[
https://issues.apache.org/jira/browse/HBASE-11462?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14140127#comment-14140127
]
Mikhail Antonov commented on HBASE-11462:
-----------------------------------------
The build seems flaky, but so far I couldn't trace it down to this patch..Last
2 builds I ran today on our internal servers read:
{code}
1)Failed tests:
TestMultiParallel.testActiveThreadsCount:157 expected:<5> but was:<4>
Tests in error:
TestEndToEndSplitTransaction.blockUntilRegionSplit:451->blockUntilRegionIsInMeta:474
NullPointer
Tests run: 1994, Failures: 1, Errors: 1, Skipped: 18
2) Failed tests:
TestSplitLogManager.testGetPreviousRecoveryMode:651 null
TestMasterFailover.testSimpleMasterFailover:154 expected:<5> but was:<6>
Tests in error:
TestMasterFailover.testMetaInTransitionWhenMasterFailover:396 ยป test timed
ou...
TestEndToEndSplitTransaction.blockUntilRegionSplit:451->blockUntilRegionIsInMeta:474
NullPointer
Tests run: 1997, Failures: 2, Errors: 2, Skipped: 19
{code}
So I was suspecting TestEndToEndSplitTransaction is reproducible failure, but
we can see this test did pass in ASF jenkins...
> MetaTableAccessor shouldn't use ZooKeeeper
> ------------------------------------------
>
> Key: HBASE-11462
> URL: https://issues.apache.org/jira/browse/HBASE-11462
> Project: HBase
> Issue Type: Improvement
> Components: Client, Zookeeper
> Affects Versions: 2.0.0
> Reporter: Mikhail Antonov
> Assignee: Mikhail Antonov
> Fix For: 2.0.0
>
> Attachments: HBASE-11462.v4.patch, HBASE-11462.v4.patch,
> HBASE-11462.v4.patch
>
>
> After committing patch for HBASE-4495, there's an further improvement which
> can be made (discussed originally on review board to that jira).
> We have MetaTableAccessor and MetaTableLocator classes. First one is used to
> access information stored in hbase:meta table. Second one is used to deal
> with ZooKeeper state to find out region server hosting hbase:meta, wait for
> it to become available and so on.
> MetaTableAccessor, in turn, should only operate on the meta table content, so
> shouldn't need ZK. The only reason why MetaTableAccessor is using ZK - when
> callers request assignment information, they can request location of meta
> table itself, which we can't read from meta, so in that case
> MetaTableAccessor relays the call to MetaTableLocator. May be the solution
> here is to declare that clients of MetaTableAccessor shall not use it to work
> with meta table itself (not it's content).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
