qdrant

mirror of https://github.com/qdrant/qdrant.git synced 2026-07-31 15:11:35 -05:00

Files

Arnaud Gourlay df6f7ba8f0 fix: durably flush optimized segments before dropping source data

Optimization moves points (merge/dedup/CoW) without recording flush
dependencies, so a segment's persistent_version can advance past data
that isn't on disk yet. The WAL acknowledgement (get_max_persisted_version)
trusts persistent_version and can truncate the WAL past those ops; once the
source segments are dropped, a reload finds neither the truncated WAL entries
nor the data in a durable segment, permanently losing points.

Force a durable flush of all segments at the end of finish_optimization,
before the source segments' data is removed, so persisted state matches
reality before the WAL can be acknowledged past it.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

2026-06-01 15:16:03 +02:00

api

Strict mode: add max_disk_usage_percent (#9212 )

2026-06-01 11:32:25 +02:00

bm25

Switch murmur3 dep to upstream murmur3_32 crate (#9193 )

2026-05-27 13:31:50 +02:00

collection

Attempt to fix flaky test_hw_metrics_cancellation test (#9233 )

2026-06-01 13:48:24 +02:00

common

Strict mode: add max_disk_usage_percent (#9212 )