Event Sourcing might halt if CommitLogStore is not available for a long period

[xirc]

RaftActor doesn't track the progress of the event-sourcing feature. If CommitLogStore is not available for a long period (due to like journal failures), the Event Sourcing feature will halt, and it won't be recovered automatically. This halt seldom happens since CommitLogStore retries to save the committed event. Users can recover this halt by restarting each node (rolling update is possible) unless compaction deletes such events.

If compaction deletes committed entries (but they are not persisted to CommitLogStore), it is difficult to recover since RaftActor doesn't know which committed entry to persist to CommitLogStore.

Related code

• RaftActor
• CommitLogStore
• CommitLogStoreActor

[xirc]

Lerna team will address this issue by the following:

• RaftActor tracks the progress of Event Sourcing (CommitLogStore).
• Compaction doesn't delete entries that are not persisted to CommitLogStore yet.

[xirc]

#142 addresses some additional challenges.