[
https://issues.apache.org/jira/browse/HDFS-10959?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15756117#comment-15756117
]
Arpit Agarwal commented on HDFS-10959:
--------------------------------------
Thanks Xiaoyu. The patch looks good.
Minor comments:
# FsVolumeImpl#metrics should be final.
# We can replace MetadataFileIo with MetadataOperation everwhere, since one
call may result in multiple IOs.
# Why do we have {{volumeName.replace(':', '-')}} in
{{DataNodeVolumeMetrics#create}}? Is it because the {{:}} character is not
accepted as part of the metric name?
# beforeFileIo and afterFileIo should use sampling. Updating global state for
every packet looks inefficient. But we can add the sampling later as the
profiling is off by default.
> Adding per disk IO statistics in DataNode.
> ------------------------------------------
>
> Key: HDFS-10959
> URL: https://issues.apache.org/jira/browse/HDFS-10959
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: datanode
> Reporter: Xiaoyu Yao
> Assignee: Xiaoyu Yao
> Attachments: HDFS-10959.00.patch
>
>
> This ticket is opened to support per disk IO statistics in DataNode based on
> HDFS-10930. The statistics added will make it easier to implement HDFS-4169
> "Add per-disk latency metrics to DataNode".
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
