Rollover During Partial Snapshot of a Datastream Should not Exclude the DS from the Snapshot #68536
Labels
:Data Management/Data streams
Data streams and their lifecycles
:Distributed Coordination/Snapshot/Restore
Anything directly related to the `_snapshot/*` APIs
>enhancement
Team:Data Management
Meta label for data/management team
Team:Distributed (Obsolete)
Meta label for distributed team (obsolete). Replaced by Distributed Indexing/Coordination.
Background: #68258 fixed the issue that rolling over a datastream during a snapshot would break the snapshot finalization because the snapshot would not include the newest index of the datastream during finalization but also would not have snapshotted the datastream metadata before rollover, leaving the cluster without a consistent pair of metadata and indices.
The current fix to this situation was to leave datastreams out of a partial snapshot if they were rolled over during the snapshot completely.
We discussed this in the core-features meeting and we decided that this behavior needs to be improved given that data-streams will be used more and more and start to work with time-based rollovers, increasing the chance of this type of collision between snapshot and rollover and thus the chance of never getting a snapshot for a certain datastream if partial snapshots are used.
I see a fairly direct fix to this and I'll open a PR for it.
The text was updated successfully, but these errors were encountered: