[
https://issues.apache.org/jira/browse/KUDU-3789?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18096878#comment-18096878
]
ASF subversion and git services commented on KUDU-3789:
-------------------------------------------------------
Commit cf54a3ecc5e3bc18289983c2d599f3574db0fb28 in kudu's branch
refs/heads/master from Gabriella Lotz
[ https://gitbox.apache.org/repos/asf?p=kudu.git;h=cf54a3ecc ]
KUDU-3789 auto_rebalancer: reliably clear replace marker after failed moves
When a rebalancing move fails, we clear the source replica's replace
marker so the master stops trying to replace it. The problem is we only
tried once, and right after a failed move the leader is usually busy with
other config changes (promoting the new NON_VOTER, stepping down, a
leader transfer), so that single attempt often gets rejected and the
marker is left set forever.
Now we retry instead of giving up: a couple of quick inline attempts, and
if those still don't land, the move is queued and retried on every
rebalancer loop iteration, even when auto-rebalancing is disabled, since
the markers still need to converge. NotFound and InvalidArgument are
treated as already-cleared.
Also reset the per-round move counters each iteration so tests don't read
a stale count after a skipped round, and update
TestRemoveReplaceFlagIfMoveFails to only check the leader's view of the
config and to tolerate slower TSAN timing.
Change-Id: Ia47698207612252bca3ec70fc090a37bd3f87809
Reviewed-on: http://gerrit.cloudera.org:8080/24525
Tested-by: Alexey Serbin <[email protected]>
Reviewed-by: Alexey Serbin <[email protected]>
> Investigate and fix flaky tests
> -------------------------------
>
> Key: KUDU-3789
> URL: https://issues.apache.org/jira/browse/KUDU-3789
> Project: Kudu
> Issue Type: Sub-task
> Reporter: Gabriella Lotz
> Assignee: Gabriella Lotz
> Priority: Major
>
> # In auto_rebalancer-test.cc
> # In auto_leader_rebalancer-test.cc
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
