[
https://issues.apache.org/jira/browse/HDDS-16758?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18124859#comment-18124859
]
Siyao Meng commented on HDDS-16758:
-----------------------------------
Another occurrence on master at d319875ec11, in the gating integration (om)
split:
{code}
testClientWithLocalLeaseEnabled -- Time elapsed: 0.021 s <<< FAILURE!
java.lang.AssertionError:
Expecting actual:
0L
to be greater than:
0L
at
org.apache.hadoop.ozone.om.TestOzoneManagerHAFollowerReadWithAllRunning.testClientWithLocalLeaseEnabled(TestOzoneManagerHAFollowerReadWithAllRunning.java:655)
{code}
- https://github.com/apache/ozone/actions/runs/37683659545/job/113009934501
This matches the analysis here. The skip over the known leader for LOCAL_LEASE
in
HadoopRpcOMFollowerReadFailoverProxyProvider#getCurrentProxy(ReadConsistency)
came in with HDDS-15089, so the failure should hit about one run in three (the
OM proxy order is shuffled). I looked over d074cf88d6 on your HDDS-16758 branch
and it looks good to me. Could you open the PR? This test runs in a gating
split, so it can block unrelated PRs.
> Intermittent failure in testClientWithLocalLeaseEnabled due to checking the
> wrong OM metrics
> --------------------------------------------------------------------------------------------
>
> Key: HDDS-16758
> URL: https://issues.apache.org/jira/browse/HDDS-16758
> Project: Apache Ozone
> Issue Type: Sub-task
> Reporter: Huang Kuan Hao
> Assignee: Huang Kuan Hao
> Priority: Major
>
> testClientWithLocalLeaseEnabled records the local-lease success count on
> getCurrentProxy() before listing volumes. LOCAL_LEASE routing skips the known
> leader, and retries may select another OM. A successful read can therefore
> leave the checked OM's count unchanged and fail the assertion.
> Record each OM's count before the request, then use getLastProxy() to check
> the count on the OM that actually served the request. Preserve normal routing
> and the local-lease assertion.
> Reproduced with a Mini HA cluster on upstream d246fff3f3: when the initial
> proxy was the known leader, its count stayed 0 while the serving follower's
> count increased to 1. The original assertion failed; the fix passed under the
> same initial condition. The complete test class (15 tests), checkstyle and
> module PMD passed.
> CI failure:
> https://github.com/rich7420/ozone/actions/runs/37608060782/job/112823872954
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
