[
https://issues.apache.org/jira/browse/HBASE-28907?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18129504#comment-18129504
]
mazhengxuan commented on HBASE-28907:
-------------------------------------
I propose two process-wide counters for FSFT manifest read and write attempts,
including failures. These would be updated in StoreFileListFile at each
single-file read/write attempt. Reading both f1/f2 would count twice; listing,
deletion and filesystem-internal retries would be excluded.
I’d reuse MetricsIO with the Counter API and verify metrics export from both
Master and RegionServer. The patch would include focused tests, leaving
latency, bytes, per-region metrics and caching for separate work.
Does this scope make sense? Is MetricsIO an appropriate home for these metrics,
including Master-side operations?
> Add metrics around FSFT backed file reads and updates
> -----------------------------------------------------
>
> Key: HBASE-28907
> URL: https://issues.apache.org/jira/browse/HBASE-28907
> Project: HBase
> Issue Type: Improvement
> Reporter: Prathyusha
> Priority: Major
>
> With HBASE-27826, number of interactions via SFT layer would increase, as all
> the interactions of StoreFiles going forward would require StoreFileTracker
> context and that would increase the number of reads to inherent backed file
> of FileBaseStoreFileTracker, this would add metrics on number of times we
> read/update backed file
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
