Commit graph

2 commits

Author SHA1 Message Date
beads/crew/lizzy
b378c1a983 fix: avoid Dolt joinIter hangs in GetReadyWork blocker computation
Three changes to improve bd ready performance on databases with many wisps:

1. Replace correlated NOT EXISTS subquery with JOIN (deferred parent
   exclusion, GH#1190) with pre-computed ID set using separate
   single-table queries — consistent with the pattern used by
   computeBlockedIDs to avoid Dolt joinIter panics.

2. Add includeWisps parameter to computeBlockedIDs. When false (default
   for bd ready without --include-ephemeral), skip scanning wisps and
   wisp_dependencies tables entirely. Verified zero cross-table blocking
   deps exist (issues only blocked by issues, wisps only by wisps).
   Reduces data scanned from ~50K rows to ~500 rows.

3. Cache-aware: a full (wisps-included) cached result satisfies
   issues-only requests; invalidation clears the new flag.

Root cause of bd ready timeouts: the hq database has 18K+ wisps and
33K+ wisp_dependencies. computeBlockedIDs was scanning all of them
even when bd ready only displays non-ephemeral issues.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-02 14:36:35 -08:00
jane
af4caf0573 test: add tests for 5xx retry and delete cache invalidation
- TestServerErrorRetry: verifies 500/502/503/504 are retried with backoff
- TestNonRetriableError: verifies 403 is NOT retried
- TestDeleteIssue_InvalidatesBlockedCache: verifies single delete clears cache
- TestDeleteIssues_InvalidatesBlockedCache: verifies batch delete clears cache
- Update TestFetchIssuesSince_Error with short timeout (500s now retried)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-01 09:59:20 -08:00