[ https://issues.apache.org/jira/browse/HDFS-14406?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16822235#comment-16822235 ]
Xue Liu edited comment on HDFS-14406 at 4/19/19 10:30 PM: ---------------------------------------------------------- Hi guys thanks for the comments! Reflected on [~xkrogen]'s comment of extending, as in patch 5, I made these changes: # Make RpcUserMetrics a subclass of RpcDetailedMetrics. Only keep 2 metrics rpcMetrics and rpcDetailedMetrics, and determine whether we create RpcUserMetrics or RpcDetailedMetrics based on enable flag at runtime. # Introduce per user per method metrics, as I am receiving suggestion that this would be even more useful because it gives us an idea on how the RPC cost of each user looks. In patch 6: 1. I found out that that this dynamic type doesn't necessarily benefit our performance, but make the code look more complicated than it should be. Now I am just creating rpcUserMetrics regardless, and only update it when there is enable flag. 2. Kept this per user per user per method metrics. What do you guys think of this idea? [~thefacts] Some other fixes: 1. [~csun] I changed String.format to String plus for better performance. 2. Docs in core-default, Metrics.md are still as user metrics. I will update them once we have impl worked out. 3. A few checkstyles, I will fix them the last. was (Author: xuel1): Hi guys thanks for the comments! Reflected on [~xkrogen]'s comment of extending, as in patch 5, I made these changes: # Make RpcUserMetrics a subclass of RpcDetailedMetrics. Only keep 2 metrics rpcMetrics and rpcDetailedMetrics, and determine whether we create RpcUserMetrics or RpcDetailedMetrics based on enable flag at runtime. # Introduce per user per method metrics, as I am receiving suggestion that this would be even more useful because it gives us an idea on how the RPC cost of each user looks. In patch 6: 1. I found out that that this dynamic type doesn't necessarily benefit our performance, but make the code look more complicated than it should be. Now I am just creating rpcUserMetrics regardless, and only update it when there is enable flag. 2. Kept this per user per user per method metrics. What do you guys think of this idea? [~thefacts] Some other fixes: 1. [~csun] I changed String.format to String plus for better performance. 2. Docs in core-default, Metrics.md are still as user metrics. I will update them once we have impl worked out. 3. A few checkstyles, I will fix them the last. > Add per user RPC Processing time > -------------------------------- > > Key: HDFS-14406 > URL: https://issues.apache.org/jira/browse/HDFS-14406 > Project: Hadoop HDFS > Issue Type: Improvement > Affects Versions: 3.2.0 > Reporter: Xue Liu > Assignee: Xue Liu > Priority: Minor > Fix For: 3.2.0 > > Attachments: HDFS-14406.001.patch, HDFS-14406.002.patch, > HDFS-14406.003.patch, HDFS-14406.004.patch, HDFS-14406.005.patch, > HDFS-14406.006.patch > > > For a shared cluster we would want to separate users' resources, as well as > having our metrics reflecting on the usage, latency, etc, for each user. > This JIRA aims to add per user RPC processing time metrics and expose it via > JMX. -- 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