[
https://issues.apache.org/jira/browse/HBASE-20251?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16409270#comment-16409270
]
Istvan Fajth commented on HBASE-20251:
--------------------------------------
Note: I am not sure if these metrics has to be reinstated or the doc would need
a fix, however as it is removed quite a long time ago, I would say fixing the
documentation can be a good approach, from the percentile metrics and the
others, performance measures can be made. But if someone thinks it is useful to
have an exact counter as well, then we need to re-add this to the metrics, and
leave the doc as is. I do not feel myself the one to decide, also I do not want
to decide on the priority, so feel free to adjust.
> Documentation issue on slow query logging
> -----------------------------------------
>
> Key: HBASE-20251
> URL: https://issues.apache.org/jira/browse/HBASE-20251
> Project: HBase
> Issue Type: Bug
> Reporter: Istvan Fajth
> Priority: Major
>
> The HBase book at the slow query monitoring section
> ([http://hbase.apache.org/book.html#ops.slow.query]) contains the following
> information:
> 1. you can set up slow query monitoring by setting
> hbase.ipc.warn.response.time and for large responses to be monitored you can
> set hbase.ipc.warn.response.size
> 2. if you set the values and it is exceeded, it promises that the count if
> slow queries will be exposed via JMX named as
> hadoop.regionserver_rpc_slowResponse, and it also contains a promise for a
> count metric for queries that were running for more then one second as
> hadoop.regionserver_rpc_methodName.aboveOneSec I assume on a per method basis.
> 3. The doc contains implications of certain tags that are searchable in the
> logs, and the lines are promised to contain json like data about the query.
> Tags mentioned:(responseTooSlow), (responseTooLarge), (operationTooSlow), and
> (operationTooLarge)
>
> After examining the code, when metrics2 and protocol buffers were introduced,
> the tags reduced to only (responseTooSlow) and (responseTooLarge) and the two
> mentioned count metric hadoop.regionserver_rpc_slowResponse and
> hadoop.regionserver_rpc_methodName.aboveOneSec was removed from the metrics.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
