[
https://issues.apache.org/jira/browse/HUDI-4829?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ethan Guo updated HUDI-4829:
----------------------------
Component/s: compaction
table-service
> Rollback of log compaction should use existing pending rollback instants
> ------------------------------------------------------------------------
>
> Key: HUDI-4829
> URL: https://issues.apache.org/jira/browse/HUDI-4829
> Project: Apache Hudi
> Issue Type: Bug
> Components: compaction, table-service
> Reporter: Surya Prasanna Yalla
> Priority: Major
>
> rollbackInflightLogCompaction(inflightInstant, s -> Option.empty());
> *Comments*
> shouldn't we pass getPendingRollbackInstantFunc in the 2nd argument.
> the reason why we introduced the pending rollback is below.
> Lets say c5 commit partially failed.
> when we are looking to start a new commit, say C7, we deduct that c5 has
> failed and we trigger a rollback named RB_5.
> But lets say the process crashed mid-way for RB_5.
> Next time we restart the pipeline for say C8. again we deduct that c5 is
> partially failed and we want to rollback. but we don't want to create RB_9
> for this,but re-use RB_5 again and get it to completion.
> So, rollbackPendingCompaction or rollbackPendingClustering should always try
> to re-use rollback instants if any.
>
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
