jackrabbit-oak-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Michael Dürig (JIRA) <j...@apache.org>
Subject [jira] [Comment Edited] (OAK-4097) Add metric for FileStore journal writes
Date Tue, 19 Jul 2016 09:25:20 GMT

    [ https://issues.apache.org/jira/browse/OAK-4097?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15383749#comment-15383749
] 

Michael Dürig edited comment on OAK-4097 at 7/19/16 9:24 AM:
-------------------------------------------------------------

I'd like some advice regarding how to approach this issue. One way would be to add a {{FileStoreMonitor}}
instance variable in {{TarRevisions}} and have a new {{FileStoreMonitor::flushed()}} method
being called at each {{TarRevisions::flush()}} invocation. IMO this would satisfy the requirement,
since {{TarRevisions::flush()}} is the only place where a journal write happens.

On the other hand, as suggested by [~frm], we could intercept calls to {{org.apache.jackrabbit.oak.segment.WriteOperationHandler::flush()}}
"since this method guarantees that pending changes in in-memory segments are persisted on
disk via a write." I think this is a more general approach, but since {{WriteOperationHandler::flush()}}
is called in {{SegmentWriter::flush()}} which in turn is called in numerous places (including
{{FileStore::flush()}} which causes also the journal write), without necessarily modifying
the journal file, I'd assume the metric collected would be one related to {{FileStore}} writes,
not to journal writes.

Any ideas on how to move this forward?


was (Author: dulceanu):
I'd like some advice regarding how to approach this issue. One way would be to add a {{FileStoreMonitor}}
instance variable in {{TarRevisions}} and have a new {{FileStoreMonitor::flushed()}} method
being called at each {{TarRevisions::flush()}} invocation. IMO this would satisfy the requirement,
since {{TarRevisions::flush()}} is the only place where a journal write happens.

On the other hand, as suggested by [~frm], we could intercept calls to {{org.apache.jackrabbit.oak.segment.WriteOperationHandler::flush()}}
"since this method guarantees that pending changes in in-memory segments are persisted on
disk via a write." I think this is a more general approach, but since {{WriteOperationHandler::flush()}}
is called in {{SegmentWriter::flush()}} which in turn is called in numerous places (including
{{FileStore::flush()}} which causes also the journal write), without necessarily modifying
the journal file, I'd assume the metric collected would be one related to {{FileStor}}} writes,
not to journal writes.

Any ideas on how to move this forward?

> Add metric for FileStore journal writes
> ---------------------------------------
>
>                 Key: OAK-4097
>                 URL: https://issues.apache.org/jira/browse/OAK-4097
>             Project: Jackrabbit Oak
>          Issue Type: Improvement
>          Components: segment-tar
>            Reporter: Chetan Mehrotra
>            Assignee: Andrei Dulceanu
>            Priority: Minor
>             Fix For: 1.6, Segment Tar 0.0.6
>
>
> TarMK flush thread should run every 5 secs and flush the current root head to journal.log.
It would be good to have a metric to capture the number of runs per minute
> This would help in confirming if flush is working at expected frequency or delay in acquiring
locks is causing some delays



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

Mime
View raw message