Par2 size matching 100x faster; release bookkeeping once per statement

matchPar2Releases checked every same-size release_files row's set and
date (18 s per batch of body-derived binaries); now an index on
release_files (release_id, size) drops the repeated sizes first: 0.16 s.

File inserts updated their release in four row triggers; one statement
trigger now updates each release once per INSERT. Completeness is noted
in release_dirty and recomputed by gc:scheduler every 30 minutes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

app/services/IngestService.ts

@@ -272,18 +272,27 @@ class IngestService {
    */
   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 b2.id, min(rf.release_id) AS release_id
-         FROM binaries b2
-         JOIN release_files rf ON rf.size = b2.size
-         JOIN releases r ON r.id = rf.release_id
-         WHERE b2.id = ANY(?) AND b2.release_id IS NULL AND b2.md5_16k IS NULL AND b2.size >= ?
-           AND abs(extract(epoch FROM (r.posted_at - b2.posted_at))) < ?
-           AND (SELECT count(*) FROM release_files x WHERE x.release_id = rf.release_id AND x.size = rf.size) = 1
-         GROUP BY b2.id
-         HAVING count(DISTINCT rf.release_id) = 1
+         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]

app/services/ReleaseCompleteness.ts

@@ -0,0 +1,35 @@
+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()