[jira] [Commented] (OAK-4824) Deadlock in TarWriter

["Alex Parvulescu (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/OAK-4824?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15504275#comment-15504275 ] 

Alex Parvulescu commented on OAK-4824:
--------------------------------------

bq. This is a regression introduced with OAK-4746: the method TarWriter.createNextGeneration() seems over eagerly synchronized.
right. looks bad. the issue is with accessing the {{access}} var which seems to always be protected by a lock on the {{TarWriter}} itself. should we wrap only the {{if (access == null)}} check in a {{synchronized (this)}} block?

> Deadlock in TarWriter
> ---------------------
>
>                 Key: OAK-4824
>                 URL: https://issues.apache.org/jira/browse/OAK-4824
>             Project: Jackrabbit Oak
>          Issue Type: Bug
>          Components: segment-tar
>            Reporter: Michael Dürig
>            Priority: Critical
>              Labels: deadlock, resilience, threading
>             Fix For: Segment Tar 0.0.14
>
>
> There is the potential for a deadlock between concurrent calls to {{TarWriter#createNextGeneration()}} and {{TarWriter#flush()}}: both methods try to acquire a lock on this and another lock on {{TarWriter.file}} but in different order. I observed the deadlock when running {{CompactionAndCleanupIT.randomAccessFileConcurrentReadAndLength()}}.
> This is a regression introduced with OAK-4746: the method {{TarWriter.createNextGeneration()}} seems over eagerly synchronized. I would argue that we could drop synchronization for that method entirely as the part after the call to {{close()}} will in any case only ever be executed once by a single thread. All other threads will fail with an {{IllegalStateException}}).
> [~alexparvulescu], WDYT?



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