[SPARK-10515] When killing executor, the pending replacement executors should not be lost

If the heartbeat receiver kills executors (and new ones are not registered to replace them), the idle timeout for the old executors will be lost (and then change a total number of executors requested by Driver), So new ones will be not to asked to replace them.
For example, executorsPendingToRemove=Set(1), and executor 2 is idle timeout before a new executor is asked to replace executor 1. Then driver kill executor 2, and sending RequestExecutors to AM. But executorsPendingToRemove=Set(1,2), So AM doesn't allocate a executor to replace 1.

see: https://github.com/apache/spark/pull/8668

Author: KaiXinXiaoLei <huleilei1@huawei.com>
Author: huleilei <huleilei1@huawei.com>

Closes #8945 from KaiXinXiaoLei/pendingexecutor.

2 files changed, 37 insertions, 0 deletions

diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala
index 18771f79b4..55a564b5c8 100644
--- a/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala
+++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala
@@ -438,6 +438,8 @@ class CoarseGrainedSchedulerBackend(scheduler: TaskSchedulerImpl, val rpcEnv: Rp
     if (!replace) {
       doRequestTotalExecutors(
         numExistingExecutors + numPendingExecutors - executorsPendingToRemove.size)
+    } else {
+      numPendingExecutors += knownExecutors.size
     }
 
     doKillExecutors(executorsToKill)
diff --git a/core/src/test/scala/org/apache/spark/deploy/StandaloneDynamicAllocationSuite.scala b/core/src/test/scala/org/apache/spark/deploy/StandaloneDynamicAllocationSuite.scala
index 2e2fa22eb4..d145e78834 100644
--- a/core/src/test/scala/org/apache/spark/deploy/StandaloneDynamicAllocationSuite.scala
+++ b/core/src/test/scala/org/apache/spark/deploy/StandaloneDynamicAllocationSuite.scala
@@ -369,6 +369,41 @@ class StandaloneDynamicAllocationSuite
     assert(apps.head.getExecutorLimit === 1)
   }
 
+  test("the pending replacement executors should not be lost (SPARK-10515)") {
+    sc = new SparkContext(appConf)
+    val appId = sc.applicationId
+    eventually(timeout(10.seconds), interval(10.millis)) {
+      val apps = getApplications()
+      assert(apps.size === 1)
+      assert(apps.head.id === appId)
+      assert(apps.head.executors.size === 2)
+      assert(apps.head.getExecutorLimit === Int.MaxValue)
+    }
+    // sync executors between the Master and the driver, needed because
+    // the driver refuses to kill executors it does not know about
+    syncExecutors(sc)
+    val executors = getExecutorIds(sc)
+    assert(executors.size === 2)
+    // kill executor 1, and replace it
+    assert(sc.killAndReplaceExecutor(executors.head))
+    eventually(timeout(10.seconds), interval(10.millis)) {
+      val apps = getApplications()
+      assert(apps.head.executors.size === 2)
+    }
+
+    var apps = getApplications()
+    // kill executor 1
+    assert(sc.killExecutor(executors.head))
+    apps = getApplications()
+    assert(apps.head.executors.size === 2)
+    assert(apps.head.getExecutorLimit === 2)
+    // kill executor 2
+    assert(sc.killExecutor(executors(1)))
+    apps = getApplications()
+    assert(apps.head.executors.size === 1)
+    assert(apps.head.getExecutorLimit === 1)
+  }
+
   // ===============================
   // | Utility methods for testing |
   // ===============================