[
https://issues.apache.org/jira/browse/HIVE-13458?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15246381#comment-15246381
]
Eugene Koifman commented on HIVE-13458:
---------------------------------------
[~wzheng]
1. DbTxnManager.stopHeartbeater():
- it would be useful for LOG.warn("Heartbeat task cannot be cancelled for
unknown reason"); to include query id (QueryPlan.getQueryId())
- e.printStackTrace(); - where will this print to?
- Below will add to the logs but I'm not sure where the value add is.
{noformat} if (heartbeatTask.isCancelled() || heartbeatTask.isDone()) {
LOG.info("Stopped " + Heartbeater.class.getName());
} {noformat}
Perhaps lower to DEBUG level or alternatively include QueryID and the "now -
startTime". I'd vote for DEBUG.
2. TezJobMonitor
- why wrap the LockException message in IOException?
- could
{noformat}
if (ctx != null && ctx.getHeartbeater() != null &&
243 ctx.getHeartbeater().getLockException() != null) {
244 throw new IOException("Need to abort execution due to
LockException: " +
245 ctx.getHeartbeater().getLockException().getMessage());
246 }
{noformat}
be wrapped in a single method "Context.checkAborted() throws Exception" for
example?
Same for HadoopJobExecHelper.
> Heartbeater doesn't fail query when heartbeat fails
> ---------------------------------------------------
>
> Key: HIVE-13458
> URL: https://issues.apache.org/jira/browse/HIVE-13458
> Project: Hive
> Issue Type: Bug
> Components: Transactions
> Affects Versions: 2.1.0
> Reporter: Wei Zheng
> Assignee: Wei Zheng
> Attachments: HIVE-13458.1.patch, HIVE-13458.2.patch,
> HIVE-13458.3.patch
>
>
> When a heartbeat fails to locate a lock, it should fail the current query.
> That doesn't happen, which is a bug.
> Another thing is, we need to make sure stopHeartbeat really stops the
> heartbeat, i.e. no additional heartbeat will be sent, since that will break
> the assumption and cause the query to fail.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
