YARN-4781. Support intra-queue preemption for fairness ordering policy. Contributed by Eric Payne.

This commit is contained in:
Sunil G 2018-05-28 16:32:53 +05:30
parent 61df174e8b
commit 7c343669ba
6 changed files with 366 additions and 10 deletions

View File

@ -34,6 +34,7 @@
import org.apache.commons.logging.LogFactory;
import org.apache.hadoop.yarn.api.records.ApplicationAttemptId;
import org.apache.hadoop.yarn.api.records.Resource;
import org.apache.hadoop.yarn.server.resourcemanager.monitor.capacity.IntraQueueCandidatesSelector.TAFairOrderingComparator;
import org.apache.hadoop.yarn.server.resourcemanager.monitor.capacity.IntraQueueCandidatesSelector.TAPriorityComparator;
import org.apache.hadoop.yarn.server.resourcemanager.monitor.capacity.ProportionalCapacityPreemptionPolicy.IntraQueuePreemptionOrderPolicy;
import org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainer;
@ -41,6 +42,8 @@
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.LeafQueue;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.SchedulingMode;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.policy.FairOrderingPolicy;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.policy.OrderingPolicy;
import org.apache.hadoop.yarn.util.resource.ResourceCalculator;
import org.apache.hadoop.yarn.util.resource.Resources;
@ -263,8 +266,17 @@ private TreeSet<TempAppPerPartition> calculateIdealAssignedResourcePerApp(
Resource queueReassignableResource,
PriorityQueue<TempAppPerPartition> orderedByPriority) {
Comparator<TempAppPerPartition> reverseComp = Collections
.reverseOrder(new TAPriorityComparator());
Comparator<TempAppPerPartition> reverseComp;
OrderingPolicy<FiCaSchedulerApp> queueOrderingPolicy =
tq.leafQueue.getOrderingPolicy();
if (queueOrderingPolicy instanceof FairOrderingPolicy
&& (context.getIntraQueuePreemptionOrderPolicy()
== IntraQueuePreemptionOrderPolicy.USERLIMIT_FIRST)) {
reverseComp = Collections.reverseOrder(
new TAFairOrderingComparator(this.rc, clusterResource));
} else {
reverseComp = Collections.reverseOrder(new TAPriorityComparator());
}
TreeSet<TempAppPerPartition> orderedApps = new TreeSet<>(reverseComp);
String partition = tq.partition;
@ -355,7 +367,16 @@ private PriorityQueue<TempAppPerPartition> createTempAppForResCalculation(
TempQueuePerPartition tq, Collection<FiCaSchedulerApp> apps,
Resource clusterResource,
Map<String, Resource> perUserAMUsed) {
TAPriorityComparator taComparator = new TAPriorityComparator();
Comparator<TempAppPerPartition> taComparator;
OrderingPolicy<FiCaSchedulerApp> orderingPolicy =
tq.leafQueue.getOrderingPolicy();
if (orderingPolicy instanceof FairOrderingPolicy
&& (context.getIntraQueuePreemptionOrderPolicy()
== IntraQueuePreemptionOrderPolicy.USERLIMIT_FIRST)) {
taComparator = new TAFairOrderingComparator(this.rc, clusterResource);
} else {
taComparator = new TAPriorityComparator();
}
PriorityQueue<TempAppPerPartition> orderedByPriority = new PriorityQueue<>(
100, taComparator);
@ -393,13 +414,12 @@ private PriorityQueue<TempAppPerPartition> createTempAppForResCalculation(
// Set ideal allocation of app as 0.
tmpApp.idealAssigned = Resources.createResource(0, 0);
orderedByPriority.add(tmpApp);
// Create a TempUserPerPartition structure to hold more information
// regarding each user's entities such as UserLimit etc. This could
// be kept in a user to TempUserPerPartition map for further reference.
String userName = app.getUser();
if (!usersPerPartition.containsKey(userName)) {
TempUserPerPartition tmpUser = usersPerPartition.get(userName);
if (tmpUser == null) {
ResourceUsage userResourceUsage = tq.leafQueue.getUser(userName)
.getResourceUsage();
@ -409,7 +429,7 @@ private PriorityQueue<TempAppPerPartition> createTempAppForResCalculation(
amUsed = (userSpecificAmUsed == null)
? Resources.none() : userSpecificAmUsed;
TempUserPerPartition tmpUser = new TempUserPerPartition(
tmpUser = new TempUserPerPartition(
tq.leafQueue.getUser(userName), tq.queueName,
Resources.clone(userResourceUsage.getUsed(partition)),
Resources.clone(amUsed),
@ -432,7 +452,10 @@ private PriorityQueue<TempAppPerPartition> createTempAppForResCalculation(
tmpUser.idealAssigned = Resources.createResource(0, 0);
tq.addUserPerPartition(userName, tmpUser);
}
tmpApp.setTempUserPerPartition(tmpUser);
orderedByPriority.add(tmpApp);
}
return orderedByPriority;
}

