[
https://issues.apache.org/jira/browse/YARN-7105?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16178611#comment-16178611
]
Sunil G edited comment on YARN-7105 at 9/25/17 6:59 AM:
--------------------------------------------------------
Thanks [~dingda6] for this work. Really appreciate the same. I will help you in
getting this in shape and to throw some more thoughts as well.
To start with, let me quote 1st and 2nd points.
bq.1. Each node would be visualized by a horizontal bar, and hierarchy would be
distinguished by indentation of each node. (e.g. If one node is a child node of
another parent node, there would be indentation before the child node.)
bq.2. Each node would show usage info on its bar, e.g. displaying used usage
like a progress bar along with the percentage.
Here each node is a Queue, and sub-nodes are child queues. And queue
usage/configured capacity (not only in percentage, also as absolute values
could be needed). So let me add few more additional points as a summary.
# Configured vs Used capacity of a queue could be shown as percentage and
absolute value.
# We have a partition, and each queue could have different capacity under each
partition. So we could overlay different partitions and its queue capacities in
a sequential way like in old RM UI. Or we could select each partition and then
visualization of queues will be for that partition. I am not very sure whether
point 5 in your list indicating same.
# We could add an option to show visualization from a given parent queue. For
eg, we have 'sales', 'engg', 'marketing'. So one can also colapse its parents
and then view child queues from that point. This could give one department view
for any sub-queues. I think you are already planning to handle in point 3. If
we could also get an option to hide some parent queues, that also would be
really great.
Thanks [~dingda6], will wait for your patches to try :)
was (Author: sunilg):
Thanks [~dingda6] for this work. Really appreciate the same. I will helping you
in getting this in shape and in comments as well.
To start with, let me quote 1st and 2nd points.
bq.1. Each node would be visualized by a horizontal bar, and hierarchy would be
distinguished by indentation of each node. (e.g. If one node is a child node of
another parent node, there would be indentation before the child node.)
bq.2. Each node would show usage info on its bar, e.g. displaying used usage
like a progress bar along with the percentage.
Here each node is a Queue, and sub-nodes are child queues. And queue
usage/configured capacity (not only in percentage, also as absolute values
could be needed). So let me add few more additional points as a summary.
# Configured vs Used capacity of a queue could be shown as percentage and
absolute.
# We have a partition, and each queue could have different capacity under each
partition. So we could overlay different partitions and its queue capacities in
a sequential way like in old RM UI. Or we could select each partition and then
visualization of queues will be for that partition. I am not very sure whether
point 5 in your list indicating same.
# We could add an option to show visualization from a given parent queue. For
eg, we have 'sales', 'engg', 'marketing'. So one can also colapse its parents
and then view child queues from that point. This could give one department view
for any sub-queues. I think you are already planning to handle in point 3. If
we could also get an option to hide some parent queues, that also would be
really great.
Thanks [~dingda6], will wait for your patches to try :)
> Refactor of Queue Visualization
> -------------------------------
>
> Key: YARN-7105
> URL: https://issues.apache.org/jira/browse/YARN-7105
> Project: Hadoop YARN
> Issue Type: Sub-task
> Reporter: Da Ding
> Assignee: Da Ding
> Attachments: Screen Shot 2017-09-24 at 8.19.36 PM.png, Screen Shot
> 2017-09-24 at 8.50.29 PM.png
>
>
> Based on the discussion with [~wangda], current implementation of Queue is
> not easy to use if number of nodes is large, and visualization is not very
> straightforward in terms of node hierarchy and details. Therefore, we want to
> refactor queue visualization to be more user friendly for different use cases.
> [^Screen Shot 2017-09-24 at 8.19.36 PM.png]
> !Screen Shot 2017-09-24 at 8.19.36 PM.png|thumbnail!
> The image above is the mockup for the new queue visualization, basically it
> should have the following features:
> 1. Each node would be visualized by a horizontal bar, and hierarchy would be
> distinguished by indentation of each node. (e.g. If one node is a child node
> of another parent node, there would be indentation before the child node.)
> 2. Each node would show usage info on its bar, e.g. displaying used usage
> like a progress bar along with the percentage.
> 3. Click +/- to expand/collapse the tree to show/hide nodes.
> 4. Click on bar would show detail charts on the bottom.
> 5. A filter on the top-right would filter the nodes based on options.
> [^Screen Shot 2017-09-24 at 8.50.29 PM.png]
> !Screen Shot 2017-09-24 at 8.50.29 PM.png|thumbnail!
> Currently I finished feature 1 and feature 2. See screenshot.
> Since this ticket is relatively big and would take a while to complete, I
> will keep the ticket updated. Don't hesitate to comment your idea or
> suggestion. Thanks!
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
