krasserm
commented
10 years ago That's a regression. View snapshotting has been broken with this commit. Problem was that view snapshots are always orphan snapshots as no corresponding journal topic exists, so they were always ignored and replay always started from scratch.
The current solution is now a good compromise to deal with:
- log purges done by Kafka under certain failure conditions
- ignore orphan snapshots for persistent actors
- loading view snapshots regardless whether the corresponding journal topic covers the snapshot's sequence number. Under certain Kafka failure conditions, this may lead to temporary view failures (which causes akka-persistence to re-attempt an incremental replay) until new events are written to the journal topic.
Actually, it should be the responsibility of akka-persistence to deal with these failure scenarios properly. I'll try to create a pull request for akka-persistence when I find some time.
Hope that helps for the moment. Please let me know and I'll cut another patch release.
Thanks, Martin
This might be related to my previous issue #7 but I can't be sure. It's the same setup as before. One producer, one view. The view doesn't seem to be able to snapshot correctly.
Testing in JDBC I can snapshot and recover but the kafka driver seems to give me very weird results. The producer actor seems to be fine but the view will always read from the beginning of the queue.
What could I look for to see what happening?