[
https://issues.apache.org/jira/browse/HUDI-6353?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Surya Prasanna Yalla updated HUDI-6353:
---------------------------------------
Description:
Couple of enhancements we can make to Multiwriter conflict resolution
strategies.
* Similar to how git rebase works, writers when creating a requested commit
should track the baseInstant on which they are executing, this will be the last
completed commit. Note, last completed commit can have commit timestamp less
than other completed commits in the timeline. Here, baseInstantTime is based on
the state transition times which is the last commit written to the timeline.
Conflict resolution strategies should happen based on this commit time. We can
also avoid the race condition by acquiring lock during requested file creation.
* Similar to IngestionPrimaryWriterBasedConflictResolutionStrategy class, we
should use state transition times to find the conflicts. By using state
transition times the conflict resolution works similar to git rebase.
was:
Couple of enhancements we can make to Multiwriter conflict resolution
strategies.
* Similar to how git rebase works, writers when creating a requested commit
should track the baseInstant on which they are executing, this will be the last
completed commit. Note, last completed commit can have commit timestamp less
than other completed commits in the timeline. Here, baseInstantTime is based on
the state transition times which is the last commit written to the timeline.
Conflict resolution strategies should happen based on this commit time.
* Similar to IngestionPrimaryWriterBasedConflictResolutionStrategy class, we
should use state transition times to find the conflicts. By using state
transition times the conflict resolution works similar to git rebase.
> Use state transition times for
> SimpleConcurrentFileWritesConflictResolutionStrategy and other enhancements.
> -----------------------------------------------------------------------------------------------------------
>
> Key: HUDI-6353
> URL: https://issues.apache.org/jira/browse/HUDI-6353
> Project: Apache Hudi
> Issue Type: Improvement
> Reporter: Surya Prasanna Yalla
> Priority: Major
>
> Couple of enhancements we can make to Multiwriter conflict resolution
> strategies.
> * Similar to how git rebase works, writers when creating a requested commit
> should track the baseInstant on which they are executing, this will be the
> last completed commit. Note, last completed commit can have commit timestamp
> less than other completed commits in the timeline. Here, baseInstantTime is
> based on the state transition times which is the last commit written to the
> timeline. Conflict resolution strategies should happen based on this commit
> time. We can also avoid the race condition by acquiring lock during requested
> file creation.
> * Similar to IngestionPrimaryWriterBasedConflictResolutionStrategy class, we
> should use state transition times to find the conflicts. By using state
> transition times the conflict resolution works similar to git rebase.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
