[
https://issues.apache.org/jira/browse/HBASE-20654?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16496859#comment-16496859
]
stack commented on HBASE-20654:
-------------------------------
bq. If expose these regions state info to jxm is very useful.
Please say more than this. Counts of RITs makes sense as a metric and to expose
as JMX or even stuff like longest RIT. I am having trouble understanding why
we need explicit names exposed in metrics. If a problem I'd imagine an operator
will notice it via review of RIT metric counts and when they go to deal with a
specific RIT, they'll figure exact region by reading WAL/log or shell output as
per Sean above or just ask hbck for list of RIT names.
Correct me if I'm wrong. I do not see why our metrics need to be fattened up
with long lists of region names. If a cluster of 1M regions on 1K nodes, at
startup, generating such a metrics few will be crippling.
> Expose regions in transition thru JMX
> -------------------------------------
>
> Key: HBASE-20654
> URL: https://issues.apache.org/jira/browse/HBASE-20654
> Project: HBase
> Issue Type: Improvement
> Reporter: Ted Yu
> Assignee: liubangchen
> Priority: Major
> Attachments: HBASE-20654-1.patch, HBASE-20654-2.patch,
> HBASE-20654-3.patch
>
>
> Currently only the count of regions in transition is exposed thru JMX.
> Here is a sample snippet of the /jmx output:
> {code}
> {
> "beans" : [ {
> ...
> }, {
> "name" : "Hadoop:service=HBase,name=Master,sub=AssignmentManager",
> "modelerType" : "Master,sub=AssignmentManager",
> "tag.Context" : "master",
> ...
> "ritCount" : 3
> {code}
> It would be desirable to expose region name, state for the regions in
> transition as well.
> We can place configurable upper bound on the number of entries returned in
> case there're a lot of regions in transition.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
