[ 
https://issues.apache.org/jira/browse/KUDU-3789?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18101924#comment-18101924
 ] 

ASF subversion and git services commented on KUDU-3789:
-------------------------------------------------------

Commit 61a882f8ffd0729c21874a515711f03a1ce30df8 in kudu's branch 
refs/heads/master from Gabriella Lotz
[ https://gitbox.apache.org/repos/asf?p=kudu.git;h=61a882f8f ]

KUDU-3789 auto_rebalancer: de-flake TSAN tests

A handful of tests in auto_rebalancer-test were flaky under TSAN
dist-test. These are test-only fixes; there are no production changes.

- OnlyLeaderDoesAutoRebalancing / NextLeaderResumesAutoRebalancing:
instead of asserting non-leader masters have exactly 0 iterations,
check that only the current leader's counts grow over a stable
leadership window. The initial election can flicker, so a brief
transient leader could have a non-zero count.

- NoRebalancingIfReplicasRecovering / TestDeletedTables: wait until the
master actually sees the killed tserver as dead
(WaitForLiveTServerCount) rather than sleeping a fixed interval, set
the unresponsive timeout only after the tables exist, and use RF=1 for
the extra table.

- PreferFollowerRebalancingTest / TestMaxMovesPerServer: give
CheckSomeMovesScheduled more time under TSAN (and a lighter workload
for the former), so the first scheduling round isn't cut off.

- ExecuteMovesCASRejectionDropsMoveGracefully: retry until the
CAS-rejection race we want to exercise actually happens, since a
heartbeat can refresh the catalog's opid before ExecuteMoves runs.

- RebalancerMetricsFollowerMoves: relax the leader-move count from ==0
to <= follower moves. prefer_follower is only a preference, so the
rebalancer can still fall back to a leader move when the source has no
follower for the table.

Change-Id: I223f83be1db6f843f24a7bf95e31f3e3f77dcfe7
Reviewed-on: http://gerrit.cloudera.org:8080/24584
Reviewed-by: Marton Greber <[email protected]>
Reviewed-by: Alexey Serbin <[email protected]>
Tested-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)

Reply via email to