[jira] [Commented] (KUDU-1473) ProcessPendingAssignments accesses tablet metadata without a lock

["Todd Lipcon (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/KUDU-1473?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15306955#comment-15306955 ] 

Todd Lipcon commented on KUDU-1473:
-----------------------------------

It's also somewhat curious why this is triggered by the java tests (TestKuduTable) reasonably often but not by any C++ tests. Perhaps we're lacking some coverage

> ProcessPendingAssignments accesses tablet metadata without a lock
> -----------------------------------------------------------------
>
>                 Key: KUDU-1473
>                 URL: https://issues.apache.org/jira/browse/KUDU-1473
>             Project: Kudu
>          Issue Type: Bug
>          Components: master
>    Affects Versions: 0.9.0
>            Reporter: Todd Lipcon
>            Assignee: Adar Dembo
>
> I've seen Java tests being flaky with this crash:
> {code}
> F0530 13:51:16.076247 30686 cow_object.h:82] Check failed: lock_.HasReaders() || lock_.HasWriteLock()
> *** Check failure stack trace: ***
>      @     0x7f8c52d0fe64  kudu::CowObject<>::state()
>      @     0x7f8c52d0848b  kudu::master::AsyncCreateReplica::AsyncCreateReplica()
>      @     0x7f8c52cf7af2  kudu::master::CatalogManager::SendCreateTabletRequests()
>      @     0x7f8c52cf7375  kudu::master::CatalogManager::ProcessPendingAssignments()
>      @     0x7f8c52ce4ad7  kudu::master::CatalogManagerBgTasks::Run()  
> {code}
> It seems like this code path doesn't acquire the read lock on the tablet, and thus fires the assertion



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)