[
https://issues.apache.org/jira/browse/HIVE-26947?focusedWorklogId=840557&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-840557
]
ASF GitHub Bot logged work on HIVE-26947:
-----------------------------------------
Author: ASF GitHub Bot
Created on: 20/Jan/23 09:32
Start Date: 20/Jan/23 09:32
Worklog Time Spent: 10m
Work Description: veghlaci05 commented on code in PR #3955:
URL: https://github.com/apache/hive/pull/3955#discussion_r1082283569
##########
ql/src/java/org/apache/hadoop/hive/ql/txn/compactor/Worker.java:
##########
@@ -118,19 +119,23 @@ public void run() {
singleRun.cancel(true);
executor.shutdownNow();
executor = getTimeoutHandlingExecutor();
+ err = true;
} catch (ExecutionException e) {
LOG.info("Exception during executing compaction", e);
+ err = true;
} catch (InterruptedException ie) {
// do not ignore interruption requests
return;
+ } catch (Throwable t) {
+ err = true;
}
doPostLoopActions(System.currentTimeMillis() - startedAt);
// If we didn't try to launch a job it either means there was no work
to do or we got
- // here as the result of a communication failure with the DB. Either
way we want to wait
+ // here as the result of an error like communication failure with the
DB, schema failures etc. Either way we want to wait
// a bit before, otherwise we can start over the loop immediately.
- if (!launchedJob && !stop.get()) {
+ if ((!launchedJob || err) && !stop.get()) {
Thread.sleep(SLEEP_TIME);
Review Comment:
@akshat0395 I think the timeout should be increased starting from the second
failed job execution in a row. (until it reaches some defined maximum value)
However, with backoff, we will need the err flag to differentiate between
errors and cases where there were no compactions to work on.
Issue Time Tracking
-------------------
Worklog Id: (was: 840557)
Time Spent: 2h 40m (was: 2.5h)
> Hive compactor.Worker can respawn connections to HMS at extremely high
> frequency
> --------------------------------------------------------------------------------
>
> Key: HIVE-26947
> URL: https://issues.apache.org/jira/browse/HIVE-26947
> Project: Hive
> Issue Type: Bug
> Reporter: Akshat Mathur
> Assignee: Akshat Mathur
> Priority: Major
> Labels: pull-request-available
> Time Spent: 2h 40m
> Remaining Estimate: 0h
>
> After catching the exception generated by the findNextCompactionAndExecute()
> task, HS2 appears to immediately rerun the task with no delay or backoff. As
> a result there are ~3500 connection attempts from HS2 to HMS over just a 5
> second period in the HS2 log
> The compactor.Worker should wait between failed attempts and maybe do an
> exponential backoff.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
