[
https://issues.apache.org/jira/browse/HIVE-22985?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17054924#comment-17054924
]
Peter Vary commented on HIVE-22985:
-----------------------------------
[~klcopp]: I am not a big fan of sending an extra HMS call just to make sure
that the transaction is aborted, or not. Shouldn't we just "unset" the
compactorTxnId when aborted?
Thanks,
Peter
> Failed compaction always throws TxnAbortedException
> ---------------------------------------------------
>
> Key: HIVE-22985
> URL: https://issues.apache.org/jira/browse/HIVE-22985
> Project: Hive
> Issue Type: Bug
> Reporter: Karen Coppage
> Assignee: Karen Coppage
> Priority: Major
> Labels: compaction
> Attachments: HIVE-22985.01.patch
>
>
> If compaction fails, its txn is aborted, however Worker attempts to commit it
> again in a finally statement. This results in a TxnAbortedException [1]
> thrown from TxnHandler#commitTxn
> We need to add a check and only try to commit at the end if the txn is not
> aborted.(TxnHandler#commitTxn does nothing if txn is already committed.)
> [1]
> {code:java}
> ERROR org.apache.hadoop.hive.metastore.RetryingHMSHandler -
> TxnAbortedException(message:Transaction txnid:16 already aborted)
> at
> org.apache.hadoop.hive.metastore.txn.TxnHandler.raiseTxnUnexpectedState(TxnHandler.java:4843)
> at
> org.apache.hadoop.hive.metastore.txn.TxnHandler.commitTxn(TxnHandler.java:1141)
> at
> org.apache.hadoop.hive.metastore.HiveMetaStore$HMSHandler.commit_txn(HiveMetaStore.java:8101)
> ...
> at
> org.apache.hadoop.hive.ql.txn.compactor.Worker.commitTxn(Worker.java:291)
> at org.apache.hadoop.hive.ql.txn.compactor.Worker.run(Worker.java:269)
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
