[
https://issues.apache.org/jira/browse/KUDU-3322?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17609151#comment-17609151
]
ASF subversion and git services commented on KUDU-3322:
-------------------------------------------------------
Commit e39411e086eeff6d77bcc640aaac34fdf8ca16df in kudu's branch
refs/heads/master from Abhishek Chennaka
[ https://gitbox.apache.org/repos/asf?p=kudu.git;h=e39411e08 ]
[master] KUDU-3322 / KUDU-3319 HMS event logging
It would be useful to have additional logging in cases where HMS
notification events have to be analyzed from Kudu side. The Jiras
have more context on the reasoning behind this but this patch adds
* an INFO level log which shows the last processed HMS event id during
startup and
* an ERROR log if the latest event id from HMS is less than
the last seen event id from the master. Was hesitant on crashing the
master and chose to log an ERROR as in the case an out of order event
id received from HMS might lead to unnecessary crash of the Kudu
master.
Change-Id: I242c9cedf170ed621867b905f1cb1875347d5887
Reviewed-on: http://gerrit.cloudera.org:8080/18866
Tested-by: Alexey Serbin <ale...@apache.org>
Reviewed-by: Alexey Serbin <ale...@apache.org>
> Handle to situation of last seen event ID in Kudu master being more than the
> latest event ID from HMS
> -----------------------------------------------------------------------------------------------------
>
> Key: KUDU-3322
> URL: https://issues.apache.org/jira/browse/KUDU-3322
> Project: Kudu
> Issue Type: Improvement
> Reporter: Abhishek
> Priority: Minor
>
> Came across a scenario where HMS database was lost and restored from an old
> backup which reset the latest event ID (monotonically increasing integer) in
> HMS to a lower value than what it should be.
> Since Kudu master has a last seen event ID greater than the one in HMS
> currently, it could not process any new events generated. For example, Kudu
> table deletion was not happening as the Kudu master expects an event ID which
> is higher than the one it has last seen but the event ID in HMS for the table
> deletion is less than the one in the Kudu master.
> This also causes discrepancy between the metadata in HMS and Kudu masters. It
> would be better if the Kudu master upon startup does the comparison of the
> last seen event ID and latest event ID in HMS and crash if the one in HMS is
> lower with a helpful message/clarifying question like:
> {code:java}
> Found the last seen event ID in the local Kudu master to be greater than the
> latest event ID in HMS. Was there any backup or restore done on HMS
> recently?{code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
