Bug #65130
Updated by Samuel Just about 2 months ago
https://github.com/ceph/ceph/pull/56376 would have reintroduced https://tracker.ceph.com/issues/61875 as it puts the snap mapper keys back into the pg meta object. Oddly, a teuthology run on that branch which seems to have included tests with both snapshots and osd restarts did not show crashes associated with this regression and at least one case that seems like it should have exercised the relevant code passed. any osd crashes. A quick glance over PGLog.cc::FuturizedShardStoreReader doesn't show any changes, so it should have crashed in the final else branch of FuturizedShardStoreLogReader::process_entry at e.decode_with_checksum. Tasks: - Confirm that the crimson-rados suite actually combines snapshots with OSD restarts - Work out why the existing suite didn't fail the above PR - Amend the tests to cover the gap