Ray Chiang commented on YARN-2868:

RE: Comment 1)

Using AtomicLong seemed like the least error prone approach.  Since we only 
expect this to change once, are you asking to implement this as a lock-free 
comparison and only do the locking when updating the value?

RE: Comment 2)

I'm fine with changing names for accuracy.

RE: Comment 3)

The goal for this metric is to measure the functionality plus overhead 
allocation time (for some definition of overhead).  I believe your version will 
differ in the following ways from the current patch:

A) Some overhead in FairScheduler#allocate() will be missed (not necessarily a 
bad thing or big deal)
B) Your suggestion will not work for unmanaged applications.  Were you making 
your suggestion just for managed applications?

Let me know your thoughts on this.

> Add metric for initial container launch time
> --------------------------------------------
>                 Key: YARN-2868
>                 URL: https://issues.apache.org/jira/browse/YARN-2868
>             Project: Hadoop YARN
>          Issue Type: Improvement
>            Reporter: Ray Chiang
>            Assignee: Ray Chiang
>              Labels: metrics, supportability
>         Attachments: YARN-2868-01.patch, YARN-2868.002.patch, 
> YARN-2868.003.patch, YARN-2868.004.patch, YARN-2868.005.patch
> Add a metric to measure the latency between "starting container allocation" 
> and "first container actually allocated".

This message was sent by Atlassian JIRA

Reply via email to