You are viewing a plain text version of this content. The canonical link for it is here.

Posted to oak-issues@jackrabbit.apache.org by "Michael Dürig (JIRA)" <ji...@apache.org> on 2015/09/23 17:31:04 UTC

[jira] [Comment Edited] (OAK-3329) TarMK cleanup blocks writers

    [ https://issues.apache.org/jira/browse/OAK-3329?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14904647#comment-14904647 ] 

Michael Dürig edited comment on OAK-3329 at 9/23/15 3:30 PM:
-------------------------------------------------------------

So far I haven't been able to come up with a systematic way to proof my patch effective. However here is some anecdotal evidence:  

!base.png! shows the number of commits over time. At 15:55:43 cleanup runs for 28 seconds. Commits stall completely during that time.

!patched.png! shows the same situation running with the attached patch. Cleanup runs at 17:06:10 for 38 seconds. Commits due not stall in this case. 


was (Author: mduerig):
So far I haven't been able to come up with a systematic way to proof my patch effective. However here is some anecdotal evidence:  

!OAK-3329-base.png! shows the number of commits over time. At 15:55:43 cleanup runs for 28 seconds. Commits stall completely during that time.

!OAK-3329-patched.png! shows the same situation running with the attached patch. Cleanup runs at 17:06:10 for 38 seconds. Commits due not stall in this case. 

> TarMK cleanup blocks writers
> ----------------------------
>
>                 Key: OAK-3329
>                 URL: https://issues.apache.org/jira/browse/OAK-3329
>             Project: Jackrabbit Oak
>          Issue Type: Improvement
>          Components: segmentmk
>            Reporter: Michael Dürig
>            Assignee: Michael Dürig
>              Labels: cleanup, gc
>             Fix For: 1.3.7
>
>         Attachments: OAK-3329.patch, base.png, patched.png
>
>
> TarMK cleanup exclusively locks the {{FileStore}}, which causes concurrent writers to block until cleanup finished. Initially cleanup was expected to be reasonably fast, however I have seen it taking dozens of minutes under certain circumstances (most likely many tar files with many small segments, aka OAK-2896).



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