View File

@ -28,6 +28,8 @@
import org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainer;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.LeafQueue;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.policy.AbstractComparatorOrderingPolicy;
import org.apache.hadoop.yarn.util.resource.ResourceCalculator;
import org.apache.hadoop.yarn.util.resource.Resources;
import java.io.Serializable;
@ -64,6 +66,44 @@ public int compare(TempAppPerPartition ta1, TempAppPerPartition ta2) {
}
}
/*
* Order first by amount used from least to most. Then order from oldest to
* youngest if amount used is the same.
*/
static class TAFairOrderingComparator
implements Comparator<TempAppPerPartition> {
private ResourceCalculator rc;
private Resource clusterRes;
TAFairOrderingComparator(ResourceCalculator rc, Resource clusterRes) {
this.rc = rc;
this.clusterRes = clusterRes;
}
@Override
public int compare(TempAppPerPartition ta1, TempAppPerPartition ta2) {
if (ta1.getUser().equals(ta2.getUser())) {
AbstractComparatorOrderingPolicy<FiCaSchedulerApp> acop =
(AbstractComparatorOrderingPolicy<FiCaSchedulerApp>)
ta1.getFiCaSchedulerApp().getCSLeafQueue().getOrderingPolicy();
return acop.getComparator()
.compare(ta1.getFiCaSchedulerApp(), ta2.getFiCaSchedulerApp());
} else {
Resource usedByUser1 = ta1.getTempUserPerPartition().getUsedDeductAM();
Resource usedByUser2 = ta2.getTempUserPerPartition().getUsedDeductAM();
if (Resources.equals(usedByUser1, usedByUser2)) {
return ta1.getApplicationId().compareTo(ta2.getApplicationId());
}
if (Resources.lessThan(rc, clusterRes, usedByUser1, usedByUser2)) {
return -1;
} else {
return 1;
}
}
}
}
IntraQueuePreemptionComputePlugin fifoPreemptionComputePlugin = null;
final CapacitySchedulerPreemptionContext context;

View File

@ -34,6 +34,7 @@ public class TempAppPerPartition extends AbstractPreemptionEntity {
// Following fields are settled and used by candidate selection policies
private final int priority;
private final ApplicationId applicationId;
private TempUserPerPartition tempUser;
FiCaSchedulerApp app;
@ -102,4 +103,12 @@ public void deductActuallyToBePreempted(ResourceCalculator resourceCalculator,
Resources.subtractFrom(getActuallyToBePreempted(), toBeDeduct);
}
}
public void setTempUserPerPartition(TempUserPerPartition tu) {
tempUser = tu;
}
public TempUserPerPartition getTempUserPerPartition() {
return tempUser;
}
}

View File

@ -26,7 +26,6 @@
import org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainer;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.*;
import org.apache.hadoop.yarn.nodelabels.CommonNodeLabelsManager;
import com.google.common.annotations.VisibleForTesting;
/**
@ -89,7 +88,6 @@ protected void entityRequiresReordering(S schedulableEntity) {
}
}
@VisibleForTesting
public Comparator<SchedulableEntity> getComparator() {
return comparator;
}

