ODIn/spark-instrumented-optimizer
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity

[SPARK-15725][YARN] Ensure ApplicationMaster sleeps for the min interval.

Browse source 
## What changes were proposed in this pull request?

Update `ApplicationMaster` to sleep for at least the minimum allocation interval before calling `allocateResources`. This prevents overloading the `YarnAllocator` that is happening because the thread is triggered when an executor is killed and its connections die. In YARN, this prevents the app from overloading the allocator and becoming unstable.

## How was this patch tested?

Tested that this allows the an app to recover instead of hanging. It is still possible for the YarnAllocator to be overwhelmed by requests, but this prevents the issue for the most common cause.

Author: Ryan Blue <blue@apache.org>

Closes #13482 from rdblue/SPARK-15725-am-sleep-work-around.
This commit is contained in:
Ryan Blue 2016-06-23 13:54:37 -05:00 committed by Tom Graves
parent 10396d9505
commit a410814c87
1 changed files with 22 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

25
yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala
View file

@ -457,8 +457,10 @@ private[spark] class ApplicationMaster(
}
try {
val numPendingAllocate = allocator.getPendingAllocate.size
var sleepStart = 0L
var sleepInterval = 200L // ms
allocatorLock.synchronized {
val sleepInterval =
sleepInterval =
if (numPendingAllocate > 0 || allocator.getNumPendingLossReasonRequests > 0) {
val currentAllocationInterval =
math.min(heartbeatInterval, nextAllocationInterval)
@ -468,10 +470,27 @@ private[spark] class ApplicationMaster(
nextAllocationInterval = initialAllocationInterval
heartbeatInterval
}
logDebug(s"Number of pending allocations is $numPendingAllocate. " +
s"Sleeping for $sleepInterval.")
sleepStart = System.currentTimeMillis()
allocatorLock.wait(sleepInterval)
}
val sleepDuration = System.currentTimeMillis() - sleepStart
if (sleepDuration < sleepInterval) {
// log when sleep is interrupted
logDebug(s"Number of pending allocations is $numPendingAllocate. " +
s"Slept for $sleepDuration/$sleepInterval ms.")
// if sleep was less than the minimum interval, sleep for the rest of it
val toSleep = math.max(0, initialAllocationInterval - sleepDuration)
if (toSleep > 0) {
logDebug(s"Going back to sleep for $toSleep ms")
// use Thread.sleep instead of allocatorLock.wait. there is no need to be woken up
// by the methods that signal allocatorLock because this is just finishing the min
// sleep interval, which should happen even if this is signalled again.
Thread.sleep(toSleep)
}
} else {
logDebug(s"Number of pending allocations is $numPendingAllocate. " +
s"Slept for $sleepDuration/$sleepInterval.")
}
} catch {
case e: InterruptedException =>
}