agrawaldevesh commented on a change in pull request #29014:
URL: https://github.com/apache/spark/pull/29014#discussion_r456724032
##########
File path: core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala
##########
@@ -1767,8 +1767,13 @@ private[spark] class DAGScheduler(
// TODO: mark the executor as failed only if there were lots of
fetch failures on it
if (bmAddress != null) {
- val hostToUnregisterOutputs = if
(env.blockManager.externalShuffleServiceEnabled &&
- unRegisterOutputOnHostOnFetchFailure) {
+ val externalShuffleServiceEnabled =
env.blockManager.externalShuffleServiceEnabled
+ val isHostDecommissioned = taskScheduler
+ .getExecutorDecommissionInfo(bmAddress.executorId)
+ .exists(_.isHostDecommissioned)
Review comment:
@holdenk .. by "local fetch", did you mean
https://github.com/apache/spark/pull/28911 ?
Typically `isHostDecommissioned` is set by the cluster manager: It knows for
sure that the host is lost or not. For example, the Master knows when a Worker
goes away for real. Something similar would have to be done for YARN.
Is the reasoning for adding this check: We don't need to clear the shuffle
state if an executor is decommissioned even if external shuffle service is
disabled, as long as there are other executors on that host ?
If I understand you correctly, the condition you are hinting at is this
(please do confirm):
```
if (host is decommissioned) {
// clear shuffle state for the entire host
} else if (!isLocalFetchEnabled || numRemainingLiveExecutorsOnThisHost ==
0) {
// clear shuffle state for just this executor that suffered the fetch
failure
// because we can be sure that no one else will serve this executor's
data
}
```
I will consider using `isShuffleLost ` here to determine if the shuffle has
been lost here.
##########
File path:
core/src/main/scala/org/apache/spark/scheduler/ExecutorLossReason.scala
##########
@@ -54,9 +54,14 @@ private [spark] object LossReasonPending extends
ExecutorLossReason("Pending los
/**
* @param _message human readable loss reason
* @param workerLost whether the worker is confirmed lost too (i.e. including
shuffle service)
+ * @param causedByApp whether the loss of the slave is the fault of the
running app. (assumed true
Review comment:
Will do
##########
File path:
core/src/main/scala/org/apache/spark/scheduler/ExecutorDecommissionInfo.scala
##########
@@ -0,0 +1,34 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.scheduler
+
+/**
+ * Provides more detail when an executor is being decommissioned.
+ * @param message Human readable reason for why the decommissioning is
happening.
+ * @param isHostDecommissioned Whether the host (aka the `node` or `worker` in
other places) is
+ * being decommissioned too. Used to infer if the
shuffle data might
+ * be lost if external shuffle service is enabled.
+ */
+private[spark]
+case class ExecutorDecommissionInfo(message: String, isHostDecommissioned:
Boolean) {
+ // Shuffle files are lost when an extecutor is decommissioned if:
+ // - No external shuffle service is present and executor is serving the
shuffle data
+ // - It's known that the node (and thus the resident external shuffle
service) is also lost
+ def isShuffleLost(externalShuffleServiceEnabled: Boolean): Boolean =
+ !externalShuffleServiceEnabled || isHostDecommissioned
Review comment:
Same question as above: By local shuffle block serving, did you mean
https://github.com/apache/spark/pull/28911 ? (And the same line of reasoning
for adding this check as well ?)
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
