Skip to content

[SPARK-17644] [CORE] Do not add failedStages when abortStage for fetch failure #15213

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
scwf wants to merge 9 commits into from
Closed

[SPARK-17644] [CORE] Do not add failedStages when abortStage for fetch failure #15213

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
8e667f5
test case
scwf Sep 23, 2016
2bfa05b
The failed stage never resubmitted
scwf Sep 23, 2016
d02cf93
test case
scwf Sep 23, 2016
7056cd6
fix style
scwf Sep 23, 2016
1f7bd88
test case imporvement
scwf Sep 23, 2016
d92adfc
comment fix
scwf Sep 23, 2016
1127ca1
address comments of zsxwing
scwf Sep 23, 2016
f91d86f
improve test
scwf Sep 28, 2016
09077cb
comment fix
scwf Sep 28, 2016
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
fix style
  • Loading branch information
@scwf
scwf committed Sep 23, 2016
commit 7056cd6efc301e8730546c6dd8b24c7f3b56c55a
6 changes: 3 additions & 3 deletions core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1235,7 +1235,7 @@ class DAGScheduler(
case FetchFailed(bmAddress, shuffleId, mapId, reduceId, failureMessage) =>
val failedStage = stageIdToStage(task.stageId)
val mapStage = shuffleIdToMapStage(shuffleId)
var abortStage = false
var abortedStage = false
Copy link
Member

@zsxwing zsxwing Sep 23, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could you move this line just above if (disallowStageRetryForTest) { since it's only used in that scope.


if (failedStage.latestInfo.attemptId != task.stageAttemptId) {
logInfo(s"Ignoring fetch failure from $task as it's from $failedStage attempt" +
Expand All @@ -1262,7 +1262,7 @@ class DAGScheduler(
s"has failed the maximum allowable number of " +
s"times: ${Stage.MAX_CONSECUTIVE_FETCH_FAILURES}. " +
s"Most recent failure reason: ${failureMessage}", None)
abortStage = true
abortedStage = true
Copy link
Member

@zsxwing zsxwing Sep 23, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There is another abortStage in if (disallowStageRetryForTest) { branch.

} else if (failedStages.isEmpty) {
Copy link
Contributor

@kayousterhout kayousterhout Sep 28, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Instead of having the abortedStage variable, how about re-writing the "else if" statement to be:

else {
if (failedStages.isEmpty) {
... stuff currently in else-if ...
}
failedStages += failedStage
failedStages += mapStage
}

That eliminates the confusion of multiple abortStage variables, as @zsxwing pointed out, and also makes the relationship between (i) adding the stage to failed stages and (ii) scheduling the Resubmit event more clear.

Copy link
Contributor Author

@scwf scwf Sep 28, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

it make sense to me, updated

// Don't schedule an event to resubmit failed stages if failed isn't empty, because
// in that case the event will already have been scheduled.
Expand All @@ -1273,7 +1273,7 @@ class DAGScheduler(
override def run(): Unit = eventProcessLoop.post(ResubmitFailedStages)
}, DAGScheduler.RESUBMIT_TIMEOUT, TimeUnit.MILLISECONDS)
}
if (!abortStage) {
if (!abortedStage) {
failedStages += failedStage
failedStages += mapStage
}
Expand Down
8 changes: 5 additions & 3 deletions core/src/test/scala/org/apache/spark/scheduler/DAGSchedulerSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,8 +34,8 @@ import org.apache.spark._
import org.apache.spark.broadcast.BroadcastManager
import org.apache.spark.rdd.RDD
import org.apache.spark.scheduler.SchedulingMode.SchedulingMode
import org.apache.spark.shuffle.MetadataFetchFailedException
import org.apache.spark.shuffle.FetchFailedException
Copy link
Contributor

@kayousterhout kayousterhout Sep 28, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: can you group this with the next import (so import org.apache.spark.shuffle.{FetchFailedException, MetadataFetchFailedException}

import org.apache.spark.shuffle.MetadataFetchFailedException
import org.apache.spark.storage.{BlockId, BlockManagerId, BlockManagerMaster}
import org.apache.spark.util._

Expand Down Expand Up @@ -2121,7 +2121,8 @@ class DAGSchedulerSuite extends SparkFunSuite with LocalSparkContext with Timeou
rdd1.dependencies.head.asInstanceOf[ShuffleDependency[_, _, _]].shuffleHandle
rdd1.map { x =>
Copy link

@Mironor Mironor Sep 23, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You may write

rdd1.map { 
    case (x, _) =>              
        if (x == 1) {
                throw new FetchFailedException(BlockManagerId("1", "1", 1), shuffleHandle.shuffleId, 0, 0, "test")
        }
        x
}.count()

instead. Pattern matching on tuples looks more readable than accessing to _1 attribute

Copy link

@Mironor Mironor Sep 23, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Or even

rdd1.map { 
    case (x, _) if (x == 1)  =>  throw new FetchFailedException(BlockManagerId("1", "1", 1), shuffleHandle.shuffleId, 0, 0, "test")             
    case (x, _) => x
}.count()

Copy link
Contributor Author

@scwf scwf Sep 23, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ok, thanks

if (x._1 == 1) {
throw new FetchFailedException(BlockManagerId("1", "1", 1), shuffleHandle.shuffleId, 0, 0, "test")
throw new FetchFailedException(
BlockManagerId("1", "1", 1), shuffleHandle.shuffleId, 0, 0, "test")
}
x._1
}.count()
Expand All @@ -2138,7 +2139,8 @@ class DAGSchedulerSuite extends SparkFunSuite with LocalSparkContext with Timeou
rdd2.dependencies.head.asInstanceOf[ShuffleDependency[_, _, _]].shuffleHandle
rdd2.map { x =>
if (x._1 == 1) {
throw new FetchFailedException(BlockManagerId("1", "1", 1), shuffleHandle.shuffleId, 0, 0, "test")
throw new FetchFailedException(
BlockManagerId("1", "1", 1), shuffleHandle.shuffleId, 0, 0, "test")
}
x._1
}.count()
Expand Down