[
https://issues.apache.org/jira/browse/HIVE-12634?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15260764#comment-15260764
]
Eugene Koifman commented on HIVE-12634:
---------------------------------------
[~wzheng]
1. TxnHandler.abortTxns() logic is wrong. If it ends up in _if (updateCnt <
txnids.size()) {_ then it will never execute _delete from HIVE_LOCKS..._ so it
will leak locks. This is OK in the context where the caller will rollback in
this case but not for the new operation you added. Perhaps you can pass in a
flag to this method indicating whether it should be 'strict' or 'best effort'
2. TxnHandler.abortTransactions() uses _lockInternal()_ - this is not needed.
Only if it was doing "Select for update" to synchronize. There are some
details in JavaDoc
3. I think it would be useful to make dbtxnmgr_abort.q provide a long list of
txn IDs. Something exceeding hive.direct.sql.max.elements.in.clause.
4. (since you are making another patch) AbortTxnDesc - class level JavaDoc
reads "compactions"...
> Add command to kill an ACID transacton
> --------------------------------------
>
> Key: HIVE-12634
> URL: https://issues.apache.org/jira/browse/HIVE-12634
> Project: Hive
> Issue Type: New Feature
> Components: Transactions
> Affects Versions: 1.0.0
> Reporter: Eugene Koifman
> Assignee: Wei Zheng
> Labels: TODOC2.1
> Attachments: HIVE-12634.1.patch, HIVE-12634.2.patch,
> HIVE-12634.3.patch, HIVE-12634.4.patch
>
>
> Should add a CLI command to abort a (runaway) transaction.
> This should clean up all state related to this txn.
> The initiator of this (if still alive) will get an error trying to
> heartbeat/commit, i.e. will become aware that the txn is dead.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
