[
https://issues.apache.org/jira/browse/HBASE-18261?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Umesh Agashe updated HBASE-18261:
---------------------------------
Attachment: HBASE-18261.master.001.patch
Looking more deep into the code, I found that Meta recovery from
* ServerCrashProcedure and
* HMaster.finishActiveMasterInitialization()
is synchronized on event serverCrashProcessingEnabled in HMaster and not on
event metaInitEvent in AssignmentManager.
Every step in ServerCrashProcedure waits on event serverCrashProcessingEnabled
making sure meta is recovered before proceeding further. There is a bug in the
test framework, patch HBASE-18261.master.001.patch fixes it by resetting
serverCrashProcessingEnabled event to false in simulated master stop().
Restarting HMaster will set it back to true.
I've verified the fix by running the test 10 times in a loop in my dev
environment. Without fix test fails 4/20 times. With fix it passes all 20 times.
There is still a problem with the code, meta will be reassigned by
HMaster.finishActiveMasterInitialization() when assignMeta() is called but when
ServerCrashProcedure is resumed after restart it will try to process and
reassign Meta based on in-memory state of meta when meta is not loaded fully.
De-deuplicating this logic and creating RecoverMetaProcedure is long term
solution that improves code readability. For now, patch
HBASE-18261.master.001.patch fixes and enables test
hbase.master.procedure.TestServerCrashProcedure#testRecoveryAndDoubleExecutionOnRsWithMeta().
Going forward:
* We can keep this JIRA open for tracking RecoverMetaProcedure work and reduce
the priority (improvements) and create a new JIRA for test
hbase.master.procedure.TestServerCrashProcedure#testRecoveryAndDoubleExecutionOnRsWithMeta()
fix.
* Re-purpose this JIRA by changing the title to enable unit test
hbase.master.procedure.TestServerCrashProcedure#testRecoveryAndDoubleExecutionOnRsWithMeta()
Let me know your thoughts. Thanks!
> [AMv2] Create new RecoverMetaProcedure and use it from ServerCrashProcedure
> and HMaster.finishActiveMasterInitialization()
> --------------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-18261
> URL: https://issues.apache.org/jira/browse/HBASE-18261
> Project: HBase
> Issue Type: Improvement
> Components: amv2
> Affects Versions: 2.0.0-alpha-1
> Reporter: Umesh Agashe
> Assignee: Umesh Agashe
> Fix For: 2.0.0-alpha-2
>
> Attachments: HBASE-18261.master.001.patch
>
>
> When unit test
> hbase.master.procedure.TestServerCrashProcedure#testRecoveryAndDoubleExecutionOnRsWithMeta()
> is enabled and run several times, it fails intermittently. Cause is meta
> recovery is done at two different places:
> * ServerCrashProcedure.processMeta()
> * HMaster.finishActiveMasterInitialization()
> and its not coordinated.
> When HMaster.finishActiveMasterInitialization() gets to submit splitMetaLog()
> first and while its running call from ServerCrashProcedure.processMeta()
> fails causing step to be retried again in a loop.
> When ServerCrashProcedure.processMeta() submits splitMetaLog after
> splitMetaLog from HMaster.finishActiveMasterInitialization() is finished,
> success is returned without doing any work.
> But if ServerCrashProcedure.processMeta() submits splitMetaLog request and
> while its going HMaster.finishActiveMasterInitialization() submits it test
> fails with exception.
> [~stack] and I discussed the possible solution:
> Create RecoverMetaProcedure and call it where required. Procedure framework
> provides mutual exclusion and requires idempotence, which should fix the
> problem.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
