Repository navigation
Reconcile-rendered re-picks truncated-tree repos every 6 hours #280
Description
The reconcile-rendered selection SQL lacks the tree_truncated_at guard the sync-candidate sweeps carry, so every 6-hour fire re-enqueues a render job for posthog/posthog whose tree always truncates, producing 4 guaranteed failed jobs on 2026-09-22 (00:25Z, 06:30Z, 12:25Z, 18:25Z) and a 38-attempt stale-reservation retry loop.
Target: layers/registry/server/tasks/reconcile-rendered.ts
Verify with: 24h after deploy: pnpm exec wrangler d1 execute skilld-db --remote --json --command "SELECT count(*) AS refails FROM failed_jobs WHERE exception = 'tree_truncated' AND failed_at >= strftime('%s','now') - 86400" returns 0 across four 6-hour fires
Estimated to change 1 file.
The daily-checkin routine opened this issue automatically on its 2026-09-22T21:40:00.000Z run. It is not Harlan's own report. Close it to reject the proposal, and the reason you give stops it being offered again.
Activity
harlan-github-agent commentedon Sep 22, 2026
🤖 ISSUE TRIAGE
r.broken_since IS NULL, while both sync-candidate sweeps (layers/registry/server/utils/sync-candidates.ts lines 32 and 57) add `AND r.tree_truncatedAND r.tree_truncated_at IS NULLto the candidate SELECT's WHERE clause beside the existingr.broken_since IS NULLguard, mirroring sync-candidates.ts; then after deploy confirm the issue's wrangler failed_jobs query returns 0 across four 6-hour fires.