It works as expected with the latest master update. In cluster mode with coarse grained enabled i should see all cpus utilized by the framework minus 1 used by the Spark Cluster. This is after Michael's changes. I have not verified it with 1.6. If there is still a problem i could disable it in case of coarse-cluster but it seems ok.
dragos
commented
8 years ago
It works as expected with the latest master update. In cluster mode with coarse grained enabled i should see all cpus utilized by the framework minus 1 used by the Spark Cluster. This is after Michael's changes. I have not verified it with 1.6. If there is still a problem i could disable it in case of coarse-cluster but it seems ok.
For an example of this failure see in the next link, 8 cpus are available and the expected value should be 7 not 6: https://ci.typesafe.com/job/spark-mesos-integration-tests-docker-nightly/94/console So this PR fixes that.