[
https://issues.apache.org/jira/browse/OAK-3155?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14645785#comment-14645785
]
Chetan Mehrotra commented on OAK-3155:
--------------------------------------
bq. What will happen is the #start method will get called again on the next
run, so the watch will reset making the intention to keep the stats running
superfluous.
That would still be fine. There are two stats we collect
# Execution count
# Time taken for each execution
In case of missed checkpoint execution count would not get updated for that run
and execution time would also be assumed to be zero. In next cycle when
checkpoint can be obtained then execution count would be increased and time
taken would only be for that run as watch is resent. So end stats are still
correct.
However if we increment the count even when checkpoint cannot be obtained then
it gives a false impression that indexing is running fine i.e. no of executions
per minute are high which is not the real case. Probably we should track
checkpoint miss separately. Say count number of checkpoint for certain duration
1 min. If the count is more than 5 then log a warning/info message
> AsyncIndex stats do not capture execution for runs where no indexing is
> performed
> ---------------------------------------------------------------------------------
>
> Key: OAK-3155
> URL: https://issues.apache.org/jira/browse/OAK-3155
> Project: Jackrabbit Oak
> Issue Type: Bug
> Components: query
> Reporter: Chetan Mehrotra
> Assignee: Chetan Mehrotra
> Priority: Minor
> Fix For: 1.3.4
>
> Attachments: OAK-3155.patch
>
>
> {{AsyncIndexStats}} currently provides a time series of the stats around when
> async indexing run and what time it took. Current logic only collect stats
> when actual indexing is performed. If no change happened in the repository
> from the last run that stats around that is ignored.
> Due to this if we see a time series of when runs happen then it might not be
> true and gives a false impression that async indexing is slow.
> As a fix we should record execution stats for every run
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
