[
https://issues.apache.org/jira/browse/HBASE-9404?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13755670#comment-13755670
]
Jayesh Thakrar commented on HBASE-9404:
---------------------------------------
Just wondering (seeking clarification) on the following line in the description
-
"Have all of the current daemons just publish everything through json and
remove the current UI. Then we can separate the ui out into a monitor daemon
that's just UI."
Is it implying that every daemon publish its own metric?
I think it might be practical to have one monitoring daemon that consolidates
metrics from all daemons/services across different servers/nodes and provides
them collectively.
This will allow further separation of metrics publisher from generators. Also,
the current metrics are at the cluster, regionserver and regionlevel. In
addition to those, I think it might be useful to have metrics at the
"table-level" (e.g. region count, read and write requests across all
regions/regionservers, flushes and compactions happening across all
regions/regionservers, compacted and compacting KVs, etc) and "DFS-level" (max,
min, mean, median, latency times across all regionservers, etc.). I know that
these are derived metrics (i.e. can be computed from existing metrics), but
rather than have every monitoring tool do that, it would be nice to have these
available "out-of-the-box". Thus cluster, regionservers, regions, tables, DFS,
etc. just serve as "roll-ups and/or dimensions" for the base-level
numbers/metrics.
> Create Monitor Daemon instead of current web ui's
> -------------------------------------------------
>
> Key: HBASE-9404
> URL: https://issues.apache.org/jira/browse/HBASE-9404
> Project: HBase
> Issue Type: Improvement
> Components: UI
> Affects Versions: 0.98.0
> Reporter: Elliott Clark
>
> It's hard to get designers and front end developers to iterate on our web ui,
> partly because it's not well separated, and partly because it's in java
> jsp/jamon. It would be awesome to have something like Accumulo's monitor.
> Have all of the current daemons just publish everything through json and
> remove the current UI. Then we can separate the ui out into a monitor daemon
> that's just UI.
> This would do a couple of things:
> * force us to publish everything that's interesting though the
> metrics system. (Lots of stuff is there but some stuff isn't there.)
> * remove a lot of resources from our hbase-server jar.
> * make it easier to add graphing other features to the ui.
> * allow us to remove some things from the rs -> master heart beat.
> * encourage front end developers to get involved (wishful ?)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
