[SPARK-8374] [YARN] Job frequently hangs after YARN preemption

[xuchenCN]

Issue description SPARK-8374

Application starve because YARN scheduler preemption,we should to keep Running containers number
Do not think about the resource ,that is Yarn scheduler's job .

[AmplabJenkins]

Can one of the admins verify this patch?

[srowen]

CC @sryza for a look

[srowen]

OK to test

[SparkQA]

Test build #978 has started for PR 7083 at commit 9555dd5.

[SparkQA]

Test build #978 has finished for PR 7083 at commit 9555dd5.

• This patch fails Scala style tests.
• This patch merges cleanly.
• This patch adds no public classes.

[tgravescs]

@xuchenCN can you add the first "[" to the description before jira number

[sryza]

Hi @xuchenCN what makes you think this is the right fix? We already decrement numExecutorsRunning when a container is killed, so I think doing it again would be double-counting.

[xuchenCN]

Thanks @sryza I notice that is mistake, I 'm trying to find way to fix this issue

[xuchenCN]

PR changes #7128