View File

@ -53,6 +53,7 @@
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.ContainerPreemptEvent;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.policy.FairOrderingPolicy;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.policy.OrderingPolicy;
import org.apache.hadoop.yarn.util.Clock;
import org.apache.hadoop.yarn.util.resource.DefaultResourceCalculator;
@ -64,6 +65,7 @@
import org.junit.Assert;
import org.junit.Before;
import org.mockito.ArgumentMatcher;
import org.mockito.Mockito;
import org.mockito.invocation.InvocationOnMock;
import org.mockito.stubbing.Answer;
@ -337,9 +339,11 @@ public Integer answer(InvocationOnMock invocation) throws Throwable {
.thenReturn(pendingForDefaultPartition);
// need to set pending resource in resource usage as well
ResourceUsage ru = new ResourceUsage();
ResourceUsage ru = Mockito.spy(new ResourceUsage());
ru.setUsed(label, used);
when(ru.getCachedUsed(anyString())).thenReturn(used);
when(app.getAppAttemptResourceUsage()).thenReturn(ru);
when(app.getSchedulingResourceUsage()).thenReturn(ru);
start = end + 1;
}
@ -637,6 +641,12 @@ public int compare(FiCaSchedulerApp a1, FiCaSchedulerApp a2) {
when(leafQueue.getApplications()).thenReturn(apps);
when(leafQueue.getAllApplications()).thenReturn(apps);
OrderingPolicy<FiCaSchedulerApp> so = mock(OrderingPolicy.class);
String opName = conf.get(CapacitySchedulerConfiguration.PREFIX
+ CapacitySchedulerConfiguration.ROOT + "." + getQueueName(q)
+ ".ordering-policy", "fifo");
if (opName.equals("fair")) {
so = Mockito.spy(new FairOrderingPolicy<FiCaSchedulerApp>());
}
when(so.getPreemptionIterator()).thenAnswer(new Answer() {
public Object answer(InvocationOnMock invocation) {
return apps.descendingIterator();

