YARN-595. Refactor fair scheduler to use common Resources. Contributed by Sandy Ryza.

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/trunk@1475670 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Thomas White 2013-04-25 08:33:43 +00:00
parent 448df18646
commit a43c344eee
12 changed files with 130 additions and 195 deletions

View File

@ -184,6 +184,9 @@ Release 2.0.5-beta - UNRELEASED
YARN-577. Add application-progress also to ApplicationReport. (Hitesh Shah
via vinodkv)
YARN-595. Refactor fair scheduler to use common Resources. (Sandy Ryza
via tomwhite)
OPTIMIZATIONS
BUG FIXES

View File

@ -208,4 +208,14 @@ public static Resource max(
Resource lhs, Resource rhs) {
return resourceCalculator.compare(clusterResource, lhs, rhs) >= 0 ? lhs : rhs;
}
public static boolean fitsIn(Resource smaller, Resource bigger) {
return smaller.getMemory() <= bigger.getMemory() &&
smaller.getVirtualCores() <= bigger.getVirtualCores();
}
public static Resource componentwiseMin(Resource lhs, Resource rhs) {
return createResource(Math.min(lhs.getMemory(), rhs.getMemory()),
Math.min(lhs.getVirtualCores(), rhs.getVirtualCores()));
}
}

View File

@ -33,6 +33,7 @@
import org.apache.hadoop.yarn.api.records.QueueACL;
import org.apache.hadoop.yarn.api.records.QueueUserACLInfo;
import org.apache.hadoop.yarn.api.records.Resource;
import org.apache.hadoop.yarn.server.resourcemanager.resource.Resources;
public class FSLeafQueue extends FSQueue {
private static final Log LOG = LogFactory.getLog(
@ -126,8 +127,8 @@ public void updateDemand() {
+ demand);
}
demand = Resources.add(demand, toAdd);
if (Resources.greaterThanOrEqual(demand, maxRes)) {
demand = maxRes;
demand = Resources.componentwiseMin(demand, maxRes);
if (Resources.equals(demand, maxRes)) {
break;
}
}
@ -153,7 +154,7 @@ public Resource assignContainer(FSSchedulerNode node) {
for (AppSchedulable sched : appScheds) {
if (sched.getRunnable()) {
assigned = sched.assignContainer(node);
if (Resources.greaterThan(assigned, Resources.none())) {
if (!assigned.equals(Resources.none())) {
break;
}
}

View File

@ -29,6 +29,7 @@
import org.apache.hadoop.yarn.api.records.QueueACL;
import org.apache.hadoop.yarn.api.records.QueueUserACLInfo;
import org.apache.hadoop.yarn.api.records.Resource;
import org.apache.hadoop.yarn.server.resourcemanager.resource.Resources;
public class FSParentQueue extends FSQueue {
private static final Log LOG = LogFactory.getLog(
@ -87,8 +88,8 @@ public void updateDemand() {
" now " + demand);
}
demand = Resources.add(demand, toAdd);
if (Resources.greaterThanOrEqual(demand, maxRes)) {
demand = maxRes;
demand = Resources.componentwiseMin(demand, maxRes);
if (Resources.equals(demand, maxRes)) {
break;
}
}
@ -135,16 +136,14 @@ public Resource assignContainer(FSSchedulerNode node) {
Resource assigned = Resources.none();
// If this queue is over its limit, reject
if (Resources.greaterThan(getResourceUsage(),
queueMgr.getMaxResources(getName()))) {
if (!assignContainerPreCheck(node)) {
return assigned;
}
Collections.sort(childQueues, policy.getComparator());
for (FSQueue child : childQueues) {
assigned = child.assignContainer(node);
if (node.getReservedContainer() != null
|| Resources.greaterThan(assigned, Resources.none())) {
if (!Resources.equals(assigned, Resources.none())) {
break;
}
}

View File

@ -32,6 +32,7 @@
import org.apache.hadoop.yarn.api.records.Resource;
import org.apache.hadoop.yarn.factories.RecordFactory;
import org.apache.hadoop.yarn.factory.providers.RecordFactoryProvider;
import org.apache.hadoop.yarn.server.resourcemanager.resource.Resources;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.Queue;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.QueueMetrics;
@ -161,7 +162,7 @@ public boolean hasAccess(QueueACL acl, UserGroupInformation user) {
* @return true if check passes (can assign) or false otherwise
*/
protected boolean assignContainerPreCheck(FSSchedulerNode node) {
if (Resources.greaterThan(getResourceUsage(),
if (!Resources.fitsIn(getResourceUsage(),
queueMgr.getMaxResources(getName()))
|| node.getReservedContainer() != null) {
return false;

