[
https://issues.apache.org/jira/browse/IMPALA-12304?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17752198#comment-17752198
]
ASF subversion and git services commented on IMPALA-12304:
----------------------------------------------------------
Commit 3a9b52faf18ab76724aaeb0bc3ea35e379c3c044 in impala's branch
refs/heads/master from wzhou-code
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=3a9b52faf ]
IMPALA-12340: Fix flaky test_two_catalogd_with_force_active
The issue could be re-produced by repeatedly running the test case for
about 50 times in Jenkins or on local machine.
The issue was introduced by IMPALA-12304, for which we tried not to
wake up the thread for update_catalogd RPC if there is no change for
elected active catalogd. Since we cannot hold mutex when calling
SendUpdateCatalogdNotification(), it may cause the notification to the
condition variable not been processed. To fix the issue, revert the
optimization added in IMPALA-12304, and add sleep in
test_two_catalogd_with_force_active.
Testing:
- Repeatedly ran test_two_catalogd_with_force_active for 8000 times on
local machine without failure, ran the test for 1000 times in Jenkins
without failure.
Repeatedly ran all test cases in test_catalogd_ha.py for 21 hours in
Jenkins without failure.
- Passed core test.
Change-Id: Ifc213422a5a3360d07a22046e1f46fdf0be1d2fd
Reviewed-on: http://gerrit.cloudera.org:8080/20323
Reviewed-by: Impala Public Jenkins <[email protected]>
Tested-by: Impala Public Jenkins <[email protected]>
> Fix the sequence number issue for update_catalogd RPC when startestore is
> restarted
> -----------------------------------------------------------------------------------
>
> Key: IMPALA-12304
> URL: https://issues.apache.org/jira/browse/IMPALA-12304
> Project: IMPALA
> Issue Type: Improvement
> Components: Backend
> Affects Versions: Impala 4.3.0
> Reporter: Wenzhe Zhou
> Assignee: Wenzhe Zhou
> Priority: Major
> Fix For: Impala 4.3.0
>
>
> Subscriber will re-register to statestore when statestore is restarted. When
> the statestore is restarted, its sending sequence number for update_catalogd
> RPC is reset. The subscribers need to reset their last receiving sequence
> number of update_catalogd RPC when they successfully re-register with
> statestore. Otherwise, they may miss the some RPCs after statestore is
> restarted.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