View File

@ -0,0 +1,276 @@
/**
* 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.hadoop.yarn.server.resourcemanager.monitor.capacity;
import static org.mockito.Matchers.argThat;
import static org.mockito.Mockito.times;
import static org.mockito.Mockito.verify;
import java.io.IOException;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacitySchedulerConfiguration;
import org.junit.Before;
import org.junit.Test;
/*
* Test class for testing intra-queue preemption when the fair ordering policy
* is enabled on a capacity queue.
*/
public class TestProportionalCapacityPreemptionPolicyIntraQueueFairOrdering
extends ProportionalCapacityPreemptionPolicyMockFramework {
@Before
public void setup() {
super.setup();
conf.setBoolean(
CapacitySchedulerConfiguration.INTRAQUEUE_PREEMPTION_ENABLED, true);
policy = new ProportionalCapacityPreemptionPolicy(rmContext, cs, mClock);
}
/*
* When the capacity scheduler fair ordering policy is enabled, preempt first
* from the application owned by the user that is the farthest over their
* user limit.
*/
@Test
public void testIntraQueuePreemptionFairOrderingPolicyEnabledOneAppPerUser()
throws IOException {
// Enable FairOrderingPolicy for yarn.scheduler.capacity.root.a
conf.set(CapacitySchedulerConfiguration.PREFIX
+ CapacitySchedulerConfiguration.ROOT + ".a.ordering-policy", "fair");
// Make sure all containers will be preempted in a single round.
conf.setFloat(CapacitySchedulerConfiguration.
INTRAQUEUE_PREEMPTION_MAX_ALLOWABLE_LIMIT,
(float) 1.0);
String labelsConfig = "=100,true;";
String nodesConfig = // n1 has no label
"n1= res=100";
String queuesConfig =
// guaranteed,max,used,pending,reserved
"root(=[100 100 100 1 0]);" + // root
"-a(=[100 100 100 1 0])"; // a
// user1/app1 has 60 resources in queue a
// user2/app2 has 40 resources in queue a
// user3/app3 is requesting 20 resources in queue a
// With 3 users, preemptable user limit should be around 35 resources each.
// With FairOrderingPolicy enabled on queue a, all 20 resources should be
// preempted from app1
String appsConfig =
// queueName\t(prio,resource,host,expression,#repeat,reserved,pending,user)
"a\t" // app1, user1 in a
+ "(1,1,n1,,60,false,0,user1);" +
"a\t" // app2, user2 in a
+ "(1,1,n1,,40,false,0,user2);" +
"a\t" // app3, user3 in a
+ "(1,1,n1,,0,false,20,user3)"
;
buildEnv(labelsConfig, nodesConfig, queuesConfig, appsConfig);
policy.editSchedule();
verify(mDisp, times(20)).handle(argThat(
new TestProportionalCapacityPreemptionPolicy.IsPreemptionRequestFor(
getAppAttemptId(1))));
}
/*
* When the capacity scheduler fifo ordering policy is enabled, preempt first
* from the youngest application until reduced to user limit, then preempt
* from next youngest app.
*/
@Test
public void testIntraQueuePreemptionFifoOrderingPolicyEnabled()
throws IOException {
// Enable FifoOrderingPolicy for yarn.scheduler.capacity.root.a
conf.set(CapacitySchedulerConfiguration.PREFIX
+ CapacitySchedulerConfiguration.ROOT + ".a.ordering-policy", "fifo");
// Make sure all containers will be preempted in a single round.
conf.setFloat(CapacitySchedulerConfiguration.
INTRAQUEUE_PREEMPTION_MAX_ALLOWABLE_LIMIT,
(float) 1.0);
String labelsConfig = "=100,true;";
String nodesConfig = // n1 has no label
"n1= res=100";
String queuesConfig =
// guaranteed,max,used,pending,reserved
"root(=[100 100 100 1 0]);" + // root
"-a(=[100 100 100 1 0])"; // a
// user1/app1 has 60 resources in queue a
// user2/app2 has 40 resources in queue a
// user3/app3 is requesting 20 resources in queue a
// With 3 users, preemptable user limit should be around 35 resources each.
// With FifoOrderingPolicy enabled on queue a, the first 5 should come from
// the youngest app, app2, until app2 is reduced to the user limit of 35.
String appsConfig =
// queueName\t(prio,resource,host,expression,#repeat,reserved,pending,user)
"a\t" // app1, user1 in a
+ "(1,1,n1,,60,false,0,user1);" +
"a\t" // app2, user2 in a
+ "(1,1,n1,,40,false,0,user2);" +
"a\t" // app3, user3 in a
+ "(1,1,n1,,0,false,5,user3)"
;
buildEnv(labelsConfig, nodesConfig, queuesConfig, appsConfig);
policy.editSchedule();
verify(mDisp, times(5)).handle(argThat(
new TestProportionalCapacityPreemptionPolicy.IsPreemptionRequestFor(
getAppAttemptId(2))));
// user1/app1 has 60 resources in queue a
// user2/app2 has 35 resources in queue a
// user3/app3 has 5 resources and is requesting 15 resources in queue a
// With 3 users, preemptable user limit should be around 35 resources each.
// The next 15 should come from app1 even though app2 is younger since app2