View File

@ -46,7 +46,7 @@ public class FSSchedulerNode extends SchedulerNode {
private static final RecordFactory recordFactory = RecordFactoryProvider
.getRecordFactory(null);
private Resource availableResource = recordFactory.newRecordInstance(Resource.class);
private Resource availableResource;
private Resource usedResource = recordFactory.newRecordInstance(Resource.class);
private volatile int numContainers;
@ -62,7 +62,7 @@ public class FSSchedulerNode extends SchedulerNode {
public FSSchedulerNode(RMNode node) {
this.rmNode = node;
this.availableResource.setMemory(node.getTotalCapability().getMemory());
this.availableResource = Resources.clone(node.getTotalCapability());
}
public RMNode getRMNode() {

View File

@ -52,6 +52,9 @@
import org.apache.hadoop.yarn.server.resourcemanager.RMAuditLogger.AuditConstants;
import org.apache.hadoop.yarn.server.resourcemanager.RMContext;
import org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore.RMState;
import org.apache.hadoop.yarn.server.resourcemanager.resource.DefaultResourceCalculator;
import org.apache.hadoop.yarn.server.resourcemanager.resource.ResourceCalculator;
import org.apache.hadoop.yarn.server.resourcemanager.resource.Resources;
import org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttemptEvent;
import org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttemptEventType;
import org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttemptState;
@ -111,6 +114,9 @@ public class FairScheduler implements ResourceScheduler {
private static final Log LOG = LogFactory.getLog(FairScheduler.class);
private static final ResourceCalculator RESOURCE_CALCULATOR =
new DefaultResourceCalculator();
// Value that container assignment methods return when a container is
// reserved
public static final Resource CONTAINER_RESERVED = Resources.createResource(-1);
@ -246,8 +252,10 @@ private void updatePreemptionVariables() {
* Is a queue below its min share for the given task type?
*/
boolean isStarvedForMinShare(FSLeafQueue sched) {
Resource desiredShare = Resources.min(sched.getMinShare(), sched.getDemand());
return Resources.lessThan(sched.getResourceUsage(), desiredShare);
Resource desiredShare = Resources.min(RESOURCE_CALCULATOR, clusterCapacity,
sched.getMinShare(), sched.getDemand());
return Resources.lessThan(RESOURCE_CALCULATOR, clusterCapacity,
sched.getResourceUsage(), desiredShare);
}
/**
@ -255,9 +263,10 @@ boolean isStarvedForMinShare(FSLeafQueue sched) {
* defined as being below half its fair share.
*/
boolean isStarvedForFairShare(FSLeafQueue sched) {
Resource desiredFairShare = Resources.max(
Resource desiredFairShare = Resources.max(RESOURCE_CALCULATOR, clusterCapacity,
Resources.multiply(sched.getFairShare(), .5), sched.getDemand());
return Resources.lessThan(sched.getResourceUsage(), desiredFairShare);
return Resources.lessThan(RESOURCE_CALCULATOR, clusterCapacity,
sched.getResourceUsage(), desiredFairShare);
}
/**
@ -283,7 +292,8 @@ protected synchronized void preemptTasksIfNecessary() {
for (FSLeafQueue sched : queueMgr.getLeafQueues()) {
resToPreempt = Resources.add(resToPreempt, resToPreempt(sched, curTime));
}
if (Resources.greaterThan(resToPreempt, Resources.none())) {
if (Resources.greaterThan(RESOURCE_CALCULATOR, clusterCapacity, resToPreempt,
Resources.none())) {
preemptResources(queueMgr.getLeafQueues(), resToPreempt);
}
}
@ -309,7 +319,8 @@ protected void preemptResources(Collection<FSLeafQueue> scheds,
// Collect running containers from over-scheduled queues
List<RMContainer> runningContainers = new ArrayList<RMContainer>();
for (FSLeafQueue sched : scheds) {
if (Resources.greaterThan(sched.getResourceUsage(), sched.getFairShare())) {
if (Resources.greaterThan(RESOURCE_CALCULATOR, clusterCapacity,
sched.getResourceUsage(), sched.getFairShare())) {
for (AppSchedulable as : sched.getAppSchedulables()) {
for (RMContainer c : as.getApp().getLiveContainers()) {
runningContainers.add(c);
@ -332,7 +343,8 @@ public int compare(RMContainer c1, RMContainer c2) {
// tasks, making sure we don't kill too many from any queue
for (RMContainer container : runningContainers) {
FSLeafQueue sched = queues.get(container);
if (Resources.greaterThan(sched.getResourceUsage(), sched.getFairShare())) {
if (Resources.greaterThan(RESOURCE_CALCULATOR, clusterCapacity,
sched.getResourceUsage(), sched.getFairShare())) {
LOG.info("Preempting container (prio=" + container.getContainer().getPriority() +
"res=" + container.getContainer().getResource() +
") from queue " + sched.getName());
@ -345,7 +357,8 @@ public int compare(RMContainer c1, RMContainer c2) {
toPreempt = Resources.subtract(toPreempt,
container.getContainer().getResource());
if (Resources.equals(toPreempt, Resources.none())) {
if (Resources.lessThanOrEqual(RESOURCE_CALCULATOR, clusterCapacity,
toPreempt, Resources.none())) {
break;
}
}
@ -369,17 +382,21 @@ protected Resource resToPreempt(FSLeafQueue sched, long curTime) {
Resource resDueToMinShare = Resources.none();
Resource resDueToFairShare = Resources.none();
if (curTime - sched.getLastTimeAtMinShare() > minShareTimeout) {
Resource target = Resources.min(sched.getMinShare(), sched.getDemand());
resDueToMinShare = Resources.max(Resources.none(),
Resources.subtract(target, sched.getResourceUsage()));
Resource target = Resources.min(RESOURCE_CALCULATOR, clusterCapacity,
sched.getMinShare(), sched.getDemand());
resDueToMinShare = Resources.max(RESOURCE_CALCULATOR, clusterCapacity,
Resources.none(), Resources.subtract(target, sched.getResourceUsage()));
}
if (curTime - sched.getLastTimeAtHalfFairShare() > fairShareTimeout) {
Resource target = Resources.min(sched.getFairShare(), sched.getDemand());
resDueToFairShare = Resources.max(Resources.none(),
Resources.subtract(target, sched.getResourceUsage()));
Resource target = Resources.min(RESOURCE_CALCULATOR, clusterCapacity,
sched.getFairShare(), sched.getDemand());
resDueToFairShare = Resources.max(RESOURCE_CALCULATOR, clusterCapacity,
Resources.none(), Resources.subtract(target, sched.getResourceUsage()));
}
Resource resToPreempt = Resources.max(resDueToMinShare, resDueToFairShare);
if (Resources.greaterThan(resToPreempt, Resources.none())) {
Resource resToPreempt = Resources.max(RESOURCE_CALCULATOR, clusterCapacity,
resDueToMinShare, resDueToFairShare);
if (Resources.greaterThan(RESOURCE_CALCULATOR, clusterCapacity,
resToPreempt, Resources.none())) {
String message = "Should preempt " + resToPreempt + " res for queue "
+ sched.getName() + ": resDueToMinShare = " + resDueToMinShare
+ ", resDueToFairShare = " + resDueToFairShare;
@ -800,7 +817,7 @@ private synchronized void nodeUpdate(RMNode nm) {
int assignedContainers = 0;
while (node.getReservedContainer() == null) {
boolean assignedContainer = false;
if (Resources.greaterThan(
if (Resources.greaterThan(RESOURCE_CALCULATOR, clusterCapacity,
queueMgr.getRootQueue().assignContainer(node),
Resources.none())) {
assignedContainer = true;

