Logflare / logflare / 35004752345-1

83%

main: 83%

Build Type

Pull #3991

github

Committed by Baishan

Commit Message

Scope spool write-health tracking to :disk and :upload independently

RotatingWal's local fsync succeeds far more often than a rotation
triggers an actual upload, so the previous single unscoped health
counter meant a healthy local disk silently cleared failures from the
upload path in :wal mode — Health could never reflect a persistently
failing GCS/queue upload. Cloud.commit/4 now reports to an :upload
scope and RotatingWal.commit/4 to a :disk scope, and the ingest gate
checks whichever scope actually backstops the active buffer mode
(:disk for :wal, since local WAL buffering is the whole point of that
mode; :upload for :mem, which has no local buffer at all).

Known remaining gap, left as-is for now: once a mode's gating scope
goes unhealthy, ingest stops calling into the spool entirely, so
nothing can ever report recovery again short of a restart. Same
behavior as before this change, just scoped correctly now.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Pull Request Pull Request #3991: Replace spool Partition/Committer stack with durable_buffer

Coverage Stats

270 of 295 new or added lines in 16 files covered. (91.53%)

2 existing lines in 1 file now uncovered.

15105 of 18215 relevant lines covered (82.93%)

1778.88 hits per line