[
https://issues.apache.org/jira/browse/KUDU-1476?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15311857#comment-15311857
]
Mike Percy commented on KUDU-1476:
----------------------------------
I think we should consider looking at our compaction heuristics as well as
potential fragmentation (holes) in the LBM containers. It's possible we're
reclaiming space but without LBM container compaction we still approach this
limit.
Above cluster reportedly has 88 tablets on that TS.
> Log block manager: Open file descriptors: 16,974. File descriptor limit:
> 32,768. Percentage in use: 51.80%. Warning threshold: 50.00%
> -------------------------------------------------------------------------------------------------------------------------------------
>
> Key: KUDU-1476
> URL: https://issues.apache.org/jira/browse/KUDU-1476
> Project: Kudu
> Issue Type: Bug
> Components: tablet
> Affects Versions: 0.8.0
> Reporter: Mike Percy
>
> Reported by Rube Q in Slack.
> Using the log block manager:
> Open file descriptors: 16,974. File descriptor limit: 32,768. Percentage in
> use: 51.80%. Warning threshold: 50.00%
> from Binglin: "like our cluster, some table with low insert rate accumulate
> lot of very small DRS, not getting compacted, and restart ts doesn't help"
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
