110 lines
5.8 KiB
TypeScript
110 lines
5.8 KiB
TypeScript
import db from '@adonisjs/lucid/services/db'
|
|
import coordinatorConfig from '#config/coordinator'
|
|
import RuleSets from '#services/rule_sets'
|
|
import Spool from '#services/spool'
|
|
import Corpus from '#services/corpus'
|
|
import CatalogSpace from '#services/catalog_space'
|
|
import type { Lease } from '#lib/edge/index'
|
|
|
|
export class LeaseLost extends Error {}
|
|
|
|
/**
|
|
* Ranges for workers (REBUILD.md, coordinator): the next range with chunks
|
|
* spooled, by priority (oldest era first), else the next planned one (the
|
|
* prefetcher fetches leased ranges first; the worker's stream waits). A
|
|
* worker never waits for prefetching to choose its range: refused leases
|
|
* left workers idle for 30 s, so fewer ranges were leased and fewer
|
|
* prefetched (LOG.md 2026-10-09). A lease whose worker went quiet for
|
|
* LEASE_TTL_SECONDS is taken over and resumes at its resume point.
|
|
*/
|
|
/**
|
|
* A group's batches wait for each other (the batch store locks the group): more workers than this on one group only
|
|
* queue. Eight on one picture group's gap ranges (priority -1, in id order) held the node API's batches for minutes
|
|
* (LOG.md 2026-10-16 (68)).
|
|
*/
|
|
const MAX_LEASES_PER_GROUP = 2
|
|
|
|
class Leases {
|
|
async lease(nodeId: number): Promise<Lease | null> {
|
|
const rules = await RuleSets.current()
|
|
// The catalog at its cap on the NVMe: the tail and the stored ranges only (CatalogSpace).
|
|
const history = (await CatalogSpace.full()) ? 'AND priority < 0' : ''
|
|
const take = (spooled: boolean) =>
|
|
db.rawQuery(
|
|
`UPDATE ranges r SET status = 'leased', node_id = ?, leased_at = now(), heartbeat_at = now(), attempts = attempts + 1, rule_version = ?
|
|
WHERE r.id = (
|
|
SELECT id FROM ranges
|
|
WHERE ${spooled ? 'spooled_chunks > 0' : 'true'} AND next_id <= end_id ${history}
|
|
AND (status = 'planned' OR (status = 'leased' AND heartbeat_at < now() - make_interval(secs => ?)))
|
|
AND (SELECT count(*) FROM ranges l
|
|
WHERE l.status = 'leased' AND l.group_id = ranges.group_id AND l.id <> ranges.id
|
|
AND l.heartbeat_at >= now() - make_interval(secs => ?)) < ${MAX_LEASES_PER_GROUP}
|
|
ORDER BY priority, id LIMIT 1 FOR UPDATE SKIP LOCKED)
|
|
RETURNING r.id, r.group_id, r.start_id, r.end_id, r.next_id`,
|
|
[nodeId, rules.version, coordinatorConfig.leaseTtlSeconds, coordinatorConfig.leaseTtlSeconds]
|
|
)
|
|
let { rows } = await take(true)
|
|
if (rows.length === 0) ({ rows } = await take(false))
|
|
const row = rows[0]
|
|
if (!row) return null
|
|
await db.from('nodes').where('id', nodeId).update({ last_seen_at: new Date() })
|
|
const group = await db.from('groups').where('id', row.group_id).select('name').firstOrFail()
|
|
return { range: { id: Number(row.id), group: group.name, start: Number(row.start_id), end: Number(row.end_id), next: Number(row.next_id) }, rules }
|
|
}
|
|
|
|
async leased(rangeId: number, nodeId: number) {
|
|
const range = await db.from('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
|
|
}
|
|
|
|
async heartbeat(rangeId: number, nodeId: number) {
|
|
await this.leased(rangeId, nodeId)
|
|
await db.from('ranges').where('id', rangeId).update({ heartbeat_at: new Date() })
|
|
await db.from('nodes').where('id', nodeId).update({ last_seen_at: new Date() })
|
|
}
|
|
|
|
/** The worker reached the range's end: done, and its spooled chunks go. */
|
|
async finish(rangeId: number, nodeId: number) {
|
|
const range = await this.leased(rangeId, nodeId)
|
|
if (Number(range.next_id) <= Number(range.end_id)) throw new LeaseLost(`Range ${rangeId} is at ${range.next_id}, not past its end ${range.end_id}`)
|
|
await db.from('ranges').where('id', rangeId).update({ status: 'done', done_at: new Date(), heartbeat_at: new Date() })
|
|
await Spool.remove(rangeId)
|
|
}
|
|
|
|
/**
|
|
* A worker got a chunk that doesn't inflate: the spooled chunk goes and
|
|
* the range back to the queue (the prefetcher fetches leased ranges'
|
|
* missing chunks again; the next worker resumes at the range's resume point).
|
|
*/
|
|
async corrupt(rangeId: number, nodeId: number, from: number) {
|
|
const range = await this.leased(rangeId, nodeId)
|
|
const span = Spool.chunksOf({ start: Number(range.start_id), end: Number(range.end_id) }).find((s) => s.from === from)
|
|
if (span) {
|
|
await Spool.removeChunk(rangeId, span)
|
|
if (range.capture) {
|
|
const group = await db.from('groups').where('id', range.group_id).select('name').first()
|
|
await Corpus.removeChunk({ group: group.name, start: Number(range.start_id), end: Number(range.end_id) }, span)
|
|
}
|
|
}
|
|
const plain: number[] = (range.plain_chunks ?? []).map(Number)
|
|
// Reported once already: the compressed copy may have been in flight (or the prefetcher's targets not
|
|
// refreshed yet) and spooled again (range 161820, LOG.md 2026-10-11), so a second try plain first.
|
|
if (plain.filter((p) => p === from).length >= 2) {
|
|
// Fetched plain (deflated here) and still not inflating: not the server's compression; set aside.
|
|
await db.from('ranges').where('id', rangeId).update({ status: 'failed', node_id: null, error: `chunk ${from} doesn't inflate, also fetched plain` })
|
|
return
|
|
}
|
|
// The server's compressed reply is cut short (the same bytes every time): fetched plain next time.
|
|
await db.rawQuery(`UPDATE ranges SET status = 'planned', node_id = NULL, plain_chunks = array_append(plain_chunks, ?::bigint) WHERE id = ?`, [from, rangeId])
|
|
}
|
|
|
|
/** A stopping worker gives its range back; the next one resumes it. */
|
|
async release(rangeId: number, nodeId: number) {
|
|
await this.leased(rangeId, nodeId)
|
|
await db.from('ranges').where('id', rangeId).update({ status: 'planned', node_id: null })
|
|
}
|
|
}
|
|
|
|
export default new Leases()
|