[
https://issues.apache.org/jira/browse/HDFS-942?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Todd Lipcon updated HDFS-942:
-----------------------------
Component/s: data-node
Affects Version/s: 0.22.0
> Read and write metrics misleading/not helpful
> ---------------------------------------------
>
> Key: HDFS-942
> URL: https://issues.apache.org/jira/browse/HDFS-942
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: data-node
> Affects Versions: 0.22.0
> Reporter: ryan rawson
>
> the metrics wrap the entire readBlock() writeBlock() call (in
> DataXciever#run())
> The problem with this is it includes the client socket read/write time. This
> means if a client is slow at streaming a write block (eg: in hbase) then it
> skews the write metrics and makes it look like the server is really slow.
> Same problem with reads.
> It would be nice if reads/writes meant actually local disk io and related
> processing (eg: checksum verification/generation)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
