exact.filter((c) => c.release_id).map((c) => Number(c.release_id)))
|
||||
let releaseId = [...counts].sort((a, b) => b[1] - a[1])[0]?.[0]
|
||||
const identity = par2ReleaseIdentity(
|
||||
set.recoverySetId,
|
||||
set.files.map((f) => f.name)
|
||||
)
|
||||
if (!releaseId) {
|
||||
const ids = await IngestService.upsertReleases(trx.knexClient, [
|
||||
{
|
||||
release: identity,
|
||||
poster: binary.poster,
|
||||
weakIdentity: true,
|
||||
groups: binary.groups,
|
||||
postedAt: binary.posted_at ? new Date(binary.posted_at) : undefined,
|
||||
},
|
||||
])
|
||||
releaseId = ids.get(identity.key)!
|
||||
}
|
||||
|
||||
await trx
|
||||
.knexClient('release_files')
|
||||
.insert(
|
||||
set.files.map((f) => ({
|
||||
release_id: releaseId,
|
||||
par2_file_id: f.fileId,
|
||||
name: f.name,
|
||||
size: f.length,
|
||||
md5_16k: f.md5_16k,
|
||||
md5: f.md5,
|
||||
set_id: set.recoverySetId,
|
||||
}))
|
||||
)
|
||||
.onConflict(['release_id', 'par2_file_id'])
|
||||
.ignore()
|
||||
|
||||
const matchIds = candidates.filter((c) => !c.release_id).map((c) => Number(c.id))
|
||||
await trx.from('binaries').whereIn('id', [binary.id, ...matchIds]).whereNull('release_id').update({ release_id: releaseId })
|
||||
// Exact matches sitting alone in a random-named release (one file of a
|
||||
// season, each posted under its own random name) join the set's release.
|
||||
await trx.rawQuery(
|
||||
`UPDATE binaries b SET release_id = ? WHERE b.id = ANY(?) AND b.release_id <> ? AND ${WEAK_RELEASE_SQL('b.release_id')}`,
|
||||
[releaseId, exact.map((c) => Number(c.id)), releaseId]
|
||||
)
|
||||
await trx.rawQuery(
|
||||
`UPDATE files f SET release_id = b.release_id FROM binaries b
|
||||
WHERE b.file_id = f.id AND b.release_id = ? AND f.release_id IS DISTINCT FROM b.release_id`,
|
||||
[releaseId]
|
||||
)
|
||||
|
||||
// The set may have joined a release named after one random filename.
|
||||
const release = await trx.from('releases').where('id', releaseId).first()
|
||||
let releaseName = release?.name
|
||||
const better = par2ReleaseName(set.files.map((f) => f.name))
|
||||
// A spot's title is kept (CONCEPT.md D29).
|
||||
if (releaseName && better && release.source !== 'spot' && isBetterName(releaseName, better)) {
|
||||
await trx.from('releases').where('id', releaseId).update({ name: better })
|
||||
releaseName = better
|
||||
}
|
||||
logger.info(
|
||||
`Par2 set ${set.recoverySetId.slice(0, 8)} → release ${releaseId} "${releaseName}": ` +
|
||||
`${set.files.length} files listed, ${matchIds.length} binaries matched by size.`
|
||||
)
|
||||
return { releaseId, releaseName, listed: set.files.length, matched: matchIds.length }
|
||||
})
|
||||
|
||||
// Real names for every known file of the set (exact fingerprint, or size within the release).
|
||||
await NameService.resolveForRelease(result.releaseId)
|
||||
return result
|
||||
}
|
||||
}
|
||||
|
||||
export default new Par2Service()
|
||||
@@ -1,65 +0,0 @@
|
||||
import { Queue, type DefaultJobOptions } from 'bullmq'
|
||||
import queueConfig from '#config/queue'
|
||||
import gcConfig from '#config/gc'
|
||||
|
||||
const defaultJobOptions: DefaultJobOptions = {
|
||||
// Transient NNTP errors (e.g. a pooled connection the server closed) are retried.
|
||||
attempts: 3,
|
||||
backoff: { type: 'exponential', delay: 2000 },
|
||||
removeOnComplete: {
|
||||
age: gcConfig.jobs.completedAgeSeconds,
|
||||
count: gcConfig.jobs.completedMaxCount,
|
||||
},
|
||||
removeOnFail: { age: gcConfig.jobs.failedAgeSeconds },
|
||||
}
|
||||
|
||||
/**
|
||||
* Queues are created (and connect to Redis) on first use. Ace imports every
|
||||
* command at boot, so eager queues would keep one-off commands from exiting.
|
||||
*/
|
||||
class QueueService {
|
||||
private readonly queues = new Map<string, Queue>()
|
||||
|
||||
private queue(name: string): Queue {
|
||||
let queue = this.queues.get(name)
|
||||
if (!queue) {
|
||||
queue = new Queue(name, { connection: queueConfig.connection, defaultJobOptions })
|
||||
this.queues.set(name, queue)
|
||||
}
|
||||
return queue
|
||||
}
|
||||
|
||||
get nntpFetchQueue() {
|
||||
return this.queue('nntp-fetch-queue')
|
||||
}
|
||||
get fileQueue() {
|
||||
return this.queue('file-queue')
|
||||
}
|
||||
get bodyQueue() {
|
||||
return this.queue('body-queue')
|
||||
}
|
||||
get collectionQueue() {
|
||||
return this.queue('collection-queue')
|
||||
}
|
||||
/** Complete archive listings, at low priority (ArchiveWalkService). */
|
||||
get archiveWalkQueue() {
|
||||
return this.queue('archive-walk-queue')
|
||||
}
|
||||
|
||||
/** Queues no longer used by the pipeline, kept here so GC can empty them. */
|
||||
get legacy(): Queue[] {
|
||||
return [this.queue('header-queue')]
|
||||
}
|
||||
|
||||
get all(): Queue[] {
|
||||
return [this.nntpFetchQueue, this.fileQueue, this.bodyQueue, this.collectionQueue, this.archiveWalkQueue]
|
||||
}
|
||||
|
||||
/** Closes the queues that were opened. */
|
||||
async closeAll() {
|
||||
await Promise.all([...this.queues.values()].map((queue) => queue.close()))
|
||||
this.queues.clear()
|
||||
}
|
||||
}
|
||||
|
||||
export default new QueueService()
|
||||
@@ -1,25 +0,0 @@
|
||||
import { createRequire } from 'module'
|
||||
import redisConfig from '#config/redis'
|
||||
|
||||
const require = createRequire(import.meta.url)
|
||||
const IORedis = require('ioredis')
|
||||
|
||||
/**
|
||||
* Connects on first use. Ace imports every command at boot, so an eager
|
||||
* connection would keep even `node ace list` from exiting.
|
||||
*/
|
||||
class RedisService {
|
||||
private instance?: any
|
||||
|
||||
public get client(): any {
|
||||
this.instance ??= new IORedis(redisConfig.connections.main)
|
||||
return this.instance
|
||||
}
|
||||
|
||||
async close() {
|
||||
if (this.instance) await this.instance.quit()
|
||||
this.instance = undefined
|
||||
}
|
||||
}
|
||||
|
||||
export default new RedisService()
|
||||
@@ -1,35 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
|
||||
/** Noted releases taken per transaction. */
|
||||
const BATCH = 2000
|
||||
|
||||
/**
|
||||
* Release completeness (D24) for releases whose files or par2 listing
|
||||
* changed: triggers only note them in release_dirty (a release can be noted
|
||||
* thousands of times between sweeps), and gc:scheduler recomputes each
|
||||
* noted release once every RELEASE_COMPLETE_INTERVAL_MS (LOG.md 2026-10-06 (6)).
|
||||
*/
|
||||
class ReleaseCompleteness {
|
||||
/** Recomputes all noted releases; returns how many distinct releases were checked. */
|
||||
async sweep(): Promise<number> {
|
||||
let checked = 0
|
||||
for (;;) {
|
||||
const done = await db.transaction(async (trx) => {
|
||||
// Taken and recomputed in one transaction: a failure leaves the notes for the next sweep.
|
||||
const { rows } = await trx.rawQuery(
|
||||
`DELETE FROM release_dirty WHERE ctid = ANY(ARRAY(SELECT ctid FROM release_dirty LIMIT ?)) RETURNING release_id`,
|
||||
[BATCH]
|
||||
)
|
||||
const ids: number[] = [...new Set<number>(rows.map((row: { release_id: string }) => Number(row.release_id)))]
|
||||
if (ids.length > 0) {
|
||||
await trx.rawQuery(`SELECT recompute_release_complete(id) FROM unnest(?::bigint[]) AS id`, [ids])
|
||||
}
|
||||
checked += ids.length
|
||||
return rows.length < BATCH
|
||||
})
|
||||
if (done) return checked
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
export default new ReleaseCompleteness()
|
||||
@@ -1,42 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
|
||||
/**
|
||||
* Uploads split per poster (CONCEPT.md D35). Some uploaders post every
|
||||
* volume of a RAR set under a different poster; the post-id grouping then
|
||||
* makes one release per volume, all with the same name
|
||||
* (`Masters.of.the.Universe.1987…`: 191 one-file releases of one 199-volume
|
||||
* set). Such releases are one upload when all their files share one stem and
|
||||
* no file name repeats (a repost would repeat them, another upload would
|
||||
* bring another stem); their files move into the oldest of them, and GC
|
||||
* deletes the emptied ones.
|
||||
*/
|
||||
class ReleaseMergeService {
|
||||
async mergeSplitPosts({ dryRun = false } = {}): Promise<{ names: number; releases: number }> {
|
||||
const { rows } = await db.rawQuery(`
|
||||
WITH dup AS (
|
||||
SELECT name FROM releases
|
||||
WHERE file_count > 0 AND source = 'post' AND is_readable_name(name)
|
||||
GROUP BY name HAVING count(*) > 1
|
||||
)
|
||||
SELECT r.name, array_agg(DISTINCT r.id ORDER BY r.id) AS ids
|
||||
FROM dup JOIN releases r ON r.name = dup.name AND r.file_count > 0 AND r.source = 'post'
|
||||
JOIN files f ON f.release_id = r.id
|
||||
GROUP BY r.name
|
||||
HAVING count(DISTINCT name_stem(coalesce(f.real_name, f.filename))) = 1
|
||||
AND count(f.id) = count(DISTINCT coalesce(f.real_name, f.filename))
|
||||
AND max(r.posted_at) - min(r.posted_at) < interval '30 days'`)
|
||||
let releases = 0
|
||||
for (const { name, ids } of rows as { name: string; ids: (string | number)[] }[]) {
|
||||
const [target, ...others] = ids.map(Number)
|
||||
releases += others.length
|
||||
if (dryRun) continue
|
||||
await db.from('files').whereIn('release_id', others).update({ release_id: target })
|
||||
await db.from('binaries').whereIn('release_id', others).update({ release_id: target })
|
||||
logger.info(`Merged ${others.length + 1} releases of one upload split per poster into ${target} "${name}".`)
|
||||
}
|
||||
return { names: rows.length, releases }
|
||||
}
|
||||
}
|
||||
|
||||
export default new ReleaseMergeService()
|
||||
@@ -1,168 +0,0 @@
|
||||
import { randomUUID } from 'node:crypto'
|
||||
import { Job, Worker } from 'bullmq'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
import queueConfig from '#config/queue'
|
||||
import nntpConfig from '#config/nntp'
|
||||
import QueueService from '#services/QueueService'
|
||||
import NntpService from '#services/NntpService'
|
||||
import ContentService, { describeContent } from '#services/ContentService'
|
||||
import ArchiveWalkService from '#services/ArchiveWalkService'
|
||||
import { isMissingArticle, type ContentTask, type InspectProbe, type InspectTask, type WalkProbe } from '#lib/content/probe'
|
||||
|
||||
/** Inspection first; archive walks (D40) only when it has nothing waiting. */
|
||||
const QUEUES = ['collection-queue', 'archive-walk-queue'] as const
|
||||
type QueueName = (typeof QUEUES)[number]
|
||||
|
||||
/** A node has this long to post a job's result; after that the job goes back to the queue (BullMQ's stalled check). */
|
||||
const LOCK_MS = 10 * 60 * 1000
|
||||
|
||||
export interface RemoteJob {
|
||||
queue: QueueName
|
||||
jobId: string
|
||||
/** The job's lock: whoever holds it may complete the job. */
|
||||
token: string
|
||||
task: ContentTask
|
||||
}
|
||||
|
||||
export class JobLost extends Error {}
|
||||
export class ArticleRelayOff extends Error {}
|
||||
|
||||
/**
|
||||
* Content inspection and archive walks on seeding nodes (CONCEPT.md D41).
|
||||
* Nodes take jobs from the same queues worker:collection works on: a lease
|
||||
* locks a job (BullMQ's manual processing) and gives the node a task that
|
||||
* was prepared from the database; the node probes it, fetching articles
|
||||
* through the relay (articles()), and posts the result, which is applied
|
||||
* here. A job whose node goes quiet is unlocked after LOCK_MS and picked up
|
||||
* again by whoever comes next.
|
||||
*/
|
||||
class RemoteContentService {
|
||||
private readonly workers = new Map<QueueName, Worker>()
|
||||
|
||||
/** Up to `max` jobs for a node, each locked for LOCK_MS. */
|
||||
async lease(nodeId: number, max: number): Promise<RemoteJob[]> {
|
||||
const jobs: RemoteJob[] = []
|
||||
for (const queue of QUEUES) {
|
||||
while (jobs.length < max) {
|
||||
const token = `node-${nodeId}:${randomUUID()}`
|
||||
const job = await this.worker(queue).getNextJob(token, { block: false })
|
||||
if (!job) break
|
||||
try {
|
||||
const task = await this.prepare(queue, job)
|
||||
if (!task) {
|
||||
await job.moveToCompleted(null, token, false)
|
||||
continue
|
||||
}
|
||||
// What applying needs, kept with the job: the node's copy of the task isn't trusted.
|
||||
if (task.kind === 'inspect') await job.updateData({ ...job.data, remote: { fileId: task.fileId, name: task.name } })
|
||||
jobs.push({ queue, jobId: job.id!, token, task })
|
||||
} catch (error: any) {
|
||||
await job.moveToFailed(error, token, false).catch(() => {})
|
||||
}
|
||||
}
|
||||
}
|
||||
return jobs
|
||||
}
|
||||
|
||||
/** Applies a node's result and completes the job. */
|
||||
async finish(queue: QueueName, jobId: string, token: string, result: InspectProbe | WalkProbe) {
|
||||
const job = await this.job(queue, jobId)
|
||||
let summary: unknown
|
||||
if (queue === 'collection-queue') {
|
||||
const remote = job.data.remote
|
||||
if (!remote) throw new JobLost(`Job ${jobId} was not leased to a node`)
|
||||
const task: InspectTask = { kind: 'inspect', fileId: Number(remote.fileId), name: String(remote.name), segments: [], sevenZip: null }
|
||||
const content = await ContentService.apply(task, result as InspectProbe)
|
||||
for (const line of describeContent(task.name, content)) logger.info(`[node] ${line}`)
|
||||
summary = content.contentType
|
||||
} else {
|
||||
const outcome = await ArchiveWalkService.apply(Number(job.data.releaseId), result as WalkProbe)
|
||||
logger.info(
|
||||
`[node] Archive walk of release ${outcome.releaseId}: ${outcome.before} → ${outcome.entries} entries from ${outcome.segments} segments` +
|
||||
(outcome.complete ? '' : ` (stopped: ${outcome.stopped})`) +
|
||||
(outcome.renamed ? `, renamed to "${outcome.renamed}"` : '')
|
||||
)
|
||||
summary = outcome
|
||||
}
|
||||
await this.locked(() => job.moveToCompleted(summary, token, false))
|
||||
}
|
||||
|
||||
/** The probe failed on the node (e.g. a missing article): retried like a local failure. */
|
||||
async fail(queue: QueueName, jobId: string, token: string, error: string) {
|
||||
const job = await this.job(queue, jobId)
|
||||
// A missing first article from a node that predates the probe's own handling: a result,
|
||||
// not something to retry (each try costs Eweka ~450 ms; LOG.md 2026-10-06 (14)).
