Varun Vasudev commented on YARN-3998:

I'm thinking of re-opening this issue because we've seen a use case for this - 
long running services which (ideally) shouldn't lose local data if the service 

Some pros about supporting some forms of restart policies on the NM -

1. Retry policies can be unified instead of every application having to 
re-implement their own.

2. Faster restarts - instead of the NM reaching out to the AM and then deciding 
what to do(and maintaining the container work dir), it can make an immediate 
decision. It's also an easier change to make - if the NMs need to talk to the 
AMs to decide whether to restart a container - we'll probably need a new state 
transition. Instead if we allow the AMs to specify a restart policy, the NM can 
make an immediate decision as soon as the container exits.

3. Similar to what Jun mentioned - when running Docker containers, it's useful 
to be able to restart containers that exit with an error code.

When I say restart policies - off the top of my head - I can think of 3 
policies - never restart(default), restart on all errors, restart on specific 
error codes.

[~jlowe], [~steve_l] - do you guys still feel that this should be done at the 
app level(and essentially re-implemented by every app)?

> Add retry-times to let NM re-launch container when it fails to run
> ------------------------------------------------------------------
>                 Key: YARN-3998
>                 URL: https://issues.apache.org/jira/browse/YARN-3998
>             Project: Hadoop YARN
>          Issue Type: New Feature
>            Reporter: Jun Gong
>            Assignee: Jun Gong
> I'd like to add a field(retry-times) in ContainerLaunchContext. When AM 
> launches containers, it could specify the value. Then NM will re-launch the 
> container 'retry-times' times when it fails to run(e.g.exit code is not 0). 
> It will save a lot of time. It avoids container localization. RM does not 
> need to re-schedule the container. And local files in container's working 
> directory will be left for re-use.(If container have downloaded some big 
> files, it does not need to re-download them when running again.) 
> We find it is useful in systems like Storm.

This message was sent by Atlassian JIRA

Reply via email to