[jira] [Resolved] (KUDU-2184) Avoid large allocations in MemRowSet arena

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

     [ https://issues.apache.org/jira/browse/KUDU-2184?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Todd Lipcon resolved KUDU-2184.
-------------------------------
       Resolution: Fixed
    Fix Version/s: 1.6.0

> Avoid large allocations in MemRowSet arena
> ------------------------------------------
>
>                 Key: KUDU-2184
>                 URL: https://issues.apache.org/jira/browse/KUDU-2184
>             Project: Kudu
>          Issue Type: Improvement
>          Components: perf
>    Affects Versions: 1.6.0
>            Reporter: Todd Lipcon
>            Assignee: Todd Lipcon
>             Fix For: 1.6.0
>
>
> tcmalloc caches metadata for spans of free space only up to a certain size (1MB). For larger allocations, an O(n) search is used, which can be quite slow according to https://github.com/gperftools/gperftools/issues/535 . Based on some recent analysis of latency outliers in a long-running cluster, I think we may be hitting this issue on the MRS Arena. We should set the max component size to 1MB to avoid it.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)