[SPARK-5093] Set spark.network.timeout to 120s consistently.

[rxin]

No description provided.

[SparkQA]

Test build #25063 has started for PR 3903 at commit 7c2138e.

• This patch merges cleanly.

[SparkQA]

Test build #25063 has finished for PR 3903 at commit 7c2138e.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
  • SparkSubmit.printErrorAndExit(s"Cannot load main class from JAR $primaryResource")
  • class BinaryClassificationMetrics(

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/25063/
Test PASSed.

[aarondav]

LGTM

[rxin]

Thanks. Merging in master.

[lianhuiwang]

if user do not set blockManagerSlaveTimeoutMs, i think we need to set spark.network.timeout to blockManagerSlaveTimeoutMs. if spark.network.timeout is not defined, we can set the default 120 to spark.network.timeout. @rxin @aarondav
and how about we remove blockManagerSlaveTimeoutMs? blockManger is in executor. we just check timeout of executors. if executor is timeout, we can make sure that blockmanager is timeout. Because in SPARK-5529 executor do not report heartbeat to driver,but driver's akka can receive executor's heartbeat. In this time, executor is not timeout but blockManager is timeout. so it is something wrong.

[rxin]

I think it's fair to remove the blockManagerSlaveTimeoutMs. The various timeouts are not useful ... Let me check with others also.

[rxin]

cc @pwendell and @tdas (I think you wrote this config?)