|
||||
if (queue === 'collection-queue' && job.data.remote && isMissingArticle(error)) {
|
||||
return this.finish(queue, jobId, token, { size: 0, contentType: 'missing', md5_16k: null, yencName: '' })
|
||||
}
|
||||
await this.locked(() => job.moveToFailed(new Error(`node: ${error}`), token, false))
|
||||
}
|
||||
|
||||
/** A stopping node gives its jobs back, without using up an attempt. */
|
||||
async release(queue: QueueName, jobId: string, token: string) {
|
||||
const job = await this.job(queue, jobId)
|
||||
await this.locked(() => job.moveToWait(token))
|
||||
}
|
||||
|
||||
/** One article body for a relay node, on this process's NNTP pool. Null when the server doesn't have it. */
|
||||
async article(messageId: string): Promise<Buffer | null> {
|
||||
// NNTP_POOL_SIZE 0 (SEED_ARTICLE_CONNECTIONS=0): no connections for this; acquire() would wait forever.
|
||||
if (nntpConfig.poolSize <= 0) throw new ArticleRelayOff('Article relay is off (SEED_ARTICLE_CONNECTIONS=0)')
|
||||
const conn = await NntpService.acquire()
|
||||
let failed = false
|
||||
try {
|
||||
return await NntpService.timed(NntpService.fetchBody(conn, messageId), `BODY ${messageId}`)
|
||||
} catch (error: any) {
|
||||
// 430 no such article: a fact about the article, the connection is fine.
|
||||
if (/^4\d\d/.test(String(error.message))) return null
|
||||
failed = true
|
||||
throw error
|
||||
} finally {
|
||||
NntpService.release(conn, failed)
|
||||
}
|
||||
}
|
||||
|
||||
/** Closes the lease workers' Redis connections (tests; a server just exits). */
|
||||
async close() {
|
||||
await Promise.all([...this.workers.values()].map((worker) => worker.close()))
|
||||
this.workers.clear()
|
||||
}
|
||||
|
||||
isQueue(name: unknown): name is QueueName {
|
||||
return QUEUES.includes(name as QueueName)
|
||||
}
|
||||
|
||||
private async prepare(queue: QueueName, job: Job): Promise<ContentTask | null> {
|
||||
if (queue === 'collection-queue') return ContentService.prepare(Number(job.data.fileId))
|
||||
const prepared = await ArchiveWalkService.prepare(Number(job.data.releaseId))
|
||||
return 'task' in prepared ? prepared.task : null
|
||||
}
|
||||
|
||||
private async job(queue: QueueName, jobId: string): Promise<Job> {
|
||||
const job = await Job.fromId(queue === 'collection-queue' ? QueueService.collectionQueue : QueueService.archiveWalkQueue, jobId)
|
||||
if (!job) throw new JobLost(`Job ${jobId} is gone`)
|
||||
return job
|
||||
}
|
||||
|
||||
/** BullMQ refuses a job whose lock another holder has now: the node's result came too late. */
|
||||
private async locked(action: () => Promise<unknown>) {
|
||||
try {
|
||||
await action()
|
||||
} catch (error: any) {
|
||||
if (/lock|not in the active state|Missing key/i.test(String(error.message))) throw new JobLost(error.message)
|
||||
throw error
|
||||
}
|
||||
}
|
||||
|
||||
/** A worker that only hands out jobs (no processor, never runs). */
|
||||
private worker(queue: QueueName): Worker {
|
||||
let worker = this.workers.get(queue)
|
||||
if (!worker) {
|
||||
worker = new Worker(queue, null, { connection: queueConfig.connection, lockDuration: LOCK_MS, autorun: false })
|
||||
this.workers.set(queue, worker)
|
||||
}
|
||||
return worker
|
||||
}
|
||||
}
|
||||
|
||||
export default new RemoteContentService()
|
||||
@@ -1,421 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import { buildNzb } from '#lib/nzb/index'
|
||||
import SpotService from '#services/SpotService'
|
||||
import { segmentsOf } from '#services/FileStore'
|
||||
import { spotCategoryLabels } from '#lib/spotnet/index'
|
||||
|
||||
/**
|
||||
* `known`: we have the password (it's in our NZB, so SABnzbd/NZBGet unpack
|
||||
* without asking); `missing`: the archive is encrypted and we don't have
|
||||
* it; `none`: no password needed as far as we know.
|
||||
*/
|
||||
export type PasswordState = 'known' | 'missing' | 'none'
|
||||
|
||||
export function passwordState(encrypted: boolean | null, hasPassword: boolean): PasswordState {
|
||||
if (hasPassword) return 'known'
|
||||
return encrypted ? 'missing' : 'none'
|
||||
}
|
||||
|
||||
export type SortOrder = 'newest' | 'oldest' | 'largest' | 'files' | 'name'
|
||||
|
||||
export interface SearchParams {
|
||||
q?: string
|
||||
/** Only releases whose name is readable (not an obfuscated token). */
|
||||
readable?: boolean
|
||||
/** Only releases that have all their files. */
|
||||
complete?: boolean
|
||||
/** Hide releases that need a password we don't have. */
|
||||
hideEncrypted?: boolean
|
||||
source?: string
|
||||
poster?: string
|
||||
sort?: SortOrder
|
||||
page?: number
|
||||
perPage?: number
|
||||
/** Newznab: exact category ids (already expanded from parents). */
|
||||
categories?: number[]
|
||||
/** Newznab: leave out releases known to be missing files. */
|
||||
notIncomplete?: boolean
|
||||
maxAgeDays?: number
|
||||
/** TV: season and episode numbers, matched as S01E02 or 1x02 in the name. */
|
||||
season?: number
|
||||
episode?: number
|
||||
/** Overrides `page` (Newznab uses offsets). */
|
||||
offset?: number
|
||||
/** Leave out releases that are only an encrypted container (nzbe). */
|
||||
hideContainers?: boolean
|
||||
/** Include incomplete releases that consist of par2 files only (hidden by default). */
|
||||
includePar2Only?: boolean
|
||||
/** Size bounds in bytes (releases without a known size never match). */
|
||||
minSize?: number
|
||||
maxSize?: number
|
||||
/** File extensions (without the dot); a release matches if a file or archive entry has one of them (`releases.extensions`). */
|
||||
extensions?: string[]
|
||||
/** Count matches up to this many past the page (default COUNT_LIMIT); `total` is capped there. */
|
||||
countLimit?: number
|
||||
}
|
||||
|
||||
const SORTS: Record<SortOrder, string> = {
|
||||
newest: 'r.posted_at DESC NULLS LAST, r.id DESC',
|
||||
oldest: 'r.posted_at ASC NULLS LAST, r.id ASC',
|
||||
largest: 'r.size DESC, r.id DESC',
|
||||
files: 'r.file_count DESC, r.id DESC',
|
||||
name: 'r.name ASC, r.id ASC',
|
||||
}
|
||||
|
||||
/**
|
||||
* Shown in search: releases with files, and Spotnet listings whose NZB is
|
||||
* fetched on first download (CONCEPT.md D29). Matches the partial indexes.
|
||||
*/
|
||||
const VISIBLE_SQL = `(r.file_count > 0 OR r.source = 'spot')`
|
||||
|
||||
/** Incomplete, and every file is a par2 file: nothing to download but repair data. */
|
||||
const PAR2_ONLY_SQL = `(r.file_count > 0 AND NOT r.complete AND NOT EXISTS (
|
||||
SELECT 1 FROM files f LEFT JOIN file_details d ON d.file_id = f.id WHERE f.release_id = r.id
|
||||
AND d.content_type IS DISTINCT FROM 'par2' AND f.filename !~* '\\.par2$'))`
|
||||
|
||||
/** A search word as an ILIKE pattern; dots, underscores, dashes and spaces are interchangeable. */
|
||||
function pattern(word: string): string {
|
||||
const escaped = word.replace(/[\\%]/g, (c) => `\\${c}`)
|
||||
return `%${escaped.replace(/[._\- ]+/g, '_')}%`
|
||||
}
|
||||
|
||||
/**
|
||||
* As SQL's search_normalize(): lowercase, and dots, dashes, underscores and
|
||||
* spaces as one space. `The.Block.AU` and `the block au` search the same.
|
||||
*/
|
||||
export function searchNormalize(text: string): string {
|
||||
return text.toLowerCase().replace(/[._\s-]+/g, ' ').trim()
|
||||
}
|
||||
|
||||
/** How long the web UI's totals are reused. */
|
||||
const STATS_TTL_MS = 60 * 1000
|
||||
|
||||
/**
|
||||
* Counting stops here: a broad search needn't count every match. Each
|
||||
* counted row is a random heap read, so the web UI counts less (it only
|
||||
* shows "1,000+"); Newznab clients page by this total and keep the default.
|
||||
*/
|
||||
const COUNT_LIMIT = 10_000
|
||||
|
||||
/**
|
||||
* Searching and browsing releases for the web interface (CONCEPT.md D22).
|
||||
*/
|
||||
class SearchService {
|
||||
async search(params: SearchParams) {
|
||||
const where: string[] = [
|
||||
VISIBLE_SQL,
|
||||
// Retracted by its poster or a Spotnet moderator (CONCEPT.md D29).
|
||||
`NOT EXISTS (SELECT 1 FROM spots s WHERE s.release_id = r.id AND s.status IN ('removed', 'dummy'))`,
|
||||
]
|
||||
const bindings: unknown[] = []
|
||||
|
||||
// Every word must occur in the release's search text: its name, archive
|
||||
// listing and file names (CONCEPT.md D33), one trigram index for all.
|
||||
const words = (params.q ?? '').trim().split(/\s+/).map(searchNormalize).filter(Boolean).slice(0, 8)
|
||||
for (const word of words) {
|
||||
where.push(`r.search_text LIKE ?`)
|
||||
bindings.push(`%${word.replace(/[\\%_]/g, (c) => `\\${c}`)}%`)
|
||||
}
|
||||
if (params.readable) where.push('is_readable_name(r.name)')
|
||||
if (params.minSize !== undefined) {
|
||||
where.push('r.size >= ?')
|
||||
bindings.push(params.minSize)
|
||||
}
|
||||
if (params.maxSize !== undefined) {
|
||||
where.push('r.size <= ?')
|
||||
bindings.push(params.maxSize)
|
||||
}
|
||||
// File types from file names and archive listings (migration …130), via their GIN index.
|
||||
const extensions = (params.extensions ?? []).map((e) => e.toLowerCase().replace(/^\./, '')).filter((e) => /^[a-z0-9]{1,8}$/.test(e)).slice(0, 8)
|
||||
if (extensions.length > 0) {
|
||||
where.push('r.extensions && ?::text[]')
|
||||
bindings.push(extensions)
|
||||
}
|
||||
if (params.hideEncrypted) where.push('(r.encrypted IS NOT TRUE OR r.password IS NOT NULL)')
|
||||
if (params.hideContainers) where.push('r.container IS NULL')
|
||||
// par2 files whose data never arrived (e.g. uploads spread over groups we can't fetch).
|
||||
if (!params.includePar2Only) where.push(`(${PAR2_ONLY_SQL}) IS NOT TRUE`)
|
||||
// Kept current by triggers (migration …860).
|
||||
if (params.complete) where.push('r.complete')
|
||||
if (params.source) {
|
||||
where.push('r.source = ?')
|
||||
bindings.push(params.source)
|
||||
}
|
||||
if (params.poster) {
|
||||
where.push('r.poster ILIKE ?')
|
||||
bindings.push(pattern(params.poster))
|
||||
}
|
||||
if (params.categories?.length) {
|
||||
where.push('r.category = ANY(?)')
|
||||
bindings.push(params.categories)
|
||||
}
|
||||
if (params.notIncomplete) {
|
||||
where.push(`(r.complete OR (r.file_total IS NULL AND NOT EXISTS (SELECT 1 FROM release_files rf WHERE rf.release_id = r.id)))`)
|
||||
}
|
||||
if (params.maxAgeDays) {
|
||||
where.push(`r.posted_at >= now() - make_interval(days => ?)`)
|
||||
bindings.push(params.maxAgeDays)
|
||||
}
|
||||
if (params.season !== undefined) {
|
||||
const s = params.season
|
||||
const e = params.episode
|
||||
where.push('r.name ~* ?')
