[
https://issues.apache.org/jira/browse/HBASE-28976?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wellington Chevreuil resolved HBASE-28976.
------------------------------------------
Resolution: Fixed
Merged to master, branch-3, branch-2 and branch-2.6.
> Fix UT flakeyness in TestBucketCache.testBlockAdditionWaitWhenCache and
> TestVerifyBucketCacheFile.testRetrieveFromFile
> ----------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-28976
> URL: https://issues.apache.org/jira/browse/HBASE-28976
> Project: HBase
> Issue Type: Bug
> Affects Versions: 3.0.0-beta-1, 2.7.0, 2.6.1, 2.5.10
> Reporter: Wellington Chevreuil
> Assignee: Wellington Chevreuil
> Priority: Major
> Labels: pull-request-available
> Fix For: 3.0.0, 2.7.0, 2.6.2
>
>
> Noticed these two tests failing intermittently on some of the pre-commit
> runs.
> 1) TestBucketCache.testBlockAdditionWaitWhenCache: The test calls
> BucketCache.cacheBlock() passing true to the waitWhenCache parameter,
> assuming the cache call will wait for a success cache, however this is only
> true if the BucketCache.QUEUE_ADDITION_WAIT_TIME property is also defined, so
> I believe this may be causing the intermittent failures when the pre-commit
> runs on slower vms.
>
> 2) TestVerifyBucketCacheFile.testRetrieveFromFile: One of the checks
> performed by this test is to delete the bucket cache file, shutdown the
> current BucketCache instance, then create a new BucketCache instance that
> would load the persistent file but should fail to recover the cache, as the
> cache file has been deleted. The way this recover works, internally, is
> async. We first read the contents of the persistent file, which includes the
> last serialized backingMap and a checksum of the cache file at that time. It
> then compares this recovered checksum against the current cache file
> checksum. If this verification fails, it starts a background thread to
> traverse all backingMap entries and check if those entries are still
> available in the cache. At this point, the test is ready to check the
> BucketCache allocator size, expecting it to be 0, but if the background
> verification has not finished yet, this assert will fail.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
