Nathan Roberts updated YARN-4287:
    Attachment: YARN-4287-v4.patch

V4 of patch.
- I moved the calculation of locality delays out of canAssign() since this is a 
very hot path and the answer only changes when the size of the cluster changes. 
This caused a few unit tests to start failing because the number of nodes in 
the cluster was not always being mocked at the right time causing the 
LocalityDelays to be 0 which confused some of the assumptions.
- I left the scaling approach in, but am willing to move to a 
rack-locality-delay that is specified as a percent. I absolutely want a 
node-locality-delay set to 5000, rack-locality-delay set to 5100, do something 
intelligent on a 3000 node cluster. 
- One argument for sticking with the scaling approach is the fact that we 
basically do it today in a simpler fashion. If you specify node-locality-delay 
of 5000 on a 3000 node cluster, it gets automatically scaled down to 3000 
without informing the user. So I'd say scale it but don't try to explain it in 
user documentation.
- Updated the documentation 

> Capacity Scheduler: Rack Locality improvement
> ---------------------------------------------
>                 Key: YARN-4287
>                 URL: https://issues.apache.org/jira/browse/YARN-4287
>             Project: Hadoop YARN
>          Issue Type: Improvement
>          Components: capacityscheduler
>    Affects Versions: 2.7.1
>            Reporter: Nathan Roberts
>            Assignee: Nathan Roberts
>         Attachments: YARN-4287-v2.patch, YARN-4287-v3.patch, 
> YARN-4287-v4.patch, YARN-4287.patch
> YARN-4189 does an excellent job describing the issues with the current delay 
> scheduling algorithms within the capacity scheduler. The design proposal also 
> seems like a good direction.
> This jira proposes a simple interim solution to the key issue we've been 
> experiencing on a regular basis:
>  - rackLocal assignments trickle out due to nodeLocalityDelay. This can have 
> significant impact on things like CombineFileInputFormat which targets very 
> specific nodes in its split calculations.
> I'm not sure when YARN-4189 will become reality so I thought a simple interim 
> patch might make sense. The basic idea is simple: 
> 1) Separate delays for rackLocal, and OffSwitch (today there is only 1)
> 2) When we're getting rackLocal assignments, subsequent rackLocal assignments 
> should not be delayed
> Patch will be uploaded shortly. No big deal if the consensus is to go 
> straight to YARN-4189. 

This message was sent by Atlassian JIRA

Reply via email to