|
||||
bindings.push(
|
||||
e !== undefined
|
||||
? `(^|[._ -])(S0*${s}E0*${e}|0*${s}x0*${e})([^0-9]|$)`
|
||||
: `(^|[._ -])(S0*${s}([._ -]|E\\d|$)|0*${s}x\\d)`
|
||||
)
|
||||
}
|
||||
|
||||
const perPage = Math.min(Math.max(params.perPage ?? 50, 1), 200)
|
||||
const page = Math.max(params.page ?? 1, 1)
|
||||
const offset = params.offset ?? (page - 1) * perPage
|
||||
const order = SORTS[params.sort ?? 'newest'] ?? SORTS.newest
|
||||
const whereSql = where.join(' AND ')
|
||||
const countLimit = offset + (params.countLimit ?? COUNT_LIMIT)
|
||||
|
||||
const [{ rows }, { rows: countRows }] = await Promise.all([
|
||||
db.rawQuery(
|
||||
`SELECT r.id, r.name, r.source, r.poster, r.groups, r.posted_at, r.file_total, r.file_count, r.category, r.container,
|
||||
r.size, r.encrypted, is_readable_name(r.name) AS readable,
|
||||
(SELECT count(*) FROM release_files rf WHERE rf.release_id = r.id) AS listed_files,
|
||||
(SELECT count(*) FROM release_files rf WHERE rf.release_id = r.id AND EXISTS (
|
||||
SELECT 1 FROM files f JOIN binaries b ON b.file_id = f.id
|
||||
WHERE f.release_id = r.id AND (b.size = rf.size OR f.real_name = rf.name))) AS listed_present,
|
||||
jsonb_array_length(coalesce(r.contents, '[]'::jsonb)) AS archive_entries,
|
||||
r.complete, ${PAR2_ONLY_SQL} AS par2_only, r.password IS NOT NULL AS has_password,
|
||||
(SELECT s.description FROM spots s WHERE s.release_id = r.id AND s.description IS NOT NULL LIMIT 1) AS spot_description
|
||||
FROM releases r WHERE ${whereSql}
|
||||
ORDER BY ${order} LIMIT ? OFFSET ?`,
|
||||
[...bindings, perPage, offset]
|
||||
),
|
||||
db.rawQuery(`SELECT count(*) AS total FROM (SELECT 1 FROM releases r WHERE ${whereSql} LIMIT ${countLimit + 1}) x`, bindings),
|
||||
])
|
||||
|
||||
return {
|
||||
page,
|
||||
perPage,
|
||||
offset,
|
||||
total: Math.min(Number(countRows[0].total), countLimit),
|
||||
/** More than `total` match (the count stops at `countLimit`). */
|
||||
totalCapped: Number(countRows[0].total) > countLimit,
|
||||
releases: rows.map((r: any) => ({
|
||||
id: Number(r.id),
|
||||
name: r.name,
|
||||
source: r.source,
|
||||
poster: r.poster,
|
||||
groups: r.groups,
|
||||
postedAt: r.posted_at,
|
||||
fileTotal: r.file_total,
|
||||
fileCount: r.file_count,
|
||||
size: Number(r.size),
|
||||
encrypted: r.encrypted,
|
||||
readable: r.readable,
|
||||
listedFiles: Number(r.listed_files),
|
||||
listedPresent: Number(r.listed_present),
|
||||
archiveEntries: Number(r.archive_entries),
|
||||
complete: r.complete,
|
||||
category: r.category,
|
||||
container: r.container,
|
||||
par2Only: r.par2_only,
|
||||
listingOnly: r.source === 'spot' && Number(r.file_count) === 0,
|
||||
hasPassword: r.has_password,
|
||||
passwordState: passwordState(r.encrypted, r.has_password),
|
||||
spotDescription: r.spot_description,
|
||||
})),
|
||||
}
|
||||
}
|
||||
|
||||
async release(id: number) {
|
||||
const release = await db.from('releases').where('id', id).first()
|
||||
if (!release) return null
|
||||
|
||||
const [files, listed, pending] = await Promise.all([
|
||||
db
|
||||
.from('files as f')
|
||||
.leftJoin('binaries as b', 'b.file_id', 'f.id')
|
||||
.where('f.release_id', id)
|
||||
.orderByRaw('coalesce(f.real_name, f.filename)')
|
||||
.select('f.id', 'f.filename', 'f.real_name', 'f.size', 'f.parts', 'f.date', 'f.poster', 'b.size as exact_size'),
|
||||
db.rawQuery(
|
||||
`SELECT rf.name, rf.size,
|
||||
EXISTS (SELECT 1 FROM files f JOIN binaries b ON b.file_id = f.id
|
||||
WHERE f.release_id = rf.release_id AND (f.real_name = rf.name OR b.size = rf.size)) AS present
|
||||
FROM release_files rf WHERE rf.release_id = ? ORDER BY rf.name`,
|
||||
[id]
|
||||
),
|
||||
db.rawQuery(
|
||||
`SELECT count(*) AS binaries, coalesce(sum(b.total_parts), 0) AS parts_total,
|
||||
coalesce(sum((SELECT count(*) FROM parts p WHERE p.binary_id = b.id)), 0) AS parts_have
|
||||
FROM binaries b WHERE b.release_id = ? AND b.completed_at IS NULL AND b.conflicted_at IS NULL`,
|
||||
[id]
|
||||
),
|
||||
])
|
||||
|
||||
const spotRow = await db.from('spots').where('release_id', id).whereNotIn('status', ['removed', 'dummy']).orderBy('posted_at', 'desc').first()
|
||||
const spot = spotRow ? await SpotService.withDetails(spotRow) : null
|
||||
return {
|
||||
id: Number(release.id),
|
||||
name: release.name,
|
||||
source: release.source,
|
||||
spot: spot
|
||||
? {
|
||||
messageId: spot.message_id,
|
||||
title: spot.title,
|
||||
tag: spot.tag,
|
||||
poster: spot.poster,
|
||||
postedAt: spot.posted_at,
|
||||
website: spot.website,
|
||||
description: spot.description,
|
||||
image: spot.image?.segments
|
||||
? { url: `/api/v1/releases/${id}/image`, width: spot.image.width ?? null, height: spot.image.height ?? null }
|
||||
: spot.image?.url
|
||||
? { external: spot.image.url }
|
||||
: null,
|
||||
categories: spotCategoryLabels(spot.category, spot.subcats),
|
||||
}
|
||||
: null,
|
||||
password: release.password ?? null,
|
||||
passwordSource: release.password_source ?? null,
|
||||
passwordState: passwordState(release.encrypted, release.password !== null),
|
||||
listingOnly: release.source === 'spot' && Number(release.file_count) === 0,
|
||||
poster: release.poster,
|
||||
groups: release.groups,
|
||||
postedAt: release.posted_at,
|
||||
fileTotal: release.file_total,
|
||||
fileCount: release.file_count,
|
||||
size: Number(release.size),
|
||||
encrypted: release.encrypted,
|
||||
container: release.container,
|
||||
contents: release.contents ?? [],
|
||||
files: files.map((f) => ({
|
||||
id: Number(f.id),
|
||||
name: f.real_name ?? f.filename,
|
||||
postedName: f.filename,
|
||||
realName: f.real_name,
|
||||
size: Number(f.exact_size ?? f.size),
|
||||
parts: f.parts,
|
||||
postedAt: f.date ? new Date(Number(f.date)) : null,
|
||||
poster: f.poster,
|
||||
})),
|
||||
par2Files: listed.rows.map((r: any) => ({ name: r.name, size: Number(r.size), present: r.present })),
|
||||
pending: {
|
||||
binaries: Number(pending.rows[0].binaries),
|
||||
partsHave: Number(pending.rows[0].parts_have),
|
||||
partsTotal: Number(pending.rows[0].parts_total),
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
/** The image of the spot behind a release: bytes and a media type sniffed from them, or null. */
|
||||
async spotImage(id: number): Promise<{ data: Buffer; type: string } | null> {
|
||||
const spot = await db.from('spots').where('release_id', id).whereNotNull('image').orderBy('posted_at', 'desc').first()
|
||||
if (!spot) return null
|
||||
const data = await SpotService.image(spot)
|
||||
if (!data) return null
|
||||
const type =
|
||||
data[0] === 0xff && data[1] === 0xd8 ? 'image/jpeg'
|
||||
: data.subarray(0, 4).toString('latin1') === '\x89PNG' ? 'image/png'
|
||||
: data.subarray(0, 3).toString('latin1') === 'GIF' ? 'image/gif'
|
||||
: data.subarray(0, 2).toString('latin1') === 'BM' ? 'image/bmp'
|
||||
: data.subarray(8, 12).toString('latin1') === 'WEBP' ? 'image/webp'
|
||||
: null
|
||||
return type ? { data, type } : null
|
||||
}
|
||||
|
||||
async nzb(id: number): Promise<{ name: string; xml: string } | null> {
|
||||
let release = await db.from('releases').where('id', id).first()
|
||||
if (!release) return null
|
||||
if (release.source === 'spot' && Number(release.file_count) === 0) {
|
||||
// A Spotnet listing: read its NZB from Usenet now (the import may join known files elsewhere).
|
||||
const imported = await SpotService.importListing(id)
|
||||
if (!imported) return null
|
||||
id = imported
|
||||
release = await db.from('releases').where('id', id).first()
|
||||
}
|
||||
const files = await db
|
||||
.from('files')
|
||||
.where('release_id', id)
|
||||
.orderByRaw('coalesce(real_name, filename)')
|
||||
.select('filename', 'real_name', 'poster', 'date', 'groups', 'segments')
|
||||
const xml = buildNzb(
|
||||
files.map((f) => ({
|
||||
name: f.real_name ?? f.filename,
|
||||
poster: f.poster,
|
||||
date: Number(f.date),
|
||||
groups: f.groups?.length ? f.groups : ['alt.binaries.test'],
|
||||
segments: segmentsOf(f),
|
||||
})),
|
||||
{ title: release.name, password: release.password ?? undefined }
|
||||
)
|
||||
return { name: release.name, xml }
|
||||
}
|
||||
|
||||
/**
|
||||
* Totals for the web UI header. They count millions of rows (minutes with
|
||||
* the backfill running), so they are computed at most once per
|
||||
* STATS_TTL_MS and never twice at a time; callers in between share it.
|
||||
*/
|
||||
async stats() {
|
||||
if (!this.statsCache || Date.now() - this.statsCache.at > STATS_TTL_MS) {
|
||||
this.statsRefresh ??= this.computeStats()
|
||||
.then((value) => {
|
||||
this.statsCache = { at: Date.now(), value }
|
||||
})
|
||||
// A failed background refresh keeps the previous totals; only a caller that waits sees the error.
|
||||
.catch((error) => {
|
||||
if (!this.statsCache) throw error
|
||||
})
|
||||
.finally(() => {
|
||||
this.statsRefresh = undefined
|
||||
})
|
||||
// Only the very first caller waits; later ones get the previous totals meanwhile.
|
||||
if (!this.statsCache) await this.statsRefresh
|
||||
}
|
||||
return this.statsCache!.value
|
||||
}
|
||||
|
||||
private statsCache?: { at: number; value: Awaited<ReturnType<SearchService['computeStats']>> }
|
||||
private statsRefresh?: Promise<void>
|
||||
|
||||
/**
|
||||
* Estimates, in milliseconds instead of minutes (CONCEPT.md D33): counts
|
||||
* from Postgres's statistics (visible releases = the entries of an index
|
||||
* that covers exactly them), ratios and sizes from a 0.5% sample of the
|
||||
* releases table. Within a few percent once autovacuum has analysed.
|
||||
*/
|
||||
private async computeStats() {
|
||||
const { rows } = await db.rawQuery(`
|
||||
WITH sample AS (
|
||||
SELECT r.name, r.size, r.file_count, r.encrypted, coalesce(r.source, 'unknown') AS source
|
||||
FROM releases r TABLESAMPLE SYSTEM (0.5) WHERE ${VISIBLE_SQL})
|
||||
SELECT (SELECT reltuples FROM pg_class WHERE relname = 'releases_category_idx') AS releases,
|
||||
-- files is partitioned (D39): its partitions hold the counts.
|
||||
(SELECT sum(c.reltuples) FROM pg_inherits i JOIN pg_class c ON c.oid = i.inhrelid WHERE i.inhparent = 'files'::regclass) AS files,
|
||||
(SELECT count(*) FROM sample) AS sampled,
|
||||
(SELECT count(*) FROM sample WHERE is_readable_name(name)) AS sampled_readable,
|
||||
-- Indexed data only: a spot listing's size is what the spot announces.
|
||||
(SELECT coalesce(sum(size), 0) FROM sample WHERE file_count > 0) AS sampled_size,
|
||||
(SELECT count(*) FROM sample WHERE encrypted) AS sampled_encrypted,
|
||||
(SELECT max(updated_at) FROM groups) AS last_indexed_at,
|
||||
(SELECT json_agg(json_build_object('source', source, 'n', n) ORDER BY n DESC)
|
||||
FROM (SELECT source, count(*) AS n FROM sample GROUP BY 1) s) AS sources`)
|
||||
const r = rows[0]
|
||||
const releases = Math.max(0, Number(r.releases))
|
||||
const sampled = Number(r.sampled)
|
||||
// Sample counts scaled to all visible releases.
|
||||
const scale = (n: unknown) => (sampled > 0 ? Math.round((Number(n) / sampled) * releases) : 0)
|
||||
return {
|
||||
approximate: true,
|
||||
releases,
|
||||
readableReleases: scale(r.sampled_readable),
|
||||
files: Math.max(0, Number(r.files)),
|
||||
size: scale(r.sampled_size),
|
||||
encryptedReleases: scale(r.sampled_encrypted),
|
||||
lastIndexedAt: r.last_indexed_at,
|
||||
sources: (r.sources ?? []).map((s: any) => ({ source: s.source, releases: scale(s.n) })),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
export default new SearchService()
|
||||
@@ -1,132 +0,0 @@
|
||||
import nntpConfig from '#config/nntp'
|
||||
import indexerConfig from '#config/indexer'
|
||||
import { XoverClient } from '#lib/nntp/xover_client'
|
||||
|
||||
interface Slot {
|
||||
client: XoverClient
|
||||
group?: string
|
||||
usedAt: number
|
||||
}
|
||||
|
||||
/**
|
||||
* A pooled connection unused this long is replaced: the news server drops
|
||||
* idle connections without telling, and a request on one then hangs until
|
||||
* the inactivity timeout (LOG.md 2026-10-05 (11)).
|
||||
*/
|
||||
const IDLE_MS = 60 * 1000
|
||||
|
||||
/**
|
||||
* NNTP connections in the seed API for relay nodes (CONCEPT.md D38): the
|
||||
* indexer fetches overview chunks and passes the gzip stream on unopened, so
|
||||
* nodes need no news server account and every connection comes from this
|
||||
* machine's IP. SEED_NNTP_CONNECTIONS (0: relay off).
|
||||
*/
|
||||
class SeedFetchPool {
|
||||
private idle: Slot[] = []
|
||||
private open = 0
|
||||
private waiters: (() => void)[] = []
|
||||
|
||||
get size() {
|
||||
return indexerConfig.seed.nntpConnections
|
||||
}
|
||||
|
||||
/**
|
||||
* Pooled connections fetch compressed unless SEED_NNTP_COMPRESS=false. A failure is retried once without compression, on a connection of its
|
||||
* own: either the pooled connection was dead (dropped while idle), or the
|
||||
* server stalled mid-reply in compressed mode. Eweka does that for some
|
||||
* chunks: it stops after most of the stream and never sends the rest
|
||||
* (a.b.ath 49701720000: 19,672 of 20,000 overviews, then silence), while
|
||||
* the same chunk uncompressed comes in 0.3 s (LOG.md 2026-10-05 (11)).
|
||||
*/
|
||||
async fetch(group: string, from: number, to: number): Promise<Buffer> {
|
||||
try {
|
||||
return await this.fetchOnce(group, from, to)
|
||||
} catch {
|
||||
return this.fetchPlain(group, from, to)
|
||||
}
|
||||
}
|
||||
|
||||
private async fetchPlain(group: string, from: number, to: number): Promise<Buffer> {
|
||||
const client = await this.connect(false)
|
||||
try {
|
||||
await client.group(group)
|
||||
return await client.xoverRaw(from, to)
|
||||
} finally {
|
||||
client.close()
|
||||
}
|
||||
}
|
||||
|
||||
private connect(compress: boolean) {
|
||||
return XoverClient.connect({
|
||||
host: nntpConfig.host,
|
||||
port: nntpConfig.port,
|
||||
secure: nntpConfig.secure,
|
||||
servername: nntpConfig.tlsServername,
|
||||
user: nntpConfig.user,
|
||||
password: nntpConfig.password?.release(),
|
||||
timeoutMs: indexerConfig.seed.nntpTimeoutMs,
|
||||
compress,
|
||||
})
|
||||
}
|
||||
|
||||
private async fetchOnce(group: string, from: number, to: number): Promise<Buffer> {
|
||||
const slot = await this.acquire()
|
||||
let ok = false
|
||||
try {
|
||||
if (slot.group !== group) {
|
||||
await slot.client.group(group)
|
||||
slot.group = group
|
||||
}
|
||||
const raw = await slot.client.xoverRaw(from, to)
|
||||
ok = true
|
||||
return raw
|
||||
} finally {
|
||||
this.release(slot, ok)
|
||||
}
|
||||
}
|
||||
|
||||
/** Closes the idle connections (tests, shutdown); busy ones close when released. */
|
||||
closeIdle() {
|
||||
for (const slot of this.idle.splice(0)) {
|
||||
slot.client.close()
|
||||
this.open--
|
||||
}
|
||||
}
|
||||
|
||||
private async acquire(): Promise<Slot> {
|
||||
while (true) {
|
||||
const slot = this.idle.pop()
|
||||
if (slot) {
|
||||
if (Date.now() - slot.usedAt < IDLE_MS) return slot
|
||||
slot.client.close()
|
||||
this.open--
|
||||
continue
|
||||
}
|
||||
if (this.open < this.size) {
|
||||
this.open++
|
||||
try {
|
||||
return { client: await this.connect(indexerConfig.seed.nntpCompress), usedAt: Date.now() }
|
||||
} catch (error) {
|
||||
this.open--
|
||||
this.waiters.shift()?.()
|
||||
throw error
|
||||
}
|
||||
}
|
||||
await new Promise<void>((resolve) => this.waiters.push(resolve))
|
||||
}
|
||||
}
|
||||
|
||||
private release(slot: Slot, ok: boolean) {
|
||||
if (ok) {
|
||||
slot.usedAt = Date.now()
|
||||
this.idle.push(slot)
|
||||
} else {
|
||||
// A failed command can leave the connection mid-reply: start afresh.
