-
v3.1.0137e2241 · ·
v3.1.0 - support naming objstorage cofigurations - add statsd probes - replace the ObjStorageFilter scaffolding by a true ObjStorage proxy - rename ReadOnlyObjStorage and NonIterableObjStorage exceptions as ReadOnlyObjStorageError and NonIterableObjStorageError - make the ReadOnlyObjStorageProxy raise a HTTPReadOnlyObjStorage instead of silently no nothing on mutating endpoints.
-
v3.0.0229c586a · ·
Release swh.objstorage 3.0.0 BREAKING CHANGES: Winery now uses a single database. All shard databases are merged as tables of the main database with the `shard_` prefix. The winery database schema is now managed with the regular `swh db` command-line interface. To bootstrap it, use `swh db init objstorage.backends.winery --dbname <dsn>` after creating an empty database. Winery deployment with gunicorn requires using a `worker_int` hook to ensure proper release of shared resources. Make sure to run gunicorn with `--config python:swh.objstorage.backends.winery.gunicorn` or to replicate the included hook. OTHER CHANGES: - Release lock on shards if packing or cleaning fails - Implement error-recovery for shard packing (by zeroing shards out) - Add support for extra options passed to `rbd device map` - Add an idle timeout to release unwritten read-write shards to the pool
-
v2.9.01ef64997 · ·
v2.9.0 Changelog: * 2024-03-15 winery tests: introduce USE_CEPH environment variable * 2024-03-07 winery: try to avoid some race conditions in RO shard mapping * 2024-02-29 winery benchmark: add an explicit Stats worker * 2024-03-06 winery benchmark: run a single rw shard cleaner * 2024-03-05 winery benchmark: don't wait for shards indefinitely in packer process * 2024-03-01 winery benchmark: make object fetching less expensive * 2024-02-29 winery benchmark: introduce time_remaining argument for work * 2024-03-05 winery: decrease verbosity of rbd commands * 2024-03-05 winery: increase verbosity of remapping a shard read-write * 2024-03-05 winery: bump all copyright years to 2024 * 2024-03-07 winery: manage the lifecycle of ConnectionPools better * 2024-03-05 winery: Only run database administration operations once per process * 2024-03-05 winery: Add postgresql connection pooling * 2024-03-05 winery: remove confusing cursor passing logic in object deletion * 2024-03-04 winery: add support for the application_name PostgreSQL setting * 2024-03-04 winery: More explicit uninit operations for shards and throttler * 2024-03-01 winery: add index for per-shard statistics * 2024-03-01 winery testing: rename default erasure code profile * 2024-03-01 winery testing: only warn on EC profile mismatch * 2024-03-01 winery: use COPY to do bulk data reads * 2024-03-01 winery: migrate to psycopg3 * 2024-02-29 winery stats: use time.time instead of monotonic for line headers * 2024-02-29 winery stats: ensure stats output csv can be reused if the pid is reused * 2024-02-29 winery: Disable autovacuum on shard object tables * 2024-03-15 tests: Migrate away from unittest to full pytest *