[
https://issues.apache.org/jira/browse/PHOENIX-4759?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16496183#comment-16496183
]
Sergey Soldatov commented on PHOENIX-4759:
------------------------------------------
That's well reproduced on 5.x branch, but may (should) affect master branch
under certain circumstances. The SYSTEM.CATALOG region get stuck during open
operation because there are 2 concurrent open region threads that are trying to
load system tables.
Loading sequence:
Thread 1:
MetaDataEndpointImpl -> PhoenixDatabaseMetaData -- (trying to load
QueryConstants)
Thread 2:
MetaDataRegionObserver -> QueryConstants -> TableProperty -> SQLExceptionCode
-> (trying to load PhoenixDatabaseMetaData)
Since only one thread is capable to load class and second thread is already
loading QueryConstants and first thread is loading PhoenixDatabaseMetaData , we
have a dead lock.
We can break this by removing the dependency between SQLExceptionCode and
PhoenixDatabaseMetaData.
> During restart RS that hosts SYSTEM.CATALOG table may get stuck
> ---------------------------------------------------------------
>
> Key: PHOENIX-4759
> URL: https://issues.apache.org/jira/browse/PHOENIX-4759
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 4.14.0, 5.0.0
> Reporter: Romil Choksi
> Assignee: Sergey Soldatov
> Priority: Blocker
> Fix For: 4.14.0, 5.0.0
>
>
> Sometimes when a cluster has restarted the regions that belong to
> SYSTEM.CATALOG and other system tables on the same RS may be stuck in RiT.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
