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

Posted to issues@kudu.apache.org by "Todd Lipcon (JIRA)" <ji...@apache.org> on 2016/05/12 02:07:13 UTC

[jira] [Updated] (KUDU-1450) LogBlockManager start-up is slow due to MemTracker atomic contention

     [ https://issues.apache.org/jira/browse/KUDU-1450?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Todd Lipcon updated KUDU-1450:
------------------------------
    Attachment: fg.svg

> LogBlockManager start-up is slow due to MemTracker atomic contention
> --------------------------------------------------------------------
>
>                 Key: KUDU-1450
>                 URL: https://issues.apache.org/jira/browse/KUDU-1450
>             Project: Kudu
>          Issue Type: Bug
>          Components: fs, perf
>    Affects Versions: 0.8.0
>            Reporter: Todd Lipcon
>            Assignee: Todd Lipcon
>         Attachments: fg.svg
>
>
> I'm looking at startup time on a cluster with lots of data per node. On this cluster, the tablet server takes about 30 seconds to start the LogBlockManager, and appears CPU bound. I grabbed a profile and it seems like most of the CPU is spent in MemTracker::Release and Consume (see attached flame graph).



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