You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hive.apache.org by "Alan Gates (JIRA)" <ji...@apache.org> on 2015/11/17 19:12:11 UTC
[jira] [Commented] (HIVE-12421) Streaming API
TransactionBatch.beginNextTransaction() does not wait for locks
[ https://issues.apache.org/jira/browse/HIVE-12421?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15009159#comment-15009159 ]
Alan Gates commented on HIVE-12421:
-----------------------------------
[~roshan_naik] can comment further, but I think the thinking was some writers have no ability to buffer their inbound data, so blocking on a lock could force them to drop data. We should change this to have the writer pass a maximum wait time. If this code fails to obtain a lock in the allowed time then it can throw an exception that makes clear what happened and let the writer decide what to do from there.
> Streaming API TransactionBatch.beginNextTransaction() does not wait for locks
> -----------------------------------------------------------------------------
>
> Key: HIVE-12421
> URL: https://issues.apache.org/jira/browse/HIVE-12421
> Project: Hive
> Issue Type: Bug
> Components: HCatalog, Transactions
> Affects Versions: 0.14.0
> Reporter: Eugene Koifman
> Assignee: Eugene Koifman
>
> TransactionBatchImpl.beginNextTransactionImpl() has
> {noformat}
> LockResponse res = msClient.lock(lockRequest);
> if (res.getState() != LockState.ACQUIRED) {
> throw new TransactionError("Unable to acquire lock on " + endPt);
> }
> {noformat}
> This means that if there are any competing locks already take, this will throw an Exception to client. This doesn't seem like the right behavior. It should block.
> We could also add TransactionBatch.beginNextTransaction(long timeoutMs) to give the client more control.
> cc [~alangates] [~sriharsha]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)