I just found this framework last night and have been looking through the code 
to understand how it all works. I like its elegance and abstraction away from 
the complexities of creating a Yarn app; however, with simplicity comes 
fidelity loss.

I'd like to know if it's possible for a runnable to spawn another runnable in 
its local container. 

Here is the scenario:

- TwillApp kicks off 3 runnables, each of type "Manager", which will get 
allocated their own containers.
- The Manager is responsible for coordinating Worker tasks and reporting local 
status up to the TwillApp; but it needs to coordinate those workers local to 
itself, not across the cluster's network. (The Manager might take advantage of 
data locality in some cases where he may have multiple local workers, etc)

As I see it, the APIs do not offer a direct way for a runnable to spawn another 
runnable in its own container. Should I just make the runnable another TwillApp 
and have the original/master TwillApp just kick off another app as if he's a 
client? That seems cumbersome and it would be nice if there were just a way a 
runnable could use a context to say "run these runnables in my local container".

Ideas?

Reply via email to