From 524ecb19efc5a0d77336708735f4def7f7b757b2 Mon Sep 17 00:00:00 2001
From: Wangda Tan <wangda@apache.org>
Date: Fri, 8 Dec 2017 21:43:28 -0800
Subject: [PATCH] YARN-7591. NPE in async-scheduling mode of CapacityScheduler.
 (Tao Yang via wangda)

Change-Id: I67e086d6d35c3166ec0f6305490ea6527bd6d799
(cherry picked from commit 41796ea60ab9dbd6a41dd8fd3c6ff50990bba2f5)
(cherry picked from commit c27a5ba65c115f72a83c3bef9599579e38926aac)
---
 .../scheduler/capacity/CapacityScheduler.java |  4 ++++
 .../scheduler/capacity/LeafQueue.java         | 21 ++++++++++++++++++-
 .../common/fica/FiCaSchedulerApp.java         |  5 +++++
 3 files changed, 29 insertions(+), 1 deletion(-)

diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java
index 408032a02f2..e2e2c544650 100644
--- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java
+++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java
@@ -1258,6 +1258,10 @@ private CSAssignment allocateContainerOnSingleNode(PlacementSet<FiCaSchedulerNod
     if (reservedContainer != null) {
       FiCaSchedulerApp reservedApplication = getCurrentAttemptForContainer(
           reservedContainer.getContainerId());
+      if (reservedApplication == null) {
+        LOG.error("Trying to schedule for a finished app, please double check.");
+        return null;
+      }
 
       // Try to fulfill the reservation
       LOG.info(
diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/LeafQueue.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/LeafQueue.java
index e8814951efc..1bbe9034a55 100644
--- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/LeafQueue.java
+++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/LeafQueue.java
@@ -1186,7 +1186,14 @@ public boolean accept(Resource cluster,
             allocation.getSchedulingMode(), null);
 
         // Deduct resources that we can release
-        Resource usedResource = Resources.clone(getUser(username).getUsed(p));
+        User user = getUser(username);
+        if (user == null) {
+          if (LOG.isDebugEnabled()) {
+            LOG.debug("User " + username + " has been removed!");
+          }
+          return false;
+        }
+        Resource usedResource = Resources.clone(user.getUsed(p));
         Resources.subtractFrom(usedResource,
             request.getTotalReleasedResource());
 
@@ -1391,6 +1398,12 @@ Resource computeUserLimitAndSetHeadroom(FiCaSchedulerApp application,
       SchedulingMode schedulingMode, Resource userLimit) {
     String user = application.getUser();
     User queueUser = getUser(user);
+    if (queueUser == null) {
+      if (LOG.isDebugEnabled()) {
+        LOG.debug("User " + user + " has been removed!");
+      }
+      return Resources.none();
+    }
 
     // Compute user limit respect requested labels,
     // TODO, need consider headroom respect labels also
@@ -1485,6 +1498,12 @@ protected boolean canAssignToUser(Resource clusterResource,
     try {
       readLock.lock();
       User user = getUser(userName);
+      if (user == null) {
+        if (LOG.isDebugEnabled()) {
+          LOG.debug("User " + userName + " has been removed!");
+        }
+        return false;
+      }
 
       currentResourceLimits.setAmountNeededUnreserve(Resources.none());
 
diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/common/fica/FiCaSchedulerApp.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/common/fica/FiCaSchedulerApp.java
index 192bfa0b515..cb6ff312406 100644
--- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/common/fica/FiCaSchedulerApp.java
+++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/common/fica/FiCaSchedulerApp.java
@@ -321,6 +321,11 @@ private boolean commonCheckContainerAllocation(
     RMContainer reservedContainerOnNode =
         schedulerContainer.getSchedulerNode().getReservedContainer();
     if (reservedContainerOnNode != null) {
+      // adding NP check as this proposal could not be allocated from reserved
+      // container in async-scheduling mode
+      if (allocation.getAllocateFromReservedContainer() == null) {
+        return false;
+      }
       RMContainer fromReservedContainer =
           allocation.getAllocateFromReservedContainer().getRmContainer();