One of our clusters runs on AWS with a portion of the nodes being spot nodes.
We would like to force the application master not to run on spot nodes. For
what ever reason, application master is not able to recover in cases the node
where it was running suddenly disappears, which is the case with spot nodes.
Standard strategy here is to label the nodes "persistent" and "spot", submit
the job into a queue which always launches in "spot".
Any guidance on this topic is appreciated.
Alex Rovner
Director, Data Engineering
o: 646.759.0052<tel:646.759.0052>
[http://www.attributionrevolution.com/wp-content/uploads/2012/08/Magnetic_Logo.png]<http://www.magnetic.com/>