[
https://issues.apache.org/jira/browse/AMBARI-6088?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14046339#comment-14046339
]
Srimanth Gunturi commented on AMBARI-6088:
------------------------------------------
[~hars.vardhan], we currently show _Queue Memory Used_ to indicate how
loaded/used a particular queue is (0% = not loaded, 100% fully loaded). We did
that normalization to 100% so that admins can see which queues were reaching
their limit. So with your example, the 'default' queue of 5% capacity will be
mapped to 0%-100%.
You are correct that this does not show excess capacity usages - we should. But
the suggested fix seems to be changing the semantics of what we are showing. If
you have queueA (capacity 30%) and queueB (capacity 70%), at full utilization,
the line for queueA will be at 30% mark and queueB will be at 70% mark. The
admin would need to remember their capacities, and then deduce that they were
running at full load/usage.
We have 2 issues here:
1. What we show currently is factually incorrect
2. We do not show excess usage for a queue
Since Ambari-1.6.1 is very close to being released, I would like to fix the
incorrect values we show when usage is less than 100%. We can then use this
JIRA to look into how best to show excess usages also. Maybe a donut or a pie
chart is a better way to represent capacities, excesses, etc. We will have
freedom in Ambari-1.7.0 release to look into such changes. Hope that is
acceptable.
> Ambari-web's YARN Metric "Queue Memory Used" is not meaningful
> --------------------------------------------------------------
>
> Key: AMBARI-6088
> URL: https://issues.apache.org/jira/browse/AMBARI-6088
> Project: Ambari
> Issue Type: Bug
> Affects Versions: 1.5.1, 1.6.0, 1.6.1
> Reporter: Hars Vardhan
> Labels: patch
> Fix For: 1.7.0
>
> Attachments: AMBARI-6088.patch
>
>
> On this webUI : http://<ambari-host-name>:8080/#/main/services/YARN/summary
> "Queue Memory Used" graph in Ambari' YARN web page seems abnormal. Sometimes,
> queue is using more than 400% of the cluster memory!! How this is calculated?
> It doesn't match with any metrics shown on Resource Manager UI page.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