|
||||
slot.client.close()
|
||||
this.open--
|
||||
}
|
||||
this.waiters.shift()?.()
|
||||
}
|
||||
}
|
||||
|
||||
export default new SeedFetchPool()
|
||||
@@ -1,234 +0,0 @@
|
||||
import { createHash, randomBytes } from 'node:crypto'
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import { cut, uncovered } from '#lib/seed/plan'
|
||||
import NntpService from '#services/NntpService'
|
||||
import SeedService from '#services/SeedService'
|
||||
import BackfillService from '#services/BackfillService'
|
||||
import SeedFetchPool from '#services/SeedFetchPool'
|
||||
import { articleAtDate } from '#lib/nntp/date_search'
|
||||
import { normalizeSeedFile, type SeedBatch } from '#lib/seed/wire'
|
||||
|
||||
/** A lease without a batch for this long is given to another node; the range resumes at next_id. */
|
||||
export const LEASE_TTL_SECONDS = 15 * 60
|
||||
/** Leases per range before it counts as failed (a range that keeps breaking nodes). */
|
||||
const MAX_ATTEMPTS = 5
|
||||
/** Articles per XOVER probe when looking for an article by date. */
|
||||
const PROBE = 50
|
||||
|
||||
export interface SeedLease {
|
||||
id: number
|
||||
group: string
|
||||
/** Resume point: first article not reported yet. */
|
||||
start: number
|
||||
end: number
|
||||
}
|
||||
|
||||
export class LeaseLost extends Error {}
|
||||
|
||||
const hash = (token: string) => createHash('sha256').update(token).digest('hex')
|
||||
|
||||
/**
|
||||
* The seed orchestrator (CONCEPT.md D38): seeding nodes lease article
|
||||
* ranges over HTTPS, report what they aggregated in batches (D37), and
|
||||
* finish or give back the range. Nodes only make outgoing requests; a lease
|
||||
* that goes quiet for LEASE_TTL_SECONDS passes to the next node that asks.
|
||||
*/
|
||||
class SeedOrchestrator {
|
||||
/** Creates a node, or gives an existing one a new token. Returns the token (stored hashed only). */
|
||||
async issueToken(name: string): Promise<string> {
|
||||
const token = `sn_${randomBytes(24).toString('base64url')}`
|
||||
await db
|
||||
.table('seed_nodes')
|
||||
.insert({ name, token_hash: hash(token), disabled: false })
|
||||
.onConflict('name')
|
||||
.merge({ token_hash: hash(token), disabled: false })
|
||||
return token
|
||||
}
|
||||
|
||||
async nodeForToken(token: string): Promise<{ id: number; name: string } | null> {
|
||||
const node = await db.from('seed_nodes').where('token_hash', hash(token)).where('disabled', false).first()
|
||||
if (!node) return null
|
||||
if (!node.last_seen_at || Date.now() - new Date(node.last_seen_at).getTime() > 60_000) {
|
||||
await db.from('seed_nodes').where('id', node.id).update({ last_seen_at: new Date() })
|
||||
}
|
||||
return { id: Number(node.id), name: node.name }
|
||||
}
|
||||
|
||||
/**
|
||||
* The next range for this node, oldest plan first; a quiet lease is taken
|
||||
* over. Nothing while the database is over the backfill's disk limits: the
|
||||
* node asks again a minute later.
|
||||
*/
|
||||
async lease(nodeId: number): Promise<SeedLease | null> {
|
||||
if ((await BackfillService.room()) !== true) return null
|
||||
// Fully reported but never finished (the node went away at the very end): nothing left to fetch.
|
||||
await db.rawQuery(
|
||||
`UPDATE seed_ranges SET status = 'done', done_at = now(), node_id = NULL
|
||||
WHERE next_id > end_id AND (status = 'queued' OR (status = 'leased' AND heartbeat_at < now() - make_interval(secs => ?)))`,
|
||||
[LEASE_TTL_SECONDS]
|
||||
)
|
||||
// Ranges that broke node after node stop being handed out.
|
||||
await db.rawQuery(
|
||||
`UPDATE seed_ranges SET status = 'failed', error = coalesce(error, 'lease expired too often')
|
||||
WHERE status = 'leased' AND heartbeat_at < now() - make_interval(secs => ?) AND attempts >= ?`,
|
||||
[LEASE_TTL_SECONDS, MAX_ATTEMPTS]
|
||||
)
|
||||
// Priorities take turns: while ranges of a higher priority are leased for less than half of
|
||||
// the work in progress they go first, otherwise the rest does (a 2026 gap fetches ~5x slower
|
||||
// than older history; LOG.md 2026-10-06 (10)).
|
||||
const { rows: mix } = await db.rawQuery(
|
||||
`SELECT count(*) FILTER (WHERE priority > 0) AS high, count(*) AS all FROM seed_ranges
|
||||
WHERE status = 'leased' AND heartbeat_at >= now() - make_interval(secs => ?)`,
|
||||
[LEASE_TTL_SECONDS]
|
||||
)
|
||||
const order = Number(mix[0].high) * 2 < Number(mix[0].all) + 1 ? 'priority DESC, id' : 'priority ASC, id'
|
||||
const { rows } = await db.rawQuery(
|
||||
`UPDATE seed_ranges SET status = 'leased', node_id = ?, leased_at = now(), heartbeat_at = now(), attempts = attempts + 1
|
||||
WHERE id = (
|
||||
SELECT id FROM seed_ranges
|
||||
WHERE status = 'queued' OR (status = 'leased' AND heartbeat_at < now() - make_interval(secs => ?))
|
||||
ORDER BY ${order} LIMIT 1 FOR UPDATE SKIP LOCKED)
|
||||
RETURNING id, group_name, next_id, end_id`,
|
||||
[nodeId, LEASE_TTL_SECONDS]
|
||||
)
|
||||
const row = rows[0]
|
||||
return row ? { id: Number(row.id), group: row.group_name, start: Number(row.next_id), end: Number(row.end_id) } : null
|
||||
}
|
||||
|
||||
/** Stores a node's batch and moves the range's resume point; LeaseLost when the range is no longer this node's. */
|
||||
async batch(nodeId: number, rangeId: number, batch: SeedBatch) {
|
||||
const range = await this.leased(nodeId, rangeId)
|
||||
const stored = await SeedService.store(batch.files.map(normalizeSeedFile))
|
||||
await SeedService.storeFragments(batch.fragments, { start: Number(range.start_id), end: Number(range.end_id) })
|
||||
await db.rawQuery(
|
||||
`UPDATE seed_ranges SET next_id = greatest(next_id, least(?, end_id + 1)), heartbeat_at = now(),
|
||||
headers = headers + ?, files = files + ?, fragments = fragments + ?
|
||||
WHERE id = ?`,
|
||||
[batch.next, batch.headers, batch.files.length, batch.fragments.length, rangeId]
|
||||
)
|
||||
return stored
|
||||
}
|
||||
|
||||
/**
|
||||
* Relay: overviews of `from`..`to` of a leased range, fetched here and handed
|
||||
* on as the gzip stream the news server sent. Keeps the lease alive.
|
||||
*/
|
||||
async chunk(nodeId: number, rangeId: number, from: number, to: number): Promise<Buffer> {
|
||||
const range = await this.leased(nodeId, rangeId)
|
||||
const start = Math.max(from, Number(range.start_id))
|
||||
const end = Math.min(to, Number(range.end_id), start + 49_999)
|
||||
if (end < start) throw new RangeError(`Articles ${from}–${to} are outside range ${rangeId}`)
|
||||
await db.from('seed_ranges').where('id', rangeId).update({ heartbeat_at: new Date() })
|
||||
return SeedFetchPool.fetch(range.group_name, start, end)
|
||||
}
|
||||
|
||||
async finish(nodeId: number, rangeId: number) {
|
||||
await this.leased(nodeId, rangeId)
|
||||
await db.from('seed_ranges').where('id', rangeId).update({ status: 'done', done_at: new Date(), heartbeat_at: new Date(), error: null })
|
||||
}
|
||||
|
||||
/**
|
||||
* A node gives a range back (it failed, or is shutting down); it's queued
|
||||
* again unless it failed too often. `uncounted`: the node never started on
|
||||
* it (couldn't reach its news server), so the lease doesn't count as an attempt.
|
||||
*/
|
||||
async release(nodeId: number, rangeId: number, error?: string, uncounted = false) {
|
||||
const range = await this.leased(nodeId, rangeId)
|
||||
const attempts = uncounted ? Math.max(0, range.attempts - 1) : range.attempts
|
||||
await db
|
||||
.from('seed_ranges')
|
||||
.where('id', rangeId)
|
||||
.update({ status: attempts >= MAX_ATTEMPTS ? 'failed' : 'queued', node_id: null, attempts, error: error?.slice(0, 2000) ?? null })
|
||||
}
|
||||
|
||||
private async leased(nodeId: number, rangeId: number) {
|
||||
const range = await db.from('seed_ranges').where('id', rangeId).first()
|
||||
if (!range || range.status !== 'leased' || Number(range.node_id) !== nodeId) throw new LeaseLost(`Range ${rangeId} is not leased to this node`)
|
||||
return range
|
||||
}
|
||||
|
||||
/**
|
||||
* Plans ranges for groups: from below what live indexing and the backfill
|
||||
* cover (or below earlier seed ranges) down to `days` back, or to the
|
||||
* server's oldest article. A group's backfill stops where its seed plan
|
||||
* begins. Ranges are interleaved across groups, newest first.
|
||||
*/
|
||||
async plan(
|
||||
groups: string[],
|
||||
options: { days?: number; size: number; priority?: number }
|
||||
): Promise<{ group: string; from: number; to: number; ranges: number; gaps: number }[]> {
|
||||
const perGroup: { group: string; ranges: [number, number][] }[] = []
|
||||
const summary: { group: string; from: number; to: number; ranges: number; gaps: number }[] = []
|
||||
const conn = await NntpService.acquire()
|
||||
let failed = false
|
||||
try {
|
||||
for (const name of groups) {
|
||||
const group = await db.from('groups').where('name', name).first()
|
||||
if (!group) throw new Error(`Unknown group ${name}`)
|
||||
const { first, last } = await NntpService.timed(NntpService.selectGroup(conn, name), `GROUP ${name}`)
|
||||
const dateOf = async (n: bigint) => {
|
||||
const { overviews } = await NntpService.timed<any>(conn.xover(n, n + BigInt(PROBE)), `XOVER ${name} ${n}`)
|
||||
const header = overviews.find(([number, h]: [number | null, any]) => number !== null && h?.date)?.[1]
|
||||
const date = header ? Date.parse(header.date) : Number.NaN
|
||||
return Number.isNaN(date) ? null : date
|
||||
}
|
||||
// Top: where live indexing began, or as far down as the backfill has come.
|
||||
let top = Number(await articleAtDate(first, last, new Date(group.created_at).getTime(), dateOf))
|
||||
if (group.backfill_next_id !== null) top = Math.min(top, Number(group.backfill_next_id))
|
||||
const backfilling = group.backfill_next_id !== null && group.backfill_until_id !== null && Number(group.backfill_next_id) > Number(group.backfill_until_id)
|
||||
const bottom = options.days
|
||||
? Number(await articleAtDate(first, last, Date.now() - options.days * 24 * 60 * 60 * 1000, dateOf))
|
||||
: Number(first)
|
||||
|
||||
// Everything below the top that no earlier range covers (in any state). Planning below
|
||||
// the lowest earlier range left a 13-month gap in a.b.ath (LOG.md 2026-10-06 (7)).
|
||||
const planned = await db.from('seed_ranges').where('group_name', name).select('start_id', 'end_id')
|
||||
const gaps = uncovered(bottom, top - 1, planned.map((r) => [Number(r.start_id), Number(r.end_id)] as [number, number]))
|
||||
const ranges = cut(gaps, options.size)
|
||||
perGroup.push({ group: name, ranges })
|
||||
summary.push({ group: name, from: bottom, to: top - 1, ranges: ranges.length, gaps: gaps.length })
|
||||
// The seed plan takes over the rest of this group's backfill.
|
||||
if (backfilling && ranges.length > 0) {
|
||||
await db.from('groups').where('id', group.id).update({ backfill_until_id: group.backfill_next_id })
|
||||
}
|
||||
}
|
||||
} catch (error) {
|
||||
failed = true
|
||||
throw error
|
||||
} finally {
|
||||
NntpService.release(conn, failed)
|
||||
}
|
||||
|
||||
// Round-robin over groups, each newest first, so every group fills in from the present backwards.
|
||||
const rows: Record<string, unknown>[] = []
|
||||
for (let i = 0; perGroup.some((g) => i < g.ranges.length); i++) {
|
||||
for (const g of perGroup) {
|
||||
const range = g.ranges[i]
|
||||
if (range) rows.push({ group_name: g.group, start_id: range[0], end_id: range[1], next_id: range[0], priority: options.priority ?? 0 })
|
||||
}
|
||||
}
|
||||
for (let i = 0; i < rows.length; i += 5000) await db.table('seed_ranges').multiInsert(rows.slice(i, i + 5000))
|
||||
return summary
|
||||
}
|
||||
|
||||
async status() {
|
||||
const { rows: ranges } = await db.rawQuery(
|
||||
`SELECT status, count(*)::int AS ranges, sum(end_id - start_id + 1)::bigint AS articles,
|
||||
sum(next_id - start_id)::bigint AS reached, sum(headers)::bigint AS headers, sum(files)::bigint AS files
|
||||
FROM seed_ranges GROUP BY status ORDER BY status`
|
||||
)
|
||||
const { rows: nodes } = await db.rawQuery(
|
||||
`SELECT n.name, n.last_seen_at, n.disabled,
|
||||
(SELECT count(*)::int FROM seed_ranges r WHERE r.node_id = n.id AND r.status = 'leased') AS leases,
|
||||
(SELECT coalesce(sum(headers), 0)::bigint FROM seed_ranges r WHERE r.node_id = n.id) AS headers
|
||||
FROM seed_nodes n ORDER BY n.name`
|
||||
)
|
||||
const { rows: recent } = await db.rawQuery(
|
||||
`SELECT coalesce(sum(headers), 0)::bigint AS headers FROM seed_ranges WHERE done_at > now() - interval '1 hour'`
|
||||
)
|
||||
const room = await BackfillService.room()
|
||||
return { ranges, nodes, headersDoneLastHour: Number(recent[0].headers), paused: room === true ? null : room }
|
||||
}
|
||||
}
|
||||
|
||||
export default new SeedOrchestrator()
|
||||
@@ -1,152 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import IngestService from '#services/IngestService'
|
||||
import FileStore, { insertLimits } from '#services/FileStore'
|
||||
import SfvService from '#services/SfvService'
|
||||
import ContentService from '#services/ContentService'
|
||||
import { fromFragment, type SeedFile, type SeedFragment } from '#lib/seed/wire'
|
||||
import InspectionQueue from '#services/InspectionQueue'
|
||||
import GarbageFilter from '#services/GarbageFilter'
|
||||
|
||||
/** Articles from a range's edge within which a file may continue in the neighbouring range. */
|
||||
const EDGE_ARTICLES = 50_000
|
||||
|
||||
export interface SeedStoreResult {
|
||||
files: number
|
||||
/** Files that existed already (same file_key, e.g. from live indexing or a crosspost). */
|
||||
existing: number
|
||||
releases: number
|
||||
inspections: number
|
||||
/** Random-named files of garbage signatures or dropped releases, not stored (D42). */
|
||||
dropped: number
|
||||
}
|
||||
|
||||
/**
|
||||
* Stores what seeding nodes found (CONCEPT.md D37): per complete file its
|
||||
* release, the file with its segment list, and one completed binary, as
|
||||
* worker:file would have left it, so par2 naming, inspection and GC treat
|
||||
* it like any other file; fragments take the regular ingest path. A batch
|
||||
* is a few statements, not one per article.
