[
https://issues.apache.org/jira/browse/PHOENIX-5311?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16853219#comment-16853219
]
Thomas D'Silva commented on PHOENIX-5311:
-----------------------------------------
Dropping tables can be an expensive operation, thats why the tests keep track
of how many tables are created and then just teardown and start a new
minicluster.
Are you running the integration tests with a real cluster (not using the
minicluster)? You could modify freeResourcesIfBeyondThreshold to drop all the
hbase tables based on a config flag.
> Integration tests leak tables when running on distributed cluster
> -----------------------------------------------------------------
>
> Key: PHOENIX-5311
> URL: https://issues.apache.org/jira/browse/PHOENIX-5311
> Project: Phoenix
> Issue Type: Bug
> Reporter: István Tóth
> Priority: Major
>
> When integration test suite is run via End2EndTestDriver on a distributed
> cluster, most tests do not clean up their tables, leaving thousands of tables
> on the cluster, and exhausting RegionServer memory.
> There are actually three problems:
> * The BaseTest.freeResourcesIfBeyondThreshold() method is called after most
> tests, and it restarts the MiniCluster, thus freeing resources, but it has no
> effect when running on a distributed cluster.
> * The TestDriver sets phoenix.schema.dropMetaData to false by default, so
> even if the Phoenix tables are dropped, the HBASE tables are not, so the
> table leak remains.
> * The phoenix.schema.dropMetaData setting cannot be easily overridden
> because of PHOENIX-5310
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
