[
https://issues.apache.org/jira/browse/MESOS-701?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Benjamin Mahler updated MESOS-701:
----------------------------------
Description: For large clusters with tens of thousands of slaves, the webui
is unusably slow. (was: For large clusters, the webui performance is abysmal.
There are two mains reasons for this:
1. We use a single state.json endpoint to export all state information from the
master.
2. We do not have paging when we need to display a large number of items in a
table. (e.g. the tasks table under a framework, the slaves table). This
translates to gigantic tables with a lot of DOM elements that need updating as
the state changes.
There are a few pieces that need to be in place to remedy this, for now I'll
capture them all here and we may want to split this apart as this gets worked
on:
1. Support path params in mesos/libprocess, e.g.
/master/state -> Returns all information
/master/state/frameworks -> Returns information for all frameworks
/master/state/frameworks/FID -> Returns information for the specific framework
/master/state/frameworks/FID/tasks -> Returns information for tasks under the
specific framework
/master/state/frameworks/FID/tasks/TID -> Returns information for the specific
task under the specificframework
2. Split the master's state endpoint to be hierarchical as described in the
example above.
3. Update the webui to take advantage of the split state.
Optional:
4. Add offset based pagination to the endpoints, e.g.
/master/state/slaves?offset=50&limit=100
5. Add paged tables to the webui.
Adding the above will make it difficult to find a specific slave / task, which
means we'd likely want to add the ability to search:
6. Add a search regex (or substring) to the relevant endpoints (/slaves and
/framework/FID/tasks).
7. Add a search field to quickly filter tables.)
Labels: scalability (was: )
> Improve webui performance for large clusters.
> ---------------------------------------------
>
> Key: MESOS-701
> URL: https://issues.apache.org/jira/browse/MESOS-701
> Project: Mesos
> Issue Type: Improvement
> Reporter: Benjamin Mahler
> Labels: scalability
>
> For large clusters with tens of thousands of slaves, the webui is unusably
> slow.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