|
||||
*/
|
||||
class SeedService {
|
||||
async store(batch: SeedFile[]): Promise<SeedStoreResult> {
|
||||
const result: SeedStoreResult = { files: 0, existing: 0, releases: 0, inspections: 0, dropped: 0 }
|
||||
// A repost can complete the same file twice in one batch; ON CONFLICT DO UPDATE takes each row once.
|
||||
const unique = [...new Map(batch.map((file) => [file.fileKey, file])).values()]
|
||||
// Garbage isn't stored at all: no release, file or binary (CONCEPT.md D42).
|
||||
const { kept: files, dropped } = await GarbageFilter.keep(unique, (file) => ({
|
||||
name: file.name,
|
||||
messageId: file.segments[0]?.[1],
|
||||
groups: file.groups,
|
||||
releaseKey: file.release?.key ?? file.fileKey,
|
||||
}))
|
||||
result.dropped = dropped
|
||||
if (files.length === 0) return result
|
||||
const knex = db.connection().getWriteClient()
|
||||
const now = new Date()
|
||||
const postedOf = (file: SeedFile) => (file.postedAt ? new Date(file.postedAt) : undefined)
|
||||
|
||||
const releaseIds = await IngestService.upsertReleases(
|
||||
knex,
|
||||
files.map((file) => ({ release: file.release, poster: file.poster, groups: file.groups, postedAt: postedOf(file) })),
|
||||
now
|
||||
)
|
||||
result.releases = releaseIds.size
|
||||
|
||||
const rows = files.map((file) => ({
|
||||
file_key: file.fileKey,
|
||||
filename: file.name,
|
||||
poster: file.poster,
|
||||
date: (postedOf(file) ?? now).getTime(),
|
||||
parts: file.segments.length,
|
||||
segments: file.segments.map(([number, id, size]) => ({ number, id, size })),
|
||||
groups: JSON.stringify(file.groups),
|
||||
release_id: file.release ? (releaseIds.get(file.release.key) ?? null) : null,
|
||||
size: file.segments.reduce((sum, [, , size]) => sum + size, 0),
|
||||
created_at: now,
|
||||
updated_at: now,
|
||||
}))
|
||||
const inserted = await FileStore.insert(rows)
|
||||
result.files = inserted.length
|
||||
result.existing = rows.length - inserted.length
|
||||
|
||||
// File ids for the binaries below (files queued for inspection only): new ones from the insert,
|
||||
// existing ones looked up. Looking up every key of the batch probed all files partitions again,
|
||||
// 15% of Postgres' active time (LOG.md 2026-10-06 (12)).
|
||||
const fileIds = new Map<string, number>()
|
||||
for (const row of inserted) if (row.file_key) fileIds.set(row.file_key, row.id)
|
||||
const missing = files.filter((file) => !fileIds.has(file.fileKey) && ContentService.isCandidate(file.name)).map((file) => file.fileKey)
|
||||
for (let i = 0; i < missing.length; i += 5000) {
|
||||
for (const row of await db.from('files').whereIn('file_key', missing.slice(i, i + 5000)).select('id', 'file_key')) {
|
||||
fileIds.set(row.file_key, Number(row.id))
|
||||
}
|
||||
}
|
||||
// A completed binary, as worker:file would have left behind, only for files queued for content
|
||||
// inspection: par2 naming (D16) needs the fingerprint inspection stores on it. For the others
|
||||
// (~85% in a.b.ath) it was only GC churn: 1.6M rows an hour inserted and deleted (LOG.md 2026-10-06 (11)).
|
||||
const insertedKeys = new Set(inserted.map((row) => row.file_key))
|
||||
const binaries = files.flatMap((file, i) => {
|
||||
if (!ContentService.isCandidate(file.name)) return []
|
||||
return [{
|
||||
file_key: file.fileKey,
|
||||
poster: file.poster,
|
||||
subject: file.subject,
|
||||
name: file.name,
|
||||
total_parts: file.totalParts,
|
||||
groups: rows[i].groups,
|
||||
posted_at: postedOf(file) ?? null,
|
||||
first_seen_at: now,
|
||||
last_seen_at: now,
|
||||
completed_at: now,
|
||||
release_id: rows[i].release_id,
|
||||
file_id: fileIds.get(file.fileKey) ?? null,
|
||||
// New files are queued for inspection below; an existing file was queued when it was added.
|
||||
awaiting_inspection: insertedKeys.has(file.fileKey) ? true : null,
|
||||
}]
|
||||
})
|
||||
binaries.sort((a, b) => a.file_key.localeCompare(b.file_key))
|
||||
for (let i = 0; i < binaries.length; i += insertLimits.rows) {
|
||||
await knex('binaries')
|
||||
.insert(binaries.slice(i, i + insertLimits.rows))
|
||||
.onConflict('file_key')
|
||||
.merge({
|
||||
file_id: knex.raw('coalesce(binaries.file_id, excluded.file_id)') as any,
|
||||
awaiting_inspection: knex.raw('coalesce(binaries.awaiting_inspection, excluded.awaiting_inspection)') as any,
|
||||
})
|
||||
}
|
||||
|
||||
// As worker:file does for a new file: an SFV posted earlier may list it (D34), then inspection.
|
||||
const rowByKey = new Map(rows.map((row) => [row.file_key, row]))
|
||||
const added = inserted.flatMap((file) => {
|
||||
const row = file.file_key ? rowByKey.get(file.file_key) : undefined
|
||||
return row ? [{ id: file.id, name: row.filename, poster: row.poster, date: row.date, release_id: row.release_id }] : []
|
||||
})
|
||||
await SfvService.claimAll(added)
|
||||
const candidates = added
|
||||
.filter((file) => ContentService.isCandidate(file.name))
|
||||
.map((file) => file.id)
|
||||
if (candidates.length > 0) await InspectionQueue.add(candidates)
|
||||
result.inspections = candidates.length
|
||||
return result
|
||||
}
|
||||
|
||||
/**
|
||||
* Segments of files that didn't complete on a node. Only files near the
|
||||
* range's edges can continue in the neighbouring range (worked on by some
|
||||
* node, maybe hours later): those go straight into binaries and parts
|
||||
* (staging would drop them after 30 minutes). The rest are files with
|
||||
* articles missing for good, and are dropped: staging them filled 22 GB
|
||||
* of churn on the NVMe (LOG.md 2026-10-05 (10)). Returns the segments kept.
|
||||
*/
|
||||
async storeFragments(fragments: SeedFragment[], range: { start: number; end: number }): Promise<number> {
|
||||
const nearEdge = new Set<string>()
|
||||
for (const fragment of fragments) {
|
||||
const n = Number(fragment.articleNumber)
|
||||
if (Number.isFinite(n) && (n < range.start + EDGE_ARTICLES || n > range.end - EDGE_ARTICLES)) nearEdge.add(fragment.fileKey)
|
||||
}
|
||||
const kept = fragments.filter((fragment) => nearEdge.has(fragment.fileKey))
|
||||
if (kept.length > 0) await IngestService.ingest(kept.map(fromFragment))
|
||||
return kept.length
|
||||
}
|
||||
}
|
||||
|
||||
export default new SeedService()
|
||||
@@ -1,36 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import { sevenZipVolume } from '#lib/sevenzip/index'
|
||||
|
||||
/**
|
||||
* The database side of 7z listings (CONCEPT.md D19): which volumes make up
|
||||
* an archive. Listing itself (first segment, tail, bsdtar) is
|
||||
* lib/sevenzip/list.ts, run by inspection here or on a node.
|
||||
*/
|
||||
class SevenZipService {
|
||||
/**
|
||||
* The volumes of the archive `file` belongs to, in order, when all of them
|
||||
* up to the highest number seen are there; null otherwise. An unsplit (or
|
||||
* unnamed) archive is its own only volume.
|
||||
*/
|
||||
async volumes(file: any): Promise<any[] | null> {
|
||||
const name = String(file.real_name ?? file.filename)
|
||||
const volume = sevenZipVolume(name)
|
||||
if (!volume) return [file]
|
||||
if (!file.release_id) return null
|
||||
const base = name.replace(/\.\d+$/, '').toLowerCase()
|
||||
const rows = await db
|
||||
.from('files')
|
||||
.where('release_id', file.release_id)
|
||||
.whereRaw(`lower(coalesce(real_name, filename)) LIKE ? ESCAPE '\\'`, [`${base.replace(/[\\%_]/g, '\\$&')}.%`])
|
||||
const byNumber = new Map<number, any>()
|
||||
for (const row of rows) {
|
||||
const n = sevenZipVolume(String(row.real_name ?? row.filename))
|
||||
if (n && !byNumber.has(n)) byNumber.set(n, row)
|
||||
}
|
||||
const max = Math.max(0, ...byNumber.keys())
|
||||
const ordered = Array.from({ length: max }, (_, i) => byNumber.get(i + 1))
|
||||
return max > 0 && ordered.every(Boolean) ? ordered : null
|
||||
}
|
||||
}
|
||||
|
||||
export default new SevenZipService()
|
||||
@@ -1,96 +0,0 @@
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
import FileStore from '#services/FileStore'
|
||||
import { parseSfv, sfvReleaseName } from '#lib/content/index'
|
||||
|
||||
/** Releases made by name rules alone; an SFV's listing outranks them. */
|
||||
const WEAK_SOURCES = ['stem', 'weak-stem', 'subject']
|
||||
/** Files further apart than this from the SFV in posting time aren't taken. */
|
||||
const WINDOW_MS = 2 * 24 * 60 * 60 * 1000
|
||||
|
||||
/**
|
||||
* SFV grouping (CONCEPT.md D34). An uploader that posts every track of an
|
||||
* album as its own file (13.5k such releases from one poster on homelab)
|
||||
* still posts the album's SFV, which names them all.
|
||||
*/
|
||||
class SfvService {
|
||||
/**
|
||||
* Records an SFV file's listing on its release, moves the listed files that
|
||||
* sit in weak releases into it, and names the release after the album when
|
||||
* it's still named after a track or the SFV itself. Returns how many files
|
||||
* moved, or null when it isn't an SFV.
|
||||
*/
|
||||
async attach(sfvFile: any, text: string): Promise<number | null> {
|
||||
const entries = parseSfv(text)
|
||||
if (entries.length === 0 || !sfvFile.release_id) return null
|
||||
const releaseId = Number(sfvFile.release_id)
|
||||
const postedAt = sfvFile.date ? new Date(Number(sfvFile.date)) : null
|
||||
await db
|
||||
.knexQuery()
|
||||
.table('sfv_entries')
|
||||
.insert(entries.map((e) => ({ release_id: releaseId, name: e.name, crc: e.crc, poster: sfvFile.poster, posted_at: postedAt })))
|
||||
.onConflict(['release_id', 'name'])
|
||||
.ignore()
|
||||
await FileStore.setDetails(Number(sfvFile.id), { content_type: 'sfv' })
|
||||
|
||||
const candidates = await db
|
||||
.from('files as f')
|
||||
.join('releases as r', 'r.id', 'f.release_id')
|
||||
.whereIn(db.raw('coalesce(f.real_name, f.filename)') as any, entries.map((e) => e.name))
|
||||
.whereNot('f.release_id', releaseId)
|
||||
.whereIn('r.source', WEAK_SOURCES)
|
||||
.select('f.id', 'f.poster', 'f.date')
|
||||
const moving = candidates.filter((f) => this.sameUpload(f, sfvFile.poster, postedAt))
|
||||
await this.move(moving.map((f) => Number(f.id)), releaseId)
|
||||
|
||||
const release = await db.from('releases').where('id', releaseId).first()
|
||||
const album = sfvReleaseName(String(sfvFile.real_name ?? sfvFile.filename))
|
||||
if (release && WEAK_SOURCES.includes(release.source) && album !== release.name && /^\d{1,3}-/.test(release.name)) {
|
||||
await db.from('releases').where('id', releaseId).update({ name: album })
|
||||
}
|
||||
logger.info(`SFV "${sfvFile.filename}": ${entries.length} files listed, ${moving.length} moved into release ${releaseId} "${album}".`)
|
||||
return moving.length
|
||||
}
|
||||
|
||||
/** A new file an SFV already listed joins that SFV's release. Returns the release, or null. */
|
||||
/**
|
||||
* claim() for a batch of new files: one lookup for all their names, the
|
||||
* per-file work only for the few an SFV lists (seeding stores thousands
|
||||
* of files per batch; CONCEPT.md D37).
|
||||
*/
|
||||
async claimAll(files: { id: number; name: string; poster?: string | null; date?: number | null; release_id?: number | null }[]): Promise<number> {
|
||||
if (files.length === 0) return 0
|
||||
const listed = new Set(
|
||||
(await db.from('sfv_entries').whereIn('name', [...new Set(files.map((f) => f.name))]).select('name')).map((e) => e.name as string)
|
||||
)
|
||||
let moved = 0
|
||||
for (const file of files) if (listed.has(file.name) && (await this.claim(file)) !== null) moved++
|
||||
return moved
|
||||
}
|
||||
|
||||
async claim(file: { id: number; name: string; poster?: string | null; date?: number | null; release_id?: number | null }): Promise<number | null> {
|
||||
const entries = await db.from('sfv_entries').where('name', file.name)
|
||||
const entry = entries.find((e) => this.sameUpload(file, e.poster, e.posted_at ? new Date(e.posted_at) : null))
|
||||
if (!entry || Number(entry.release_id) === Number(file.release_id)) return null
|
||||
if (file.release_id) {
|
||||
const current = await db.from('releases').where('id', file.release_id).first()
|
||||
if (current && !WEAK_SOURCES.includes(current.source)) return null
|
||||
}
|
||||
await this.move([file.id], Number(entry.release_id))
|
||||
return Number(entry.release_id)
|
||||
}
|
||||
|
||||
private sameUpload(file: { poster?: string | null; date?: number | string | null }, poster: string | null, postedAt: Date | null) {
|
||||
if (poster && file.poster && file.poster !== poster) return false
|
||||
if (postedAt && file.date && Math.abs(Number(file.date) - postedAt.getTime()) > WINDOW_MS) return false
|
||||
return true
|
||||
}
|
||||
|
||||
private async move(fileIds: number[], releaseId: number) {
|
||||
if (fileIds.length === 0) return
|
||||
await db.from('files').whereIn('id', fileIds).update({ release_id: releaseId })
|
||||
await db.from('binaries').whereIn('file_id', fileIds).update({ release_id: releaseId })
|
||||
}
|
||||
}
|
||||
|
||||
export default new SfvService()
|
||||
@@ -1,45 +0,0 @@
|
||||
import { setTimeout as sleep } from 'node:timers/promises'
|
||||
|
||||
/** After a signal, give cleanup this long before exiting anyway. */
|
||||
const FORCE_EXIT_MS = 25_000
|
||||
|
||||
/**
|
||||
* Graceful shutdown for the long-running commands. SIGTERM (docker stop) and
|
||||
* SIGINT wake up `wait()` and `sleep()`, so workers can finish their active
|
||||
* jobs and close their connections before the process exits.
