fix(wiki): cap requeueFailedOps retry count to prevent queue pile-up

Root cause: requeueFailedOps unconditionally RPush'd every failed op back to
wiki:pending, causing unbounded growth when a document consistently triggers
LLM timeouts. Observed: 553 entries in wiki:pending:08134644 (one KB), mostly
duplicates for ~5 unique documents that each timed out on every batch cycle.

Fix:
- Add wikiFailCountKeyPrefix ("wiki:failcount:") and wikiMaxFailRetries (5)
  constants in wiki_ingest.go
- requeueFailedOps now atomically Incr(wiki:failcount:{kbID}:{knowledgeID})
  before each RPush; skips re-queue (drops op permanently) if count > 5
- Expire the fail-count key at wikiPendingTTL (24h) on every update
- wiki_ingest_batch.go: Del the fail-count key at ingestSucceeded++ so
  transient errors don't permanently burn through a document's retry budget

Evidence of problem severity: 5 persistently-failing docs × ~110 requeue
cycles each = 553 pending entries; had to use Lua atomic dedup to recover.

Behaviour after this fix:
- First 5 failures → normal retry (re-queued, follow-up batch scheduled)
- 6th failure → logged as WARN "dropping op … after 6 failures (limit 5)",
  not re-queued; prevents the feedback loop
- If document is fixed upstream and re-ingested via EnqueueWikiIngest, the
  op is a fresh entry with no existing fail-count key → full 5-retry budget

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
toy0116
2026-05-07 17:50:21 +08:00
committed by lyingbug
parent edbcfa80ab
commit bf12868923
2 changed files with 40 additions and 1 deletions
+33 -1
View File
@@ -57,6 +57,21 @@ const (
// and are picked up by the follow-up task.
wikiMaxDocsPerBatch = 5
// wikiFailCountKeyPrefix is the Redis key prefix for per-document failure
// counters. Key: wiki:failcount:{kbID}:{knowledgeID} → integer.
// Incremented each time requeueFailedOps retries an op; reset to 0 on
// successful ingest. Once the counter exceeds wikiMaxFailRetries the op is
// dropped instead of re-queued, preventing LLM-timeout storms from causing
// unbounded wiki:pending growth.
wikiFailCountKeyPrefix = "wiki:failcount:"
// wikiMaxFailRetries is the maximum number of times a single document op
// may be re-queued via requeueFailedOps before it is permanently dropped.
// 5 retries ≈ five full batch cycles (each with a ~30 s delay), giving
// transient LLM errors a fair chance to recover without letting a
// persistently-broken doc clog the queue indefinitely.
wikiMaxFailRetries = 5
// wikiDeletedKeyPrefix is the Redis key prefix for "recently deleted
// knowledge" tombstones. Key: wiki:deleted:{kbID}:{knowledgeID}. Written
// by cleanupWikiOnKnowledgeDelete so that any wiki_ingest task still in
@@ -353,6 +368,23 @@ func (s *wikiIngestService) requeueFailedOps(ctx context.Context, payload WikiIn
if s.redisClient != nil {
pendingKey := wikiPendingKeyPrefix + payload.KnowledgeBaseID
for _, op := range ops {
// Increment failure counter; drop the op permanently once it
// exceeds wikiMaxFailRetries to prevent unbounded queue growth
// caused by persistent LLM timeouts or extraction errors.
failKey := wikiFailCountKeyPrefix + payload.KnowledgeBaseID + ":" + op.KnowledgeID
count, err := s.redisClient.Incr(ctx, failKey).Result()
if err != nil {
logger.Warnf(ctx, "wiki ingest: failed to increment fail count for %s: %v", op.KnowledgeID, err)
// Fall through and requeue anyway — better to retry than silently drop
} else {
// Refresh TTL on every update so the key doesn't outlast its usefulness
s.redisClient.Expire(ctx, failKey, wikiPendingTTL)
if count > wikiMaxFailRetries {
logger.Warnf(ctx, "wiki ingest: dropping op %s (%s) after %d failures (limit %d)", op.KnowledgeID, op.DocTitle, count, wikiMaxFailRetries)
continue
}
}
data, err := json.Marshal(op)
if err != nil {
logger.Warnf(ctx, "wiki ingest: failed to marshal op for requeue: %v", err)
@@ -362,7 +394,7 @@ func (s *wikiIngestService) requeueFailedOps(ctx context.Context, payload WikiIn
logger.Warnf(ctx, "wiki ingest: failed to requeue op %s: %v", op.KnowledgeID, err)
continue
}
logger.Infof(ctx, "wiki ingest: re-queued failed op %s (%s) for retry", op.KnowledgeID, op.DocTitle)
logger.Infof(ctx, "wiki ingest: re-queued failed op %s (%s) for retry (attempt %d/%d)", op.KnowledgeID, op.DocTitle, count, wikiMaxFailRetries)
}
return
}
@@ -345,6 +345,13 @@ func (s *wikiIngestService) ProcessWikiIngest(ctx context.Context, t *asynq.Task
slugUpdates[u.Slug] = append(slugUpdates[u.Slug], u)
}
mapMu.Unlock()
// Reset the per-document failure counter on success so transient
// errors don't permanently exhaust the retry budget.
if s.redisClient != nil {
failKey := wikiFailCountKeyPrefix + payload.KnowledgeBaseID + ":" + op.KnowledgeID
s.redisClient.Del(mapCtx, failKey)
}
}
return nil
})