View File

@ -42,6 +42,7 @@
import org.apache.hadoop.yarn.api.records.QueueACL;
import org.apache.hadoop.yarn.api.records.Resource;
import org.apache.hadoop.yarn.conf.YarnConfiguration;
import org.apache.hadoop.yarn.server.resourcemanager.resource.Resources;
import org.w3c.dom.Document;
import org.w3c.dom.Element;
import org.w3c.dom.Node;
@ -474,8 +475,8 @@ private void loadQueue(String parentName, Element element, Map<String, Resource>
}
queueAcls.put(queueName, acls);
if (maxQueueResources.containsKey(queueName) && minQueueResources.containsKey(queueName)
&& Resources.lessThan(maxQueueResources.get(queueName),
minQueueResources.get(queueName))) {
&& !Resources.fitsIn(minQueueResources.get(queueName),
maxQueueResources.get(queueName))) {
LOG.warn(String.format("Queue %s has max resources %d less than min resources %d",
queueName, maxQueueResources.get(queueName), minQueueResources.get(queueName)));
}
@ -504,7 +505,7 @@ public Resource getMaxResources(String queueName) {
if (maxQueueResource != null) {
return maxQueueResource;
} else {
return Resources.createResource(Integer.MAX_VALUE);
return Resources.createResource(Integer.MAX_VALUE, Integer.MAX_VALUE);
}
}