|
||||
*/
|
||||
class Shutdown {
|
||||
private readonly controller = new AbortController()
|
||||
|
||||
constructor() {
|
||||
for (const signal of ['SIGTERM', 'SIGINT'] as const) {
|
||||
process.once(signal, () => {
|
||||
this.controller.abort()
|
||||
setTimeout(() => process.exit(0), FORCE_EXIT_MS).unref()
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
get requested(): boolean {
|
||||
return this.controller.signal.aborted
|
||||
}
|
||||
|
||||
/** Resolves when a shutdown signal arrives. */
|
||||
wait(): Promise<void> {
|
||||
if (this.requested) return Promise.resolve()
|
||||
return new Promise((resolve) =>
|
||||
this.controller.signal.addEventListener('abort', () => resolve(), { once: true })
|
||||
)
|
||||
}
|
||||
|
||||
/** Sleeps, but returns early on shutdown. */
|
||||
async sleep(ms: number): Promise<void> {
|
||||
try {
|
||||
await sleep(ms, undefined, { signal: this.controller.signal })
|
||||
} catch {
|
||||
// aborted by a shutdown signal
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
export default new Shutdown()
|
||||
@@ -1,146 +0,0 @@
|
||||
import { createHash } from 'node:crypto'
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import IngestService from '#services/IngestService'
|
||||
import NameService from '#services/NameService'
|
||||
import { splitVerified } from '#lib/split/index'
|
||||
|
||||
/** Conflicted binaries handled per run; the rest follow on the next tick. */
|
||||
const BATCH = 200
|
||||
|
||||
export interface SplitResult {
|
||||
parents: number
|
||||
volumes: number
|
||||
completed: number
|
||||
}
|
||||
|
||||
/**
|
||||
* Splits conflicted obfuscated binaries (several equal-sized files under one
|
||||
* size+parts identity) into one binary per volume. Only volumes whose
|
||||
* combined part CRCs match the file CRC of their last part are created, so a
|
||||
* split can never produce a corrupt file; the rest waits for more articles.
|
||||
* See CONCEPT.md (D20).
|
||||
*/
|
||||
class SplitService {
|
||||
async splitPending(): Promise<SplitResult> {
|
||||
const parents = await db
|
||||
.from('binaries')
|
||||
.whereNotNull('conflicted_at')
|
||||
.whereNull('parent_binary_id')
|
||||
// split_at = 'infinity' marks binaries that cannot be split (see migration …810).
|
||||
.where((q) => q.whereNull('split_at').orWhereRaw('last_seen_at > split_at'))
|
||||
// Most recently active first: those are the posts still coming in.
|
||||
.orderBy('last_seen_at', 'desc')
|
||||
.limit(BATCH)
|
||||
|
||||
const volumeIds: number[] = []
|
||||
for (const parent of parents) {
|
||||
volumeIds.push(...(await this.split(parent)))
|
||||
}
|
||||
|
||||
// Former conflicts that turned out to be reposts (unflagged, keep their
|
||||
// split_at): no new article may arrive to trigger their completion.
|
||||
const unflagged = await db
|
||||
.from('binaries')
|
||||
.whereNull('conflicted_at')
|
||||
.whereNull('completed_at')
|
||||
.whereNull('parent_binary_id')
|
||||
.whereNotNull('split_at')
|
||||
.whereRaw("split_at <> 'infinity'")
|
||||
.limit(BATCH)
|
||||
.select('id')
|
||||
volumeIds.push(...unflagged.map((b) => Number(b.id)))
|
||||
|
||||
await IngestService.matchPar2Releases(volumeIds)
|
||||
await NameService.resolve(volumeIds)
|
||||
const completed = await IngestService.claimCompleted(volumeIds)
|
||||
await IngestService.enqueue(completed)
|
||||
return { parents: parents.length, volumes: volumeIds.length, completed: completed.length }
|
||||
}
|
||||
|
||||
/** Creates the newly verified volumes of one conflicted binary; returns their ids. */
|
||||
private async split(parent: any): Promise<number[]> {
|
||||
return db.transaction(async (trx) => {
|
||||
// Unverified volumes from before verification existed can never complete.
|
||||
await trx.from('binaries').where('parent_binary_id', parent.id).whereNull('completed_at').delete()
|
||||
|
||||
const { rows } = await trx.rawQuery(
|
||||
`WITH taken AS (
|
||||
SELECT p.message_id FROM parts p JOIN binaries v ON v.id = p.binary_id
|
||||
WHERE v.parent_binary_id = ?
|
||||
), candidates AS (
|
||||
SELECT number, message_id, bytes, article_number, pcrc32, part_size, file_crc32 FROM parts WHERE binary_id = ?
|
||||
UNION ALL
|
||||
SELECT number, message_id, bytes, article_number, pcrc32, part_size, file_crc32 FROM part_conflicts WHERE binary_id = ?
|
||||
)
|
||||
SELECT * FROM candidates WHERE message_id NOT IN (SELECT message_id FROM taken)`,
|
||||
[parent.id, parent.id, parent.id]
|
||||
)
|
||||
const volumes = splitVerified(
|
||||
rows.map((r: any) => ({
|
||||
number: r.number,
|
||||
messageId: r.message_id,
|
||||
bytes: r.bytes,
|
||||
articleNumber: r.article_number === null ? null : Number(r.article_number),
|
||||
pcrc32: r.pcrc32,
|
||||
partSize: r.part_size,
|
||||
fileCrc32: r.file_crc32,
|
||||
})),
|
||||
{ totalParts: parent.total_parts }
|
||||
)
|
||||
|
||||
const now = new Date()
|
||||
const ids: number[] = []
|
||||
for (const articles of volumes) {
|
||||
// A verified volume is fixed, so its first article identifies it.
|
||||
const fileKey = createHash('sha1').update(`${parent.file_key}#${articles[0].messageId}`).digest('hex')
|
||||
const inserted = await trx
|
||||
.knexClient('binaries')
|
||||
.insert({
|
||||
file_key: fileKey,
|
||||
poster: parent.poster,
|
||||
subject: parent.subject,
|
||||
name: parent.name,
|
||||
total_parts: parent.total_parts,
|
||||
groups: JSON.stringify(parent.groups),
|
||||
posted_at: parent.posted_at,
|
||||
size: parent.size,
|
||||
release_id: parent.release_id,
|
||||
parent_binary_id: parent.id,
|
||||
first_seen_at: now,
|
||||
last_seen_at: now,
|
||||
})
|
||||
.onConflict('file_key')
|
||||
.ignore()
|
||||
.returning('id')
|
||||
if (inserted.length === 0) continue
|
||||
const id = Number(inserted[0].id)
|
||||
await trx.knexClient('parts').insert(
|
||||
articles.map((a) => ({
|
||||
binary_id: id,
|
||||
number: a.number,
|
||||
message_id: a.messageId,
|
||||
bytes: a.bytes,
|
||||
article_number: a.articleNumber,
|
||||
pcrc32: a.pcrc32,
|
||||
part_size: a.partSize,
|
||||
file_crc32: a.fileCrc32,
|
||||
}))
|
||||
)
|
||||
// The fingerprint of a volume is that of its part 1 article.
|
||||
await trx.rawQuery(
|
||||
`UPDATE binaries SET md5_16k = (
|
||||
SELECT md5_16k FROM part_conflicts WHERE binary_id = ? AND message_id = ?
|
||||
UNION ALL SELECT md5_16k FROM parts WHERE binary_id = ? AND message_id = ? LIMIT 1)
|
||||
WHERE id = ?`,
|
||||
[parent.id, articles[0].messageId, parent.id, articles[0].messageId, id]
|
||||
)
|
||||
ids.push(id)
|
||||
}
|
||||
|
||||
await trx.from('binaries').where('id', parent.id).update({ split_at: now })
|
||||
return ids
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
export default new SplitService()
|
||||
@@ -1,493 +0,0 @@
|
||||
import { createHash } from 'node:crypto'
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import logger from '@adonisjs/core/services/logger'
|
||||
import indexerConfig from '#config/indexer'
|
||||
import NntpService from '#services/NntpService'
|
||||
import NzbImportService from '#services/NzbImportService'
|
||||
import IngestService, { type OverviewHeader } from '#services/IngestService'
|
||||
import { parseNzb } from '#lib/nzb/index'
|
||||
import {
|
||||
moderationTarget,
|
||||
passwordFromSpot,
|
||||
spotterId,
|
||||
decodeSpotBinary,
|
||||
decodeSpotText,
|
||||
parseSpotArticleHeaders,
|
||||
parseSpotHeader,
|
||||
parseSpotXml,
|
||||
verifySpotHeader,
|
||||
type SpotHeader,
|
||||
type SpotXml,
|
||||
} from '#lib/spotnet/index'
|
||||
|
||||
/** Spots whose NZB is larger than this are skipped (Spotweb skips >50 KB of XML too). */
|
||||
const MAX_NZB_SEGMENTS = 20
|
||||
/** A personal dispose only counts within this long after the spot (Spotweb: 5 days). */
|
||||
const DISPOSE_WINDOW_SECONDS = 5 * 24 * 60 * 60
|
||||
/** Rows per INSERT: 14 columns each, under Postgres' 65,535 bind parameters. */
|
||||
const INSERT_CHUNK = 2000
|
||||
/** Spot images larger than this many articles aren't fetched. */
|
||||
const MAX_IMAGE_SEGMENTS = 10
|
||||
/** A claimed spot whose worker died is taken again after this long. */
|
||||
const RECLAIM_AFTER = '1 hour'
|
||||
|
||||
/** The NZB's articles were read, but they don't decode to an NZB (e.g. "invalid code lengths set"). */
|
||||
export class NzbUnreadable extends Error {}
|
||||
|
||||
/**
|
||||
* A spot's title: from its XML, which is decoded properly (the Subject of a
|
||||
* raw header is read as ISO-8859-1), else from the header.
|
||||
*/
|
||||
function spotTitle(head: { header: SpotHeader; xml: SpotXml | null }): string {
|
||||
return head.xml?.title || head.header.title
|
||||
}
|
||||
|
||||
/**
|
||||
* A placeholder NZB (e.g. Dummy.rar + Dummy.nzb, 2.6 KB, shared by many
|
||||
* spots): less than 1% of the size the spot announces (the real NZB is
|
||||
* handed out elsewhere), or under 16 KB in all (text-only spots: requests,
|
||||
* announcements). Its files are not imported, and the spot is hidden.
|
||||
*/
|
||||
export function isDummyNzb(nzb: { files: { segments: { bytes: number }[] }[] }, spotSize: number): boolean {
|
||||
const bytes = nzb.files.reduce((sum, f) => sum + f.segments.reduce((s, x) => s + x.bytes, 0), 0)
|
||||
return bytes < 16 * 1024 || (spotSize > 0 && bytes < spotSize / 100)
|
||||
}
|
||||
|
||||
export function isSpotGroup(name: string): boolean {
|
||||
return indexerConfig.spotGroups.includes(name)
|
||||
}
|
||||
|
||||
/**
|
||||
* Spotnet (CONCEPT.md D29). Spots are signed posts in free.pt; their full
|
||||
* header holds XML with the message ids of a compressed NZB, which is
|
||||
* imported like a posted NZB (D26). Spot NZBs name many obfuscated posts
|
||||
* (nyuu, ngPost, the skipped multi-group uploader).
|
||||
*/
|
||||
class SpotService {
|
||||
/** Queues the spots among a batch of one spot group's headers. Returns how many. */
|
||||
async recordHeaders(groupName: string, overviews: [number, OverviewHeader][]): Promise<number> {
|
||||
const now = new Date()
|
||||
const rows = []
|
||||
for (const [articleNumber, header] of overviews) {
|
||||
if (!header?.from || !header['message-id']) continue
|
||||
const spot = parseSpotHeader(header.from, header.subject ?? '', header['message-id'], header.date)
|
||||
if (!spot?.title) continue
|
||||
rows.push({
|
||||
message_id: spot.messageId,
|
||||
group_name: groupName,
|
||||
article_number: Number(articleNumber),
|
||||
title: spot.title,
|
||||
tag: spot.tag || null,
|
||||
poster: spot.poster,
|
||||
category: spot.category,
|
||||
subcats: JSON.stringify(spot.subcats),
|
||||
size: spot.size || null,
|
||||
posted_at: spot.stamp ? new Date(spot.stamp * 1000) : null,
|
||||
key_id: spot.keyId,
|
||||
spotter_id: spotterId(spot.selfSignedPubKey) || null,
|
||||
// From the overview; titles with accents may fail here (UTF-8 decoding)
|
||||
// and are verified again on their raw header by worker:spots.
|
||||
verified: verifySpotHeader(spot),
|
||||
created_at: now,
|
||||
})
|
||||
}
|
||||
for (let i = 0; i < rows.length; i += INSERT_CHUNK) {
|
||||
await db.knexQuery().table('spots').insert(rows.slice(i, i + INSERT_CHUNK)).onConflict('message_id').ignore()
|
||||
}
|
||||
return rows.length
|
||||
}
|
||||
|
||||
/**
|
||||
* Takes up to `limit` pending spots, marked `processing`: recent ones
|
||||
* (to be imported) first, then the history in posting order, so live spots
|
||||
* don't wait behind a backfill and moderation follows its targets.
|
||||
*/
|
||||
async claim(limit: number): Promise<any[]> {
|
||||
const claim = async (filter: string, bindings: unknown[], n: number) =>
|
||||
(
|
||||
await db.rawQuery(
|
||||
`UPDATE spots SET status = 'processing', processed_at = now()
|
||||
WHERE message_id IN (
|
||||
SELECT message_id FROM spots WHERE ${filter}
|
||||
ORDER BY article_number LIMIT ? FOR UPDATE SKIP LOCKED)
|
||||
RETURNING *`,
|
||||
[...bindings, n]
|
||||
)
|
||||
).rows as any[]
|
||||
const rows = await claim(`status = 'pending' AND posted_at >= ?`, [this.importCutoff()], limit)
|
||||
if (rows.length < limit) rows.push(...(await claim(`status = 'pending'`, [], limit - rows.length)))
|
||||
if (rows.length < limit) {
|
||||
rows.push(...(await claim(`status = 'processing' AND processed_at < now() - interval '${RECLAIM_AFTER}'`, [], limit - rows.length)))
|
||||
}
|
||||
return rows
|
||||
}
|
||||
|
||||
/** Spots posted before this are listed, not imported: their NZB is read on first download. */
|
||||
importCutoff(): Date {
|
||||
return new Date(Date.now() - indexerConfig.spotImportDays * 24 * 60 * 60 * 1000)
|
||||
}
|
||||
|
||||
/** Old, verified spots that aren't moderation: listed in bulk, without reading anything more. */
|
||||
isListable(spot: any): boolean {
|
||||
return (
|
||||
spot.verified === true &&
|
||||
spot.posted_at !== null &&
|
||||
new Date(spot.posted_at) < this.importCutoff() &&
|
||||
!moderationTarget({ keyId: spot.key_id, title: spot.title })
|
||||
)
|
||||
}
|
||||
|
||||
/**
|
||||
* Creates the listing releases (source `spot`, no files) for these spots.
