Uh oh!
There was an error while loading. Please reload this page.
[SPARK-1825] Make Windows Spark client work fine with Linux YARN cluster - #3943
[SPARK-1825] Make Windows Spark client work fine with Linux YARN cluster#3943tsudukim wants to merge 3 commits into
Conversation
Modified environment strings and path separators to platform-independent style if possible.
SparkQA
commented
Jan 8, 2015
Test build #25204 has started for PR 3943 at commit
|
SparkQA
commented
Jan 8, 2015
Test build #25204 has finished for PR 3943 at commit
|
AmplabJenkins
commented
Jan 8, 2015
Test PASSed. |
andrewor14
commented
Jan 8, 2015
Hey @tsudukim have you tested this on a real YARN cluster? Also, does this work across YARN versions, say 2.2 through 2.5? |
tsudukim
commented
Jan 9, 2015
thanks @andrewor14 for following this. I have tested it on two YARN clusters: 2.3 and 2.5. Both has 1 master and 3 slaves. This patch is implemented as to work across YARN versions. When we apply this patch on old YARN clusters (2.2, 2.3), the behaviour doesn't change from now because just the function which the current code uses ( |
sarutak
commented
Jan 12, 2015
@tsudukim Recently, YARN module was refactored so could you rebase your change? |
tgravescs
commented
Jan 20, 2015
@tsudukim sorry for the delay in looking at this, could you rebase to latest master? |
There was a problem hiding this comment.
This is probably not a big deal, since this is not in any hot path, but it seems wasteful to look for the right method every time this method is called instead of caching the method to be called. I'd do this:
private lazy val expandMethod = Try(classOf[Environment].getMethod("$$"))
.getOrElse(classOf[Environment].getMethod("$"))
Then your method becomes just:
def expandEnvironment(environment: Environment): String = \
expandMethod.invoke(environment).asInstanceOf[String]
vanzin
commented
Jan 20, 2015
@tsudukim were you able to actually run the unit tests on Windows? I tried that before but failed... |
tsudukim
commented
Jan 21, 2015
@sarutak@tgravescs@vanzin Thank you for your comments! |
aniketbhatnagar
commented
Jan 28, 2015
@tsudukim Would you have time this week to look at this please? I was hoping for this fix to make it in 1.2.1! |
…-1825 Conflicts: yarn/src/main/scala/org/apache/spark/deploy/yarn/ClientBase.scala yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnSparkHadoopUtil.scala yarn/src/test/scala/org/apache/spark/deploy/yarn/ClientSuite.scala
SparkQA
commented
Jan 29, 2015
Test build #26316 has started for PR 3943 at commit
|
tsudukim
commented
Jan 29, 2015
@vanzin Actually, test for org.apache.spark.deploy.yarn.* fails in Windows even in master branch. @aniketbhatnagar Sorry for the delay. I managed to do it, but I don't think this patch goes into 1.2.1. 1.3 might be good. |
SparkQA
commented
Jan 29, 2015
Test build #26316 has finished for PR 3943 at commit
|
AmplabJenkins
commented
Jan 29, 2015
Test PASSed. |
There was a problem hiding this comment.
you missed one Utils.isWindows
andrewor14
commented
Feb 2, 2015
Ok LGTM aside from the minor comments I pointed out. I will fix these myself when I merge this into master. Thanks a lot @tsudukim |
Modified environment strings and path separators to platform-independent style if possible.