View File

@ -1,150 +0,0 @@
/**
* 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.scheduler.fair;
import org.apache.hadoop.classification.InterfaceAudience.Private;
import org.apache.hadoop.classification.InterfaceStability.Evolving;
import org.apache.hadoop.yarn.api.records.Resource;
import org.apache.hadoop.yarn.util.Records;
@Private
@Evolving
public class Resources {
// Java doesn't have const :(
private static final Resource NONE = new Resource() {
@Override
public int getMemory() {
return 0;
}
@Override
public void setMemory(int memory) {
throw new RuntimeException("NONE cannot be modified!");
}
@Override
public int getVirtualCores() {
return 0;
}
@Override
public void setVirtualCores(int cores) {
throw new RuntimeException("NONE cannot be modified!");
}
@Override
public int compareTo(Resource o) {
int diff = 0 - o.getMemory();
if (diff == 0) {
diff = 0 - o.getVirtualCores();
}
return diff;
}
};
public static Resource createResource(int memory) {
return createResource(memory, (memory > 0) ? 1 : 0);
}
public static Resource createResource(int memory, int cores) {
Resource resource = Records.newRecord(Resource.class);
resource.setMemory(memory);
resource.setVirtualCores(cores);
return resource;
}
public static Resource none() {
return NONE;
}
public static Resource clone(Resource res) {
return createResource(res.getMemory(), res.getVirtualCores());
}
public static Resource addTo(Resource lhs, Resource rhs) {
lhs.setMemory(lhs.getMemory() + rhs.getMemory());
return lhs;
}
public static Resource add(Resource lhs, Resource rhs) {
return addTo(clone(lhs), rhs);
}
public static Resource subtractFrom(Resource lhs, Resource rhs) {
lhs.setMemory(lhs.getMemory() - rhs.getMemory());
return lhs;
}
public static Resource subtract(Resource lhs, Resource rhs) {
return subtractFrom(clone(lhs), rhs);
}
public static Resource negate(Resource resource) {
return subtract(NONE, resource);
}
public static Resource multiplyTo(Resource lhs, int by) {
lhs.setMemory(lhs.getMemory() * by);
return lhs;
}
public static Resource multiply(Resource lhs, int by) {
return multiplyTo(clone(lhs), by);
}
/**
* Mutliply a resource by a {@code double}. Note that integral
* resource quantites are subject to rounding during cast.
*/
public static Resource multiply(Resource lhs, double by) {
Resource out = clone(lhs);
out.setMemory((int) (lhs.getMemory() * by));
return out;
}
public static boolean equals(Resource lhs, Resource rhs) {
return lhs.getMemory() == rhs.getMemory();
}
public static boolean lessThan(Resource lhs, Resource rhs) {
return lhs.getMemory() < rhs.getMemory();
}
public static boolean lessThanOrEqual(Resource lhs, Resource rhs) {
return lhs.getMemory() <= rhs.getMemory();
}
public static boolean greaterThan(Resource lhs, Resource rhs) {
return lhs.getMemory() > rhs.getMemory();
}
public static boolean greaterThanOrEqual(Resource lhs, Resource rhs) {
return lhs.getMemory() >= rhs.getMemory();
}
public static Resource min(Resource lhs, Resource rhs) {
return (lhs.getMemory() < rhs.getMemory()) ? lhs : rhs;
}
public static Resource max(Resource lhs, Resource rhs) {
return (lhs.getMemory() > rhs.getMemory()) ? lhs : rhs;
}}

