pnowojski commented on a change in pull request #17774:
URL: https://github.com/apache/flink/pull/17774#discussion_r767889774
##########
File path:
flink-state-backends/flink-statebackend-rocksdb/src/main/java/org/apache/flink/contrib/streaming/state/snapshot/RocksIncrementalSnapshotStrategy.java
##########
@@ -258,30 +268,37 @@ private SnapshotDirectory
prepareLocalSnapshotDirectory(long checkpointId) throw
}
Review comment:
> The purpose of notifyCheckpointAborted is purely cleanup. It's called
from state backend, so it's not dead code.
In this context by dead code I meant "not having any meaningful effect". The
code would be working just as well without this `notifyCheckpointAborted()`,
right? The `materializedSstFiles` would still be cleaned up in
`notifyCheckpointComplete()`?
About my proposal of unifying `lastUploadedSstFiles` and
`materializedSstFiles` I missed that the former one has also
`StreamStateHandle`s. So to unify them, we would need to change
`materializedSstFiles` type to: `SortedMap<Long, Map<StateHandleID,
StreamStateHandle>>`. (so keeping them separate makes sense).
However I'm still confused by those variables and the inconsistency between
them. Once checkpoint is aborted, those sst files are no longer "materialized"?
But they are still considered as uploaded? At the very least naming is
confusing.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