|
||||
* The release key is the one an import of the spot's NZB uses, so the
|
||||
* import on first download fills this release.
|
||||
*/
|
||||
async list(spots: any[]): Promise<number> {
|
||||
if (spots.length === 0) return 0
|
||||
const keyOf = (spot: any) => createHash('sha1').update(`nzb\n${spot.message_id}`).digest('hex')
|
||||
const ids = await IngestService.upsertReleases(
|
||||
db.connection().getWriteClient(),
|
||||
spots.map((spot) => ({
|
||||
release: { key: keyOf(spot), name: spot.title, source: 'spot' as const },
|
||||
poster: spot.poster,
|
||||
groups: [spot.group_name],
|
||||
postedAt: spot.posted_at ? new Date(spot.posted_at) : undefined,
|
||||
}))
|
||||
)
|
||||
const rows = spots.map((spot) => ({ message_id: spot.message_id, release_id: ids.get(keyOf(spot)), size: spot.size }))
|
||||
await db.rawQuery(
|
||||
`UPDATE releases r SET size = v.size FROM jsonb_to_recordset(?::jsonb) AS v(release_id bigint, size bigint)
|
||||
WHERE r.id = v.release_id AND r.file_count = 0 AND v.size IS NOT NULL`,
|
||||
[JSON.stringify(rows)]
|
||||
)
|
||||
await db.rawQuery(
|
||||
`UPDATE spots s SET status = 'listed', release_id = v.release_id, processed_at = now()
|
||||
FROM jsonb_to_recordset(?::jsonb) AS v(message_id text, release_id bigint)
|
||||
WHERE s.message_id = v.message_id`,
|
||||
[JSON.stringify(rows)]
|
||||
)
|
||||
return rows.length
|
||||
}
|
||||
|
||||
/**
|
||||
* Verifies the spot on its raw header (the signature covers ISO-8859-1
|
||||
* bytes, which the overview, read as UTF-8, may have mangled), then
|
||||
* applies moderation, lists an old spot or imports a recent one's NZB.
|
||||
* Returns the new status.
|
||||
*/
|
||||
async process(spot: any): Promise<string> {
|
||||
const update: Record<string, unknown> = { processed_at: new Date(), error: null }
|
||||
try {
|
||||
const head = await this.readHead(spot.message_id)
|
||||
if (!head) {
|
||||
update.status = 'unverified'
|
||||
} else if (moderationTarget(head.header)) {
|
||||
update.status = 'moderation'
|
||||
await this.moderate(head.header, moderationTarget(head.header)!)
|
||||
} else {
|
||||
update.title = spotTitle(head)
|
||||
update.poster = head.header.poster
|
||||
update.spotter_id = spotterId(head.header.selfSignedPubKey) || null
|
||||
Object.assign(update, this.details(head.xml))
|
||||
if (new Date(head.header.stamp * 1000) < this.importCutoff()) {
|
||||
await this.list([{ ...spot, title: spotTitle(head), poster: head.header.poster }])
|
||||
update.status = 'listed'
|
||||
} else {
|
||||
Object.assign(update, await this.importNzb(spot.message_id, head))
|
||||
}
|
||||
}
|
||||
} catch (error: any) {
|
||||
update.status = 'failed'
|
||||
update.error = String(error.message).slice(0, 500)
|
||||
logger.warn(`Spot ${spot.message_id} "${spot.title}": ${error.message}`)
|
||||
}
|
||||
await db.from('spots').where('message_id', spot.message_id).update(update)
|
||||
if (update.status === 'imported' || update.status === 'listed') await this.applyPassword(spot.message_id)
|
||||
return update.status as string
|
||||
}
|
||||
|
||||
/**
|
||||
* The NZB of a listing, on first download: reads and imports it. Returns
|
||||
* the release that now holds its files (the listing's own, unless the NZB
|
||||
* joined files already known elsewhere), or null.
|
||||
*/
|
||||
async importListing(releaseId: number): Promise<number | null> {
|
||||
const spot = await db.from('spots').where('release_id', releaseId).where('status', 'listed').first()
|
||||
if (!spot) return null
|
||||
return this.reimport(spot)
|
||||
}
|
||||
|
||||
/**
|
||||
* Reads a spot's NZB again and imports it, so its files end up in the
|
||||
* spot's own release under the spot's title. Returns that release, or null.
|
||||
*/
|
||||
async reimport(spot: any): Promise<number | null> {
|
||||
const head = await this.readHead(spot.message_id)
|
||||
if (!head) return null
|
||||
const result = await this.importNzb(spot.message_id, head)
|
||||
await db
|
||||
.from('spots')
|
||||
.where('message_id', spot.message_id)
|
||||
.update({ ...result, ...this.details(head.xml), processed_at: new Date() })
|
||||
await this.applyPassword(spot.message_id)
|
||||
return (result.release_id as number | undefined) ?? null
|
||||
}
|
||||
|
||||
/**
|
||||
* A password the spotter wrote in the spot's title or description (not in
|
||||
* the NZB) goes on its release, unless the release has one already.
|
||||
*/
|
||||
async applyPassword(messageId: string): Promise<string | null> {
|
||||
const spot = await db.from('spots').where('message_id', messageId).whereNotNull('release_id').first()
|
||||
const password = spot ? passwordFromSpot(spot.title, spot.description) : null
|
||||
if (!password) return null
|
||||
await db
|
||||
.from('releases')
|
||||
.where('id', spot.release_id)
|
||||
.whereNull('password')
|
||||
.update({ password, password_source: 'spot' })
|
||||
return password
|
||||
}
|
||||
|
||||
/** Columns for what the spot's XML adds to its overview: description, image, website. */
|
||||
private details(xml: SpotXml | null): Record<string, unknown> {
|
||||
return {
|
||||
description: xml?.description || null,
|
||||
image: xml?.image ? JSON.stringify(xml.image) : null,
|
||||
...(xml?.website ? { website: xml.website } : {}),
|
||||
details_at: new Date(),
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* A spot row with its description and image, reading its full header
|
||||
* first when that hasn't happened yet (old spots are listed from their
|
||||
* overview only). Returns the row as it is when Usenet can't be reached.
|
||||
*/
|
||||
async withDetails(spot: any): Promise<any> {
|
||||
if (spot.details_at) return spot
|
||||
try {
|
||||
const head = await this.readHead(spot.message_id)
|
||||
const update = head ? this.details(head.xml) : { details_at: new Date() }
|
||||
await db.from('spots').where('message_id', spot.message_id).update(update)
|
||||
await this.applyPassword(spot.message_id)
|
||||
return { ...spot, ...update, image: head?.xml?.image ?? null }
|
||||
} catch (error: any) {
|
||||
logger.warn(`Spot ${spot.message_id} details: ${error.message}`)
|
||||
// The article is gone (430): nothing to read later either.
|
||||
if (/^4\d\d/.test(String(error.message))) {
|
||||
await db.from('spots').where('message_id', spot.message_id).update({ details_at: new Date() })
|
||||
}
|
||||
return spot
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Reads the details of up to `limit` imported spots that don't have them
|
||||
* (imported before details were kept), newest first. Returns how many.
|
||||
* Listed spots are left to their first view: there are millions.
|
||||
* Returns how many were read (0 when Usenet can't be reached).
|
||||
*/
|
||||
async fillDetails(limit: number): Promise<number> {
|
||||
const spots = await db
|
||||
.from('spots')
|
||||
.where('status', 'imported')
|
||||
.whereNull('details_at')
|
||||
.orderBy('posted_at', 'desc')
|
||||
.limit(limit)
|
||||
let read = 0
|
||||
for (const spot of spots) if ((await this.withDetails(spot)).details_at) read++
|
||||
return read
|
||||
}
|
||||
|
||||
/**
|
||||
* Takes up to `limit` listed spots whose article and NZB haven't been read,
|
||||
* newest first, marking them so another reader skips them.
|
||||
*/
|
||||
async claimListings(limit: number): Promise<any[]> {
|
||||
const { rows } = await db.rawQuery(
|
||||
`UPDATE spots SET nzb_read_at = now()
|
||||
WHERE message_id IN (
|
||||
SELECT message_id FROM spots WHERE status = 'listed' AND nzb_read_at IS NULL
|
||||
ORDER BY posted_at DESC LIMIT ? FOR UPDATE SKIP LOCKED)
|
||||
RETURNING *`,
|
||||
[limit]
|
||||
)
|
||||
return rows
|
||||
}
|
||||
|
||||
/**
|
||||
* Reads a listed spot's article and NZB without importing its files
|
||||
* (LOG.md 2026-10-08 (10)): the details (description, image), the NZB's
|
||||
* own password (`password_source = 'nzb'`, over one from the spot's text),
|
||||
* else one in the spot's text; a placeholder NZB hides the spot (`dummy`).
|
||||
* Returns what it found. A connection failure gives the spot back for later.
|
||||
*/
|
||||
async readListing(spot: any): Promise<string> {
|
||||
try {
|
||||
const head = await this.readHead(spot.message_id)
|
||||
if (!head) {
|
||||
await db.from('spots').where('message_id', spot.message_id).update({ details_at: new Date() })
|
||||
return 'unverified'
|
||||
}
|
||||
const update: Record<string, unknown> = this.details(head.xml)
|
||||
let outcome = 'nothing'
|
||||
let nzbXml: string | null = null
|
||||
try {
|
||||
nzbXml = await this.readNzb(head)
|
||||
} catch (error: any) {
|
||||
if (error instanceof NzbUnreadable) {
|
||||
// Corrupt: reading it again won't help (LOG.md 2026-10-08 (12)).
|
||||
update.error = `NZB unreadable: ${error.message}`.slice(0, 500)
|
||||
outcome = 'nzb-unreadable'
|
||||
} else if (/^4\d\d/.test(String(error.message))) {
|
||||
// The NZB's articles are gone: the listing can't be downloaded any more.
|
||||
update.error = 'NZB articles gone'
|
||||
outcome = 'nzb-gone'
|
||||
} else {
|
||||
throw error
|
||||
}
|
||||
}
|
||||
const nzb = nzbXml ? parseNzb(nzbXml) : null
|
||||
if (nzb && nzb.files.length > 0 && isDummyNzb(nzb, head.header.size)) {
|
||||
update.status = 'dummy'
|
||||
outcome = 'dummy'
|
||||
} else if (nzb?.meta.password && spot.release_id) {
|
||||
await db
|
||||
.from('releases')
|
||||
.where('id', spot.release_id)
|
||||
.where((q) => q.whereNull('password').orWhere('password_source', 'spot'))
|
||||
.update({ password: nzb.meta.password, password_source: 'nzb' })
|
||||
outcome = 'nzb-password'
|
||||
}
|
||||
await db.from('spots').where('message_id', spot.message_id).update(update)
|
||||
if (outcome !== 'dummy' && (await this.applyPassword(spot.message_id)) && outcome !== 'nzb-password') outcome = 'spot-password'
|
||||
return outcome
|
||||
} catch (error: any) {
|
||||
// The spot's own article is gone (430): nothing to read later either.
|
||||
if (/^4\d\d/.test(String(error.message))) {
|
||||
await db.from('spots').where('message_id', spot.message_id).update({ details_at: new Date() })
|
||||
return 'gone'
|
||||
}
|
||||
await db.from('spots').where('message_id', spot.message_id).update({ nzb_read_at: null })
|
||||
logger.warn(`Spot ${spot.message_id} listing: ${error.message}`)
|
||||
return 'retry'
|
||||
}
|
||||
}
|
||||
|
||||
/** Reads up to `limit` unread listings, `concurrency` at a time. Returns the outcomes counted. */
|
||||
async readListings(limit: number, concurrency: number): Promise<Record<string, number>> {
|
||||
const spots = await this.claimListings(limit)
|
||||
const counts: Record<string, number> = {}
|
||||
let next = 0
|
||||
await Promise.all(
|
||||
Array.from({ length: Math.max(1, concurrency) }, async () => {
|
||||
while (next < spots.length) {
|
||||
const outcome = await this.readListing(spots[next++])
|
||||
counts[outcome] = (counts[outcome] ?? 0) + 1
|
||||
}
|
||||
})
|
||||
)
|
||||
return counts
|
||||
}
|
||||
|
||||
/** The spot's image as posted on Usenet (Spotnet posts them uncompressed), or null. */
|
||||
async image(spot: any): Promise<Buffer | null> {
|
||||
const segments: string[] = spot.image?.segments ?? []
|
||||
if (segments.length === 0 || segments.length > MAX_IMAGE_SEGMENTS) return null
|
||||
const bodies = []
|
||||
for (const id of segments) bodies.push(await this.fetch(`BODY <${id.replace(/^<|>$/g, '')}>`))
|
||||
return decodeSpotBinary(bodies, false)
|
||||
}
|
||||
|
||||
/** The spot's raw header, verified; null when it doesn't verify. */
|
||||
private async readHead(messageId: string): Promise<{ header: SpotHeader; xml: SpotXml | null } | null> {
|
||||
const raw = await this.fetch(`HEAD <${messageId}>`)
|
||||
const { headers, xml } = parseSpotArticleHeaders(raw.toString('latin1'))
|
||||
const header = parseSpotHeader(headers.from ?? '', headers.subject ?? '', messageId, headers.date)
|
||||
if (!header || !verifySpotHeader(header)) return null
|
||||
return { header, xml: parseSpotXml(decodeSpotText(xml)) }
|
||||
}
|
||||
|
||||
/** Downloads and decodes the spot's compressed NZB; null when it has none (or a too large one). */
|
||||
private async readNzb(head: { xml: SpotXml | null }): Promise<string | null> {
|
||||
const segments = head.xml?.nzbSegments ?? []
|
||||
if (segments.length === 0 || segments.length > MAX_NZB_SEGMENTS) return null
|
||||
const bodies = []
|
||||
for (const id of segments) bodies.push(await this.fetch(`BODY <${id.replace(/^<|>$/g, '')}>`))
|
||||
try {
|
||||
return decodeSpotBinary(bodies).toString('utf8')
|
||||
} catch (error: any) {
|
||||
throw new NzbUnreadable(String(error.message))
|
||||
}
|
||||
}
|
||||
|
||||
/** Downloads the spot's compressed NZB and imports it as a release (D26). */
|
||||
private async importNzb(messageId: string, head: { header: SpotHeader; xml: SpotXml | null }): Promise<Record<string, unknown>> {
|
||||
const nzbXml = await this.readNzb(head)
|
||||
if (!nzbXml) return { status: 'no-nzb' }
|
||||
const nzb = parseNzb(nzbXml)
|
||||
if (!nzb || nzb.files.length === 0) return { status: 'no-nzb' }
|
||||
const { header } = head
|
||||
if (isDummyNzb(nzb, header.size)) return { status: 'dummy' }
|
||||
const title = spotTitle(head)
|
||||
const result = await NzbImportService.import(
|
||||
{
|
||||
id: null,
|
||||
filename: `${title}.nzb`,
|
||||
poster: header.poster,
|
||||
groups: [...new Set(nzb.files.flatMap((f) => f.groups))],
|
||||
date: header.stamp * 1000,
|
||||
first_message_id: messageId,
|
||||
},
|
||||
nzbXml,
|
||||
{ name: title, source: 'spot', moveContainer: false, trusted: true }
|
||||
)
|
||||
return result ? { status: 'imported', release_id: result.releaseId } : { status: 'no-nzb' }
|
||||
}
|
||||
|
||||
/**
|
||||
* Applies a delete/dispose/remove: the target spot is marked `removed`
|
||||
* (its release is then hidden from search). As in Spotweb, a personal
|
||||
* dispose must come from the same spotter within 5 days; a moderator's
|
||||
* (signed with Spotnet's key, no personal key) always counts.