View File

@ -22,7 +22,8 @@
import java.util.Comparator;
import org.apache.hadoop.yarn.api.records.Resource;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.Resources;
import org.apache.hadoop.yarn.server.resourcemanager.resource.DefaultResourceCalculator;
import org.apache.hadoop.yarn.server.resourcemanager.resource.Resources;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.Schedulable;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.SchedulingPolicy;
@ -31,6 +32,8 @@
public class FairSharePolicy extends SchedulingPolicy {
@VisibleForTesting
public static final String NAME = "Fairshare";
private static final DefaultResourceCalculator RESOURCE_CALCULATOR =
new DefaultResourceCalculator();
private FairShareComparator comparator = new FairShareComparator();
@Override
@ -59,15 +62,19 @@ private static class FairShareComparator implements Comparator<Schedulable>,
public int compare(Schedulable s1, Schedulable s2) {
double minShareRatio1, minShareRatio2;
double useToWeightRatio1, useToWeightRatio2;
Resource minShare1 = Resources.min(s1.getMinShare(), s1.getDemand());
Resource minShare2 = Resources.min(s2.getMinShare(), s2.getDemand());
boolean s1Needy = Resources.lessThan(s1.getResourceUsage(), minShare1);
boolean s2Needy = Resources.lessThan(s2.getResourceUsage(), minShare2);
Resource minShare1 = Resources.min(RESOURCE_CALCULATOR, null,
s1.getMinShare(), s1.getDemand());
Resource minShare2 = Resources.min(RESOURCE_CALCULATOR, null,
s2.getMinShare(), s2.getDemand());
boolean s1Needy = Resources.lessThan(RESOURCE_CALCULATOR, null,
s1.getResourceUsage(), minShare1);
boolean s2Needy = Resources.lessThan(RESOURCE_CALCULATOR, null,
s2.getResourceUsage(), minShare2);
Resource one = Resources.createResource(1);
minShareRatio1 = (double) s1.getResourceUsage().getMemory()
/ Resources.max(minShare1, one).getMemory();
/ Resources.max(RESOURCE_CALCULATOR, null, minShare1, one).getMemory();
minShareRatio2 = (double) s2.getResourceUsage().getMemory()
/ Resources.max(minShare2, one).getMemory();
/ Resources.max(RESOURCE_CALCULATOR, null, minShare2, one).getMemory();
useToWeightRatio1 = s1.getResourceUsage().getMemory() / s1.getWeight();
useToWeightRatio2 = s2.getResourceUsage().getMemory() / s2.getWeight();
int res = 0;
@ -161,9 +168,11 @@ public static void computeFairShares(
for (Schedulable sched : schedulables) {
Resources.addTo(totalDemand, sched.getDemand());
}
Resource cap = Resources.min(totalDemand, totalResources);
Resource cap = Resources.min(RESOURCE_CALCULATOR, null, totalDemand,
totalResources);
double rMax = 1.0;
while (Resources.lessThan(resUsedWithWeightToResRatio(rMax, schedulables),
while (Resources.lessThan(RESOURCE_CALCULATOR, null,
resUsedWithWeightToResRatio(rMax, schedulables),
cap)) {
rMax *= 2.0;
}
@ -172,7 +181,8 @@ public static void computeFairShares(
double right = rMax;
for (int i = 0; i < COMPUTE_FAIR_SHARES_ITERATIONS; i++) {
double mid = (left + right) / 2.0;
if (Resources.lessThan(resUsedWithWeightToResRatio(mid, schedulables),
if (Resources.lessThan(RESOURCE_CALCULATOR, null,
resUsedWithWeightToResRatio(mid, schedulables),
cap)) {
left = mid;
} else {

View File

@ -22,8 +22,8 @@
import java.util.Comparator;
import org.apache.hadoop.yarn.api.records.Resource;
import org.apache.hadoop.yarn.server.resourcemanager.resource.Resources;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.Schedulable;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.Resources;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.SchedulingPolicy;
import com.google.common.annotations.VisibleForTesting;

View File

@ -0,0 +1,43 @@
/**
* 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.resource;
import static org.apache.hadoop.yarn.server.resourcemanager.resource.Resources.*;
import static org.junit.Assert.*;
import org.junit.Test;
public class TestResources {
@Test(timeout=1000)
public void testFitsIn() {
assertTrue(fitsIn(createResource(1, 1), createResource(2, 2)));
assertTrue(fitsIn(createResource(2, 2), createResource(2, 2)));
assertFalse(fitsIn(createResource(2, 2), createResource(1, 1)));
assertFalse(fitsIn(createResource(1, 2), createResource(2, 1)));
assertFalse(fitsIn(createResource(2, 1), createResource(1, 2)));
}
@Test(timeout=1000)
public void testComponentwiseMin() {
assertEquals(createResource(1, 1),
componentwiseMin(createResource(1, 1), createResource(2, 2)));
assertEquals(createResource(1, 1),
componentwiseMin(createResource(2, 2), createResource(1, 1)));
assertEquals(createResource(1, 1),
componentwiseMin(createResource(1, 2), createResource(2, 1)));
}
}