[
https://issues.apache.org/jira/browse/KUDU-3341?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17452480#comment-17452480
]
ASF subversion and git services commented on KUDU-3341:
-------------------------------------------------------
Commit 0222c3163129b1d6c1c37b216482aa64f921c415 in kudu's branch
refs/heads/master from zhangyifan27
[ https://gitbox.apache.org/repos/asf?p=kudu.git;h=0222c31 ]
KUDU-3341: Stop retrying to DeleteTablet on wrong server
This patch improves catalog_manager's behavior when delete tablet with
a 'WRONG_SERVER_UUID' error. It's better to mark this RetryTask failed
than keep retrying to send too many requests. Because master would
always receive same error message until the "wrong uuid server" restarts
with a "correct uuid", at that time the tserver would send full tablets
report and then trigger the deletion of outdated tablets.
I also add a test that reproduces the scenario described in the JIRA.
Change-Id: Ieaa36086300bda7f958570c690b951dc090c342a
Reviewed-on: http://gerrit.cloudera.org:8080/18057
Tested-by: Kudu Jenkins
Reviewed-by: Andrew Wong <aw...@cloudera.com>
Reviewed-by: Attila Bukor <abu...@apache.org>
> Catalog Manager should stop retrying DeleteTablet when receive
> WRONG_SERVER_UUID error
> --------------------------------------------------------------------------------------
>
> Key: KUDU-3341
> URL: https://issues.apache.org/jira/browse/KUDU-3341
> Project: Kudu
> Issue Type: Improvement
> Components: master
> Reporter: YifanZhang
> Assignee: YifanZhang
> Priority: Minor
>
> Sometimes a tablet server could be shutdown because of detected disk
> failures, and this server would be re-added to the cluster with all data
> cleared.
> Replicas could be replicated afterÂ
> {{\-\-follower_unavailable_considered_failed_sec}} seconds. And then master
> send DeleteTablet RPCs to this tserver, but receive either a RPC
> failure(tserver was shutdown) or a WRONG_SERVER_UUID error(tserver started
> with a new uuid), and keep retrying to delete tablets after
> {{{}--unresponsive_ts_rpc_timeout_ms{}}}(default 1 hour).
> It's not so necessary to retry when receive WRONG_SERVER_UUID errors, because
> the server uuid could only be corrected by restarting the tablet server, at
> that time full tablet reports would sent to master and if any, outdated
> replicas could be deleted finally.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