|
||||
*/
|
||||
private async moderate(header: SpotHeader, targetId: string) {
|
||||
const target = await db.from('spots').where('message_id', targetId).first()
|
||||
if (!target) return
|
||||
const personal = header.selfSignedPubKey.length > 50 ? spotterId(header.selfSignedPubKey) : ''
|
||||
if (personal) {
|
||||
const age = header.stamp - (target.posted_at ? new Date(target.posted_at).getTime() / 1000 : 0)
|
||||
if (target.spotter_id !== personal || age > DISPOSE_WINDOW_SECONDS) return
|
||||
}
|
||||
await db.from('spots').where('message_id', targetId).update({ status: 'removed' })
|
||||
logger.info(`Spot "${target.title}" removed by ${personal ? 'its poster' : 'a moderator'}.`)
|
||||
// Spots that joined its release (a correction, a repost) take the files over now.
|
||||
if (target.release_id) {
|
||||
const sharing = await db
|
||||
.from('spots')
|
||||
.where('release_id', target.release_id)
|
||||
.where('status', 'imported')
|
||||
.whereNot('message_id', targetId)
|
||||
for (const spot of sharing) await this.reimport(spot)
|
||||
}
|
||||
}
|
||||
|
||||
/** One HEAD or BODY, as raw bytes. */
|
||||
private async fetch(command: string): Promise<Buffer> {
|
||||
const conn = await NntpService.acquire()
|
||||
let failed = false
|
||||
try {
|
||||
const { data } = await NntpService.timed<any>(conn._artcmd(command), command)
|
||||
return Buffer.from(data.buffer, data.byteOffset, data.byteLength)
|
||||
} catch (error: any) {
|
||||
// 430 (no such article) is an answer, not a broken connection.
|
||||
failed = !/^4\d\d/.test(String(error.message))
|
||||
throw error
|
||||
} finally {
|
||||
NntpService.release(conn, failed)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
export default new SpotService()
|
||||
@@ -1,252 +0,0 @@
|
||||
import { createHash } from 'node:crypto'
|
||||
import db from '@adonisjs/lucid/services/db'
|
||||
import indexerConfig from '#config/indexer'
|
||||
import IngestService, { type IngestSegment } from '#services/IngestService'
|
||||
import { counterFileKey } from '#lib/subject/index'
|
||||
import { fileStem } from '#lib/release/index'
|
||||
|
||||
/** Rows per statement (a key and a JSON array each). */
|
||||
const CHUNK = 5000
|
||||
|
||||
/**
|
||||
* Staging of lone header segments (CONCEPT.md D35). Uploads that give every
|
||||
* article its own random subject turned each article into a binary, a part
|
||||
* and a release, which GC deleted two hours later: ~3.9M binaries an hour
|
||||
* against ~123k real files. A segment whose file is new and alone in its
|
||||
* fetch batch now waits in `staged_segments` (unlogged, one index) until a
|
||||
* second segment of that file arrives; only then do both go into the real
|
||||
* tables. GC drops what waited longer than STAGING_TTL_SECONDS (30 min).
|
||||
*
|
||||
* Counter grouping (D36): those random-subject uploads still keep poster,
|
||||
* file counter `[i/M]` and part count per file. Staged segments also carry
|
||||
* that key (`counter_key`); once two staged segments from different subjects
|
||||
* share it, they become one file under that key, and later segments with it
|
||||
* join that file directly.
|
||||
*/
|
||||
class StagingService {
|
||||
enabled(groupName: string): boolean {
|
||||
const groups = indexerConfig.stagingGroups
|
||||
return groups.includes('*') || groups.includes(groupName)
|
||||
}
|
||||
|
||||
/** Whether counter grouping (D36) applies to this group. */
|
||||
countersEnabled(groupName: string): boolean {
|
||||
const groups = indexerConfig.counterGroupingGroups
|
||||
return groups.includes('*') || groups.includes(groupName)
|
||||
}
|
||||
|
||||
/**
|
||||
* The segments of a header batch that should be ingested now: the batch's
|
||||
* own, minus lone segments of new files (staged instead), plus staged
|
||||
* segments whose file this batch shows to be real. `staged` counts the
|
||||
* segments that went to staging.
|
||||
*/
|
||||
async filter(segments: IngestSegment[], { counters = false } = {}): Promise<{ ingest: IngestSegment[]; staged: number }> {
|
||||
const byKey = new Map<string, IngestSegment[]>()
|
||||
for (const segment of segments) {
|
||||
const list = byKey.get(segment.fileKey)
|
||||
if (list) list.push(segment)
|
||||
else byKey.set(segment.fileKey, [segment])
|
||||
}
|
||||
|
||||
const ingest: IngestSegment[] = []
|
||||
const lone: IngestSegment[] = []
|
||||
for (const list of byKey.values()) {
|
||||
// Single-segment files are complete; two different articles of a file in one batch make it real.
|
||||
// (One article twice is a crosspost, LOG.md 2026-10-02.)
|
||||
const articles = new Set(list.map((s) => s.messageId)).size
|
||||
if (list[0].totalParts <= 1 || articles >= 2 || list[0].weakIdentity || list[0].size != null) ingest.push(...list)
|
||||
else lone.push(list[0])
|
||||
}
|
||||
|
||||
const counterOf = new Map<IngestSegment, string>()
|
||||
if (counters) {
|
||||
for (const segment of lone) {
|
||||
const key = counterFileKey(segment)
|
||||
if (key) counterOf.set(segment, key)
|
||||
}
|
||||
}
|
||||
|
||||
// Files we already have a binary for (by subject, or by counters) take their segments directly.
|
||||
const known = new Set<string>()
|
||||
const lookup = [...new Set([...lone.map((s) => s.fileKey), ...counterOf.values()])]
|
||||
for (let i = 0; i < lookup.length; i += CHUNK) {
|
||||
const rows = await db.from('binaries').whereIn('file_key', lookup.slice(i, i + CHUNK)).select('file_key')
|
||||
for (const row of rows) known.add(row.file_key)
|
||||
}
|
||||
const toStage: IngestSegment[] = []
|
||||
const knownCounters = new Set<string>()
|
||||
for (const segment of lone) {
|
||||
const counter = counterOf.get(segment)
|
||||
if (known.has(segment.fileKey)) {
|
||||
ingest.push(segment)
|
||||
} else if (counter && known.has(counter)) {
|
||||
ingest.push(asCounterFile(segment, counter))
|
||||
knownCounters.add(counter)
|
||||
} else {
|
||||
toStage.push(segment)
|
||||
}
|
||||
}
|
||||
|
||||
// Stage the rest; a file that now has two staged segments comes out again.
|
||||
const promoted: string[] = []
|
||||
// Key order, so concurrent jobs lock rows in the same order.
|
||||
toStage.sort((a, b) => a.fileKey.localeCompare(b.fileKey))
|
||||
for (let i = 0; i < toStage.length; i += CHUNK) {
|
||||
const rows = toStage.slice(i, i + CHUNK).map((s) => ({
|
||||
file_key: s.fileKey,
|
||||
counter_key: counterOf.get(s) ?? null,
|
||||
segments: [serialize(s)],
|
||||
staged_at: new Date(),
|
||||
}))
|
||||
const { rows: result } = await retryDeadlock(() => db.rawQuery(
|
||||
`INSERT INTO staged_segments (file_key, counter_key, segments, staged_at)
|
||||
SELECT file_key, counter_key, segments, staged_at
|
||||
FROM jsonb_to_recordset(?::jsonb) AS v(file_key text, counter_key text, segments jsonb, staged_at timestamptz)
|
||||
ON CONFLICT (file_key) DO UPDATE SET segments = CASE
|
||||
-- The same article again (a crosspost, or a re-fetched range) is no second segment.
|
||||
WHEN EXISTS (SELECT 1 FROM jsonb_array_elements(staged_segments.segments) e
|
||||
WHERE e->>'messageId' = EXCLUDED.segments->0->>'messageId') THEN staged_segments.segments
|
||||
ELSE staged_segments.segments || EXCLUDED.segments END
|
||||
RETURNING file_key, jsonb_array_length(segments) AS n`,
|
||||
[JSON.stringify(rows)]
|
||||
))
|
||||
for (const row of result) if (Number(row.n) >= 2) promoted.push(row.file_key)
|
||||
}
|
||||
|
||||
// Counter files: two staged rows (two subjects) with one counter key make a file.
|
||||
const stagedCounters = [...new Set(toStage.map((s) => counterOf.get(s)).filter((k): k is string => Boolean(k)))]
|
||||
const readyCounters: string[] = []
|
||||
for (let i = 0; i < stagedCounters.length; i += CHUNK) {
|
||||
const { rows } = await db.rawQuery(
|
||||
`SELECT counter_key FROM staged_segments WHERE counter_key = ANY(?) GROUP BY counter_key HAVING count(*) >= 2`,
|
||||
[stagedCounters.slice(i, i + CHUNK)]
|
||||
)
|
||||
for (const row of rows) readyCounters.push(row.counter_key)
|
||||
}
|
||||
|
||||
// Promoted files, staged segments of files this batch ingests anyway, and ready counter files.
|
||||
const fileKeys = [...new Set([...promoted, ...ingest.map((s) => s.fileKey)])]
|
||||
const counterKeys = [...new Set([...readyCounters, ...knownCounters])]
|
||||
const seen = new Set(ingest.map((s) => s.messageId))
|
||||
for (const segment of await this.take(fileKeys, counterKeys)) {
|
||||
if (!seen.has(segment.messageId)) ingest.push(segment)
|
||||
seen.add(segment.messageId)
|
||||
}
|
||||
const staged = toStage.filter((s) => !seen.has(s.messageId)).length
|
||||
return { ingest, staged }
|
||||
}
|
||||
|
||||
/**
|
||||
* GC: drops staged segments that waited longer than `maxAgeSeconds`, then
|
||||
* ingests those whose file got a binary meanwhile (two fetch jobs racing).
|
||||
* Expiry comes first so the table stays bounded whatever the budget; the
|
||||
* adoption starts from recently seen binaries (index on last_seen_at,
|
||||
* which moves in 10-minute steps) rather than joining every staged row.
|
||||
*/
|
||||
async sweep(maxAgeSeconds: number, budget: number): Promise<{ adopted: number; expired: number }> {
|
||||
let adopted = 0
|
||||
let expired = 0
|
||||
while (Date.now() < budget) {
|
||||
const result = await db.rawQuery(
|
||||
`DELETE FROM staged_segments WHERE ctid IN (
|
||||
SELECT ctid FROM staged_segments WHERE staged_at < now() - make_interval(secs => ?) LIMIT 20000)`,
|
||||
[maxAgeSeconds]
|
||||
)
|
||||
expired += result.rowCount
|
||||
if (result.rowCount < 20000) break
|
||||
}
|
||||
let lastId = 0
|
||||
while (Date.now() < budget) {
|
||||
const recent = await db
|
||||
.from('binaries')
|
||||
.where('last_seen_at', '>', new Date(Date.now() - (maxAgeSeconds + 15 * 60) * 1000))
|
||||
.where('id', '>', lastId)
|
||||
.orderBy('id')
|
||||
.limit(CHUNK)
|
||||
.select('id', 'file_key')
|
||||
if (recent.length === 0) break
|
||||
lastId = Number(recent[recent.length - 1].id)
|
||||
const keys = recent.map((b) => b.file_key as string)
|
||||
// A binary's key is a subject key or a counter key; staged rows are looked up by both.
|
||||
const segments = await this.take(keys, keys)
|
||||
if (segments.length > 0) await IngestService.ingest(segments)
|
||||
adopted += segments.length
|
||||
}
|
||||
return { adopted, expired }
|
||||
}
|
||||
|
||||
/**
|
||||
* Removes and returns the staged segments of these files: by subject key,
|
||||
* or by counter key (returned as segments of the counter file).
|
||||
*/
|
||||
private async take(fileKeys: string[], counterKeys: string[] = []): Promise<IngestSegment[]> {
|
||||
const taken: IngestSegment[] = []
|
||||
for (let i = 0; i < fileKeys.length; i += CHUNK) {
|
||||
const { rows } = await retryDeadlock(() =>
|
||||
db.rawQuery(`DELETE FROM staged_segments WHERE file_key = ANY(?) RETURNING segments`, [fileKeys.slice(i, i + CHUNK)])
|
||||
)
|
||||
for (const row of rows) taken.push(...(row.segments as any[]).map(deserialize))
|
||||
}
|
||||
for (let i = 0; i < counterKeys.length; i += CHUNK) {
|
||||
const { rows } = await retryDeadlock(() =>
|
||||
db.rawQuery(`DELETE FROM staged_segments WHERE counter_key = ANY(?) RETURNING counter_key, segments`, [counterKeys.slice(i, i + CHUNK)])
|
||||
)
|
||||
for (const row of rows) taken.push(...(row.segments as any[]).map((s) => asCounterFile(deserialize(s), row.counter_key)))
|
||||
}
|
||||
return taken
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* A segment as part of its counter file (D36): the counter key is its file
|
||||
* key, it is a weak identity (a repeated part number from another upload
|
||||
* marks the binary conflicted), and its release is a weak one of its own,
|
||||
* which the upload's par2 set can gather later.
|
||||
*/
|
||||
function asCounterFile(segment: IngestSegment, counterKey: string): IngestSegment {
|
||||
return {
|
||||
...segment,
|
||||
fileKey: counterKey,
|
||||
weakIdentity: true,
|
||||
release: {
|
||||
key: createHash('sha1').update(`counter-release\n${counterKey}`).digest('hex'),
|
||||
name: fileStem(segment.name),
|
||||
source: 'weak-stem',
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Runs a staging statement again when it lost a deadlock: concurrent fetch
|
||||
* jobs insert by file key and delete by counter key, so their row locks can
|
||||
* cross. Postgres rolled the statement back, and each one is safe to repeat.
|
||||
*/
|
||||
async function retryDeadlock<T>(statement: () => Promise<T>, attempts = 5): Promise<T> {
|
||||
for (let attempt = 1; ; attempt++) {
|
||||
try {
|
||||
return await statement()
|
||||
} catch (error: any) {
|
||||
if (error?.code !== '40P01' || attempt >= attempts) throw error
|
||||
await new Promise((resolve) => setTimeout(resolve, 50 * attempt + Math.random() * 100))
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
function serialize(segment: IngestSegment) {
|
||||
return {
|
||||
...segment,
|
||||
postedAt: segment.postedAt?.toISOString(),
|
||||
articleNumber: segment.articleNumber?.toString(),
|
||||
}
|
||||
}
|
||||
|
||||
function deserialize(stored: any): IngestSegment {
|
||||
return {
|
||||
...stored,
|
||||
postedAt: stored.postedAt ? new Date(stored.postedAt) : undefined,
|
||||
articleNumber: stored.articleNumber ?? undefined,
|
||||
}
|
||||
}
|
||||
|
||||
export default new StagingService()
|
||||
Reference in new issue
Block a user