[SPARK-11771][YARN][TRIVIAL] maximum memory in yarn is controlled by two params have both in error msg

[holdenk]

When we exceed the max memory tell users to increase both params instead of just the one.

[SparkQA]

Test build #46057 has finished for PR 9758 at commit 4e530a1.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[srowen]

"and/or"? it could really be due to either one -- the max the scheduler will schedule, or the limit on the max size of a container. It may not need an increase in both.

[tgravescs]

hmm, it would definitely and/or.. but I kind of question why we need this at all? The scheduler max is really what its compared to. If your nodemanagers aren't configured to have enough memory to meet that then I would say you have misconfigured your cluster. I'm fine with leaving the error in there if people fine is useful but I agree with @srowen I think this should be reworded to say something like check the values of ... rather then please increase the values.

[holdenk]

I've switched it to and/or + check the values. I think its a useful error for people trying to run Spark on YARN who might not know what the configuration params are they need here (much better than the alternative of dumping a stack trace to nowhere in particular).

[SparkQA]

Test build #46120 has finished for PR 9758 at commit 5e38c39.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #46121 has finished for PR 9758 at commit 9125fcf.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[andrewor14]

OK, merging into master 1.6