// has already been reduced to its user limit.
appsConfig =
// queueName\t(prio,resource,host,expression,#repeat,reserved,pending,user)
"a\t" // app1, user1 in a
+ "(1,1,n1,,60,false,0,user1);" +
"a\t" // app2, user2 in a
+ "(1,1,n1,,35,false,0,user2);" +
"a\t" // app3, user3 in a
+ "(1,1,n1,,5,false,15,user3)"
;
buildEnv(labelsConfig, nodesConfig, queuesConfig, appsConfig);
policy.editSchedule();
verify(mDisp, times(15)).handle(argThat(
new TestProportionalCapacityPreemptionPolicy.IsPreemptionRequestFor(
getAppAttemptId(1))));
}
/*
* When the capacity scheduler fair ordering policy is enabled, preempt first
* from the youngest application from the user that is the farthest over their
* user limit.
*/
@Test
public void testIntraQueuePreemptionFairOrderingPolicyMulitipleAppsPerUser()
throws IOException {
// Enable FairOrderingPolicy for yarn.scheduler.capacity.root.a
conf.set(CapacitySchedulerConfiguration.PREFIX
+ CapacitySchedulerConfiguration.ROOT + ".a.ordering-policy", "fair");
// Make sure all containers will be preempted in a single round.
conf.setFloat(CapacitySchedulerConfiguration.
INTRAQUEUE_PREEMPTION_MAX_ALLOWABLE_LIMIT,
(float) 1.0);
String labelsConfig = "=100,true;";
String nodesConfig = // n1 has no label
"n1= res=100";
String queuesConfig =
// guaranteed,max,used,pending,reserved
"root(=[100 100 100 1 0]);" + // root
"-a(=[100 100 100 1 0])"; // a
// user1/app1 has 35 resources in queue a
// user1/app2 has 25 resources in queue a
// user2/app3 has 40 resources in queue a
// user3/app4 is requesting 20 resources in queue a
// With 3 users, preemptable user limit should be around 35 resources each.
// With FairOrderingPolicy enabled on queue a, all 20 resources should be
// preempted from app1 since it's the most over served app from the most
// over served user
String appsConfig =
// queueName\t(prio,resource,host,expression,#repeat,reserved,pending,user)
"a\t" // app1 and app2, user1 in a
+ "(1,1,n1,,35,false,0,user1);" +
"a\t"
+ "(1,1,n1,,25,false,0,user1);" +
"a\t" // app3, user2 in a
+ "(1,1,n1,,40,false,0,user2);" +
"a\t" // app4, user3 in a
+ "(1,1,n1,,0,false,20,user3)"
;
buildEnv(labelsConfig, nodesConfig, queuesConfig, appsConfig);
policy.editSchedule();
verify(mDisp, times(20)).handle(argThat(
new TestProportionalCapacityPreemptionPolicy.IsPreemptionRequestFor(
getAppAttemptId(1))));
}
/*
* When the capacity scheduler fifo ordering policy is enabled and a user has
* multiple apps, preempt first from the youngest application.
*/
@Test
public void testIntraQueuePreemptionFifoOrderingPolicyMultipleAppsPerUser()
throws IOException {
// Enable FifoOrderingPolicy for yarn.scheduler.capacity.root.a
conf.set(CapacitySchedulerConfiguration.PREFIX
+ CapacitySchedulerConfiguration.ROOT + ".a.ordering-policy", "fifo");
// Make sure all containers will be preempted in a single round.
conf.setFloat(CapacitySchedulerConfiguration.
INTRAQUEUE_PREEMPTION_MAX_ALLOWABLE_LIMIT,
(float) 1.0);
String labelsConfig = "=100,true;";
String nodesConfig = // n1 has no label
"n1= res=100";
String queuesConfig =
// guaranteed,max,used,pending,reserved
"root(=[100 100 100 1 0]);" + // root
"-a(=[100 100 100 1 0])"; // a
// user1/app1 has 40 resources in queue a
// user1/app2 has 20 resources in queue a
// user3/app3 has 40 resources in queue a
// user4/app4 is requesting 20 resources in queue a
// With 3 users, preemptable user limit should be around 35 resources each.
String appsConfig =
// queueName\t(prio,resource,host,expression,#repeat,reserved,pending,user)
"a\t" // app1, user1 in a
+ "(1,1,n1,,40,false,0,user1);" +
"a\t" // app2, user1 in a
+ "(1,1,n1,,20,false,0,user1);" +
"a\t" // app3, user3 in a
+ "(1,1,n1,,40,false,0,user3);" +
"a\t" // app4, user4 in a
+ "(1,1,n1,,0,false,25,user4)"
;
buildEnv(labelsConfig, nodesConfig, queuesConfig, appsConfig);
policy.editSchedule();
// app3 is the younges and also over its user limit. 5 should be preempted
// from app3 until it comes down to user3's user limit.
verify(mDisp, times(5)).handle(argThat(
new TestProportionalCapacityPreemptionPolicy.IsPreemptionRequestFor(
getAppAttemptId(3))));
// User1's app2 is its youngest. 19 should be preempted from app2, leaving
// only the AM
verify(mDisp, times(19)).handle(argThat(
new TestProportionalCapacityPreemptionPolicy.IsPreemptionRequestFor(
getAppAttemptId(2))));
// Preempt the remaining resource from User1's oldest app1.
verify(mDisp, times(1)).handle(argThat(
new TestProportionalCapacityPreemptionPolicy.IsPreemptionRequestFor(
getAppAttemptId(1))));
}
}