From 99fffe30f4f4f689a38fcb20f14330b22e244b34 Mon Sep 17 00:00:00 2001
From: Daan Meijer <daan@studioseptember.nl>
Date: Fri, 9 Oct 2026 17:49:45 +0200
Subject: [PATCH] Rebuild: stats from a fixed-size sample, refreshed every 5
 minutes, given up after 2

TABLESAMPLE SYSTEM (0.5) read 30 s of random pages on the HDD at 7M
releases, growing with the catalog; a refresh that never finished kept
the header's totals 3.5 hours old (it blocked every later refresh).
Now TABLESAMPLE SYSTEM_ROWS (5000) (tsm_system_rows), every 5 minutes,
and a refresh is given up after 2 minutes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 app/services/search_service.ts                   | 16 +++++++++++-----
 .../migrations/1770000001600_tsm_system_rows.ts  | 12 ++++++++++++
 2 files changed, 23 insertions(+), 5 deletions(-)
 create mode 100644 database/migrations/1770000001600_tsm_system_rows.ts

diff --git a/app/services/search_service.ts b/app/services/search_service.ts
index ef3e34a..3dc0c88 100644
--- a/app/services/search_service.ts
+++ b/app/services/search_service.ts
@@ -61,7 +61,9 @@ const SORTS: Record<SortOrder, string> = {
 
 /** A broad search needn't count every match (each counted row is a heap read). */
 const COUNT_LIMIT = 10_000
-const STATS_TTL_MS = 60 * 1000
+const STATS_TTL_MS = 5 * 60 * 1000
+/** A refresh that takes longer is given up (the next request tries again). */
+const STATS_TIMEOUT_MS = 2 * 60 * 1000
 
 /** ILIKE pattern of one word (the words are split on dots, dashes, underscores and spaces). */
 function contains(word: string): string {
@@ -188,7 +190,10 @@ class SearchService {
   /** Totals for the web interface, computed at most once a minute (callers in between share them). */
   async stats() {
     if (!this.statsCache || Date.now() - this.statsCache.at > STATS_TTL_MS) {
-      this.statsRefresh ??= this.computeStats()
+      this.statsRefresh ??= Promise.race([
+        this.computeStats(),
+        new Promise<never>((_, reject) => setTimeout(() => reject(new Error('timed out')), STATS_TIMEOUT_MS).unref()),
+      ])
         .then((value) => {
           this.statsCache = { at: Date.now(), value }
         })
@@ -210,13 +215,14 @@ class SearchService {
   /**
    * Estimates (a catalog of 100M+ rows can't be counted on every visit):
    * the row count from Postgres's statistics, below a million rows exact;
-   * readable share, files and size from a sample; ranges counted exactly
-   * (a few hundred thousand rows).
+   * readable share, files and size from a sample of 5,000 rows; ranges
+   * counted exactly (a few hundred thousand rows).
    */
   private async computeStats() {
     const { rows } = await db.rawQuery(`SELECT reltuples::bigint AS n FROM pg_class WHERE relname = 'releases'`)
     const estimate = Number(rows[0]?.n ?? 0)
-    const sample = estimate < 1_000_000 ? '' : 'TABLESAMPLE SYSTEM (0.5)'
+    // A fixed-size sample: a percentage of a big catalog is many random reads on the HDD.
+    const sample = estimate < 1_000_000 ? '' : 'TABLESAMPLE SYSTEM_ROWS (5000)'
     const { rows: s } = await db.rawQuery(
       `SELECT count(*) AS sampled, count(*) FILTER (WHERE readable) AS readable, coalesce(sum(files), 0) AS files, coalesce(sum(bytes), 0) AS bytes
        FROM releases ${sample} WHERE files > 0`
diff --git a/database/migrations/1770000001600_tsm_system_rows.ts b/database/migrations/1770000001600_tsm_system_rows.ts
new file mode 100644
index 0000000..c5fc7a8
--- /dev/null
+++ b/database/migrations/1770000001600_tsm_system_rows.ts
@@ -0,0 +1,12 @@
+import { BaseSchema } from '@adonisjs/lucid/schema'
+
+/** TABLESAMPLE SYSTEM_ROWS: the stats' sample of the catalog at a fixed size (a percentage read 30 s of random pages on the HDD at 7M releases). */
+export default class extends BaseSchema {
+  async up() {
+    this.schema.raw(`CREATE EXTENSION IF NOT EXISTS tsm_system_rows`)
+  }
+
+  async down() {
+    this.schema.raw(`DROP EXTENSION IF EXISTS tsm_system_rows`)
+  }
+}