Rongsong Shen
d26ff8870d
The buglog was one 27K-line JSON file whose fields covered only symptom,
root cause and fix, as free text: 2% of entries named the code version a
bug was found on, 12% the build profile, 36% a way to reproduce, and 58% of
the file was OpenWolf hook auto-detections ("Missing guard clause",
"Function not marked async" on .md files).
- .claude/skills/buglog/bug-record.md: the record standard. Eight
sections (Summary as the title, Description, Environment, Reproduction;
Root cause, Fix, Verification, References once fixed), each with the
content it must contain -- e.g. Environment needs the OS, the libc the
failing binary loads (from its ELF interpreter: glibc 2.42 in the nix
shell vs 2.40 on the host), toolchain, build profile and commit.
- .wolf/bugs/<id>.md: 500 records migrated from buglog.json; every field
verified to round-trip. They carry legacy: true and an explicit "not
recorded" where the old log had no such field; component derived from
file paths, severity from keywords (else unknown), fix_commits from
hashes in status_note/fix that git resolves (91 of 463 fixed).
- .wolf/buglog-archive.json: the 710 hook auto-detections, verbatim.
- .wolf/buglog.json is now an index generated from the records, keeping
the fields OpenWolf's hooks and `openwolf bug search` read.
- scripts/buglog.py rewritten: new (template + Environment collected from
the machine and --binary), check, status (refuses a status the record
doesn't support; fixed needs Root cause, Fix, Verification and
--commit hashes git resolves), note, recur, set, show, list, triage,
env, index. 29 unittest cases in scripts/tests/test_buglog.py.
- scripts/git-hooks/pre-commit runs `check --staged`: incomplete staged
records, a legacy record hand-changed to fixed without the fix-time
sections, a new record claiming legacy: true, or a stale index refuse
the commit.
- scripts/gitea-sync.py reads the records: issue body = the record's
sections plus linked fix commits (origin remote, GITEA_COMMIT_REPO to
override), and a body hash so an issue's text is refreshed when the
record changes.
- bug-1866 rewritten to the full standard as the worked example.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
3.6 KiB
| id | status | status_updated | status_note | component | severity | timestamp | last_seen | occurrences | files | tags | related | legacy | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| bug-1564 | fixed | 2026-09-24 | controller | unknown | 2026-08-17T07:15:00.000Z | 2026-09-19T09:00:00Z | 1 |
|
|
|
true |
bug-1564: controller_unit_test: test_durable_majority_commit_end_to_end FAIL: Expected 0 Was -110…
Description
controller_unit_test: test_durable_majority_commit_end_to_end FAIL: Expected 0 Was -110 (commit_submit(WAIT) must return 0 after durable-majority) -- deterministic 3/3 direct runs of the binary, not a flake. Surfaced while live-testing an unrelated /xce-deploy-ha enhancement, right after rebuilding controller on top of bug-1563's fix.
Environment
Not recorded (migrated from buglog.json).
Reproduction
Not recorded (migrated from buglog.json).
Root cause
persist_flush() (the synchronous durability point tests and clean-shutdown use in place of waiting for the background flusher) advances durable_lsn directly under p->lock but never calls p->on_durable_advance -- only fsync_worker's own successful flusher_flush_once() call fires that callback (persist.c, 'after > before' gate). Any WAIT-blocked commit caller (commit_engine's COMMIT_MODE_WAIT) that relies on on_durable_advance to be re-evaluated is therefore never woken by a persist_flush()-driven advance -- only by fsync_worker's own next successful cycle. This gap was LATENT and never triggered before bug-1563: pre-1563, fsync_worker's un-throttled retry loop (the livelock bug-1563 fixed) spun continuously while work was pending, so it happened to notice any state change (e.g. a test clearing test_force_fsync_fail) and independently re-attempt+succeed+fire the callback within microseconds, completely masking that persist_flush's own advance was silently un-observed. bug-1563's fix (correct: fsync_worker now backs off up to wal_sync_ms between failed attempts instead of spinning) removed that accidental fast-repoll cover, so a persist_flush()-only advance during a slow-cadence window (make_durability_persist uses wal_sync_ms=5000 deliberately, 'durability comes from the explicit flush, not the timer') now genuinely times out the 2000ms-budget WAIT with nothing left to wake it for the remaining ~4.7s fsync_worker was still asleep.
Fix
FIXED (TDD RED->GREEN). Added test_persist_flush_fires_durable_advance_callback (test_commit_durability.c, next to test_durable_majority_commit_end_to_end): registers a counting callback via persist_set_durable_cb, appends an entry, asserts persist_flush() fires it exactly once on a genuine advance and not again on a no-op flush. RED confirmed against the pre-fix persist_flush (0 fires). Fix: persist_flush() now fires p->on_durable_advance(p->durable_cb_ctx) after releasing p->lock, gated on advanced==true (only when durable_lsn actually moved) -- identical gate and identical lock-ordering discipline (strictly outside p->lock, so the callback's own ce->lock acquisition has no p->lock held underneath it) to fsync_worker's own existing callback-firing code. GREEN: both test_persist_flush_fires_durable_advance_callback and the originally-broken test_durable_majority_commit_end_to_end pass, confirmed deterministic across 3 direct repeat runs of controller_unit_test. Full suite: 1104 tests, 2 failures (both pre-existing, unrelated overlayfs fdatasync sandbox flakiness in test_config_repl.c, unchanged from before this fix).
Verification
Not recorded (migrated from buglog.json).
References
- Related: bug-1563