[
https://issues.apache.org/jira/browse/HBASE-18902?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16188061#comment-16188061
]
Ted Yu commented on HBASE-18902:
--------------------------------
At the beginning of Appy's comment on Friday, he listed 3 items. I think Using
single cluster, for this test, is feasible and should be chosen over using 3
mini cluster starts/stops (as shown in patch v4).
One benefit is reduction in test run time (42s to 22s on my Mac).
bq. Why do have flaky tests?
There are many reasons for tests to behave differently. Most are timing issues
involving, e.g. region splitting / merging / compacting / flushing / moving.
Discussing on a case by case basis would be better.
bq. the flakyness can be due to wrong functionality
Could be: if code behaves incorrectly, that would ultimately surface in some
test(s). For TestCoprocessorServiceBackwardCompatibility, since compatibility
is kept, there is not much room for flakiness.
I took a brief look at TestAdmin1 where multiple scenarios are tested:
non-existent table,
existent table but non-existent column family,
table creation,
table truncation,
table truncation preserving splits,
...
It would be very difficult to merge them considering the divergence of test
scenarios.
Toward the end of Appy's comment on Friday, seems he already looked at the test
composition from a different angle. I appreciate the summary.
I was involved in customer support case during day time of Friday - otherwise I
should have put up patch v4 earlier.
> TestCoprocessorServiceBackwardCompatibility fails in master branch
> ------------------------------------------------------------------
>
> Key: HBASE-18902
> URL: https://issues.apache.org/jira/browse/HBASE-18902
> Project: HBase
> Issue Type: Test
> Reporter: Ted Yu
> Assignee: Ted Yu
> Attachments: 18902.v1.txt, 18902.v2.txt, 18902.v3.txt, 18902.v4.txt,
> HBASE-18902.master.001.patch, HBASE-18902.master.001.patch
>
>
> Starting mini cluster in subtests without shutting down resulted in:
> {code}
> java.lang.IllegalStateException: A mini-cluster is already running
> at
> org.apache.hadoop.hbase.coprocessor.TestCoprocessorServiceBackwardCompatibility.testCoprocessorServiceLoadedByRegion(TestCoprocessorServiceBackwardCompatibility.java:109)
> {code}
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
