[
https://issues.apache.org/jira/browse/OAK-5368?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15772404#comment-15772404
]
Stefan Eissing commented on OAK-5368:
-------------------------------------
[~chetanm] I agree that it will work for the compacted changes and it always
recovers quite fast in my tests. If it has performance impact I cannot say,
since I investigate other, larger performance blocks. I mainly want to clear
the error logs from things that are not relevant warnings and give a notice
that a length of 5 is not realistic for a busy server.
I think, if it weren't for reusing the BackgroundObserver, a queue length of 1
would be fine where the producer always replaces any existing change with a
new, info=null instance. Which goes into queue compaction strategies,
For this ticket, if we could eliminate the unnecessary warning, I'd be good.
> Not configurable/Unnecessary short Lucene Observation Queue length
> ------------------------------------------------------------------
>
> Key: OAK-5368
> URL: https://issues.apache.org/jira/browse/OAK-5368
> Project: Jackrabbit Oak
> Issue Type: Improvement
> Components: lucene
> Affects Versions: 1.4.10
> Reporter: Stefan Eissing
> Assignee: Chetan Mehrotra
> Priority: Minor
> Fix For: 1.8
>
> Attachments: LuceneIndexConfigObserver-under-load.png
>
>
> The maximum queue length in the {{LuceneIndexConfigObserver}} is hard coded
> to 5. This is unreasonable short for production systems experiencing heavy
> load.
> Tests with a patched version resulted in observed queue length of >100
> entries. This floods warnings into the error log, produces additional load.
> The fix would be to increase the queue max, make it configurable or rely on
> the system default (which can be configured).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
