Staging instance, all changes can be removed at any time

Skip to content

Make winery packer and rw-shard-mapper self-heal after failures

Nicolas Dandrimont requested to merge olasd/swh-objstorage:mr/winery-failures-resilience into master

(Stacked on top of !183 (merged), only 3 commits included)

Currently, winery locks shards to perform state changes (FULL -[via PACKING]-> PACKED, PACKED -[via CLEANING]-> READONLY).

When the operation that locked the shard fails, the state change isn't reverted, so the locked shard stays lingering.

In most cases, if the failure is caught by the packer or the rw shard cleaner, the shard state can be reverted to the original status before exiting. This will reduce the amount of manual operations we'll need to perform.

Merge request reports