[
https://issues.apache.org/jira/browse/HBASE-15206?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15129214#comment-15129214
]
Jonathan Hsieh commented on HBASE-15206:
----------------------------------------
For #2.
Got it. The catalogjanitor cleans up meta after splits. So I think we either
disable catalogjanitor and assert that the parent is still present (e.g. assert
that hbck didn't remove the parent) , or force the catalog janitor to run and
then assert that the parent is not (e.g. either hbck or the janitor removed).
Since the first case is more deterministic I prefer it. Can you add some
comments to where we disable the catalog janitor with the justifcation for
disabling? Then I'll commit.
> Flakey testSplitDaughtersNotInMeta test
> ---------------------------------------
>
> Key: HBASE-15206
> URL: https://issues.apache.org/jira/browse/HBASE-15206
> Project: HBase
> Issue Type: Bug
> Components: flakey
> Reporter: huaxiang sun
> Assignee: huaxiang sun
> Priority: Minor
> Attachments: HBASE-15206-001.patch
>
>
> Run into the following failure with hbase 1.0.0.
> Stacktrace
> java.lang.AssertionError: null
> at org.junit.Assert.fail(Assert.java:86)
> at org.junit.Assert.assertTrue(Assert.java:41)
> at org.junit.Assert.assertNotNull(Assert.java:712)
> at org.junit.Assert.assertNotNull(Assert.java:722)
> at
> org.apache.hadoop.hbase.util.TestHBaseFsck.testSplitDaughtersNotInMeta(TestHBaseFsck.java:1723)
> From the log, the ntp issue caused clock skew and it woke up CatalogJanitor
> earlier. The CatalogJanitor cleaned up the parent region. It could happen
> with master branch as well. The fix is to disable CatalogJanitor to make sure
> this will not happen.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
