[jira] [Commented] (OAK-3819) Collect and expose statistics related to Segment FileStore operations

["Michael Dürig (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/OAK-3819?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15082745#comment-15082745 ] 

Michael Dürig commented on OAK-3819:
------------------------------------

Looks useful. Some notes re. the patch:

* {{FileStoreStats#getSize}} calls {{FileStore#size}}. The latter is very prone to contention and shouldn't be called too often. Since the former is exposed via {{FileStoreStatsMBean}} we can't control this however. I experienced this issue when experimenting with {{SegmentCompactionIT}}, which also monitors the store size. Now I'm using a scheduler to sample the store size in 1 minute intervals (see {{SegmentCompactionIT#scheduleSizeMonitor}}). 
For the same reason [~frm] came up with {{FileStore#approximateSize}}, which your patch seems to upgrade to return the exact size. So why not return {{FileStore#approximateSize}} from {{FileStore#size}} and rename the field accordingly?

* {{FileStoreStats#addWrites}} is called from {{FileStore}}. However the amount of data written is rather a concern of {{TarWriter}} and the method should be called from there. This offset twiddling in {{FileStore#newWriter}} is actually a symptom of this. 

* I don't like the extra call to {{FileStoreStats#reclaimed}} in {{FileStore#cleanup}}. {{GCMonitor}} already provides means for tracking the GC progress.

> Collect and expose statistics related to Segment FileStore operations
> ---------------------------------------------------------------------
>
>                 Key: OAK-3819
>                 URL: https://issues.apache.org/jira/browse/OAK-3819
>             Project: Jackrabbit Oak
>          Issue Type: New Feature
>          Components: segmentmk
>            Reporter: Chetan Mehrotra
>            Assignee: Chetan Mehrotra
>             Fix For: 1.4
>
>         Attachments: OAK-3819-v1.patch, segment-writes.png
>
>
> It would be useful to collect some more stats around FileStore operations in SegmentMK
> * Writes per second - How many bytes are being written measured per second
> * Repository size time series - There is a time series in {{GCMonitorMBean}} but that looks like get updated upon GC and hence bit stale



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