[
https://issues.apache.org/jira/browse/HDDS-961?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16802223#comment-16802223
]
Nanda kumar commented on HDDS-961:
----------------------------------
[~swagle], +1 on reporting the metrics from datanode itself.
{quote}bq. Alternatively, could we include completion time in the CommandStatus
object? Should we create something like SCMCommandMetrics per command type and
it stores these global counters for the average completion time of the specific
command?
{quote}
We don't send the status for all the commands to SCM, and also there is a
proposal (HDDS-895) to remove command watcher from Replication Manager (We
won't be needing CommandStatus for Replicate and Delete command if this is
done). After these changes, the number of commands for which a datanode sends
its status to SCM will be very small (only Delete Block Command). Maintaining a
global counter in SCM for just one command type feels like overkill. The
metrics exposed from datanode should be sufficient.
I will change the Jira summary and the description.
> Send command execution metrics from Datanode to SCM
> ---------------------------------------------------
>
> Key: HDDS-961
> URL: https://issues.apache.org/jira/browse/HDDS-961
> Project: Hadoop Distributed Data Store
> Issue Type: Improvement
> Components: Ozone Datanode, SCM
> Reporter: Nanda kumar
> Priority: Major
>
> The CommandHandlers in datanode calculates and tracks the time taken to
> execute each command that is sent by SCM. It would be nice to report these
> values to SCM so that we can build average time, std dev etc for those
> operations.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
