[
https://issues.apache.org/jira/browse/HBASE-14017?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14613299#comment-14613299
]
Matteo Bertozzi commented on HBASE-14017:
-----------------------------------------
you are looking at the specific table implementation, try to look more at the
RunQueue object alone.
that acquireDelete() is a ref-count of poor people. it has nothing to do with
insert/update/delete.
if we implement refcount for that object you'll have an unref() == 0, instead
of acquireDelete().
the fact that we have read/write lock in the table is because we have
read/write operation support, and since we don't have refcount in the base
RunQueue object we can just implement acquireDelete() as a tryExclusiveLock().
but the acquireDelete() has no knowledge of "delete" operation in term of
"delete table". it is equivalent to a refcounted unref() == 0, how it is
implemented is just a shortcut to use what we have already.
> Procedure v2 - MasterProcedureQueue fix concurrency issue on table queue
> deletion
> ---------------------------------------------------------------------------------
>
> Key: HBASE-14017
> URL: https://issues.apache.org/jira/browse/HBASE-14017
> Project: HBase
> Issue Type: Sub-task
> Components: proc-v2
> Affects Versions: 2.0.0, 1.2.0, 1.1.1, 1.3.0
> Reporter: Matteo Bertozzi
> Assignee: Matteo Bertozzi
> Priority: Blocker
> Fix For: 2.0.0, 1.2.0, 1.1.2
>
> Attachments: HBASE-14017-v0.patch, HBASE-14017.v1-branch1.1.patch
>
>
> [~syuanjiang] found a concurrecy issue in the procedure queue delete where we
> don't have an exclusive lock before deleting the table
> {noformat}
> Thread 1: Create table is running - the queue is empty and wlock is false
> Thread 2: markTableAsDeleted see the queue empty and wlock= false
> Thread 1: tryWrite() set wlock=true; too late
> Thread 2: delete the queue
> Thread 1: never able to release the lock - NPE when trying to get the queue
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
