Skip to content
Permalink
Fetching contributors…
Cannot retrieve contributors at this time
1140 lines (1032 sloc) 45.8 KB
/**
* 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 java.io.IOException;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashMap;
import java.util.HashSet;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import java.util.PriorityQueue;
import java.util.Set;
import java.util.TreeSet;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.yarn.api.records.ApplicationAttemptId;
import org.apache.hadoop.yarn.api.records.NodeId;
import org.apache.hadoop.yarn.api.records.Priority;
import org.apache.hadoop.yarn.api.records.Resource;
import org.apache.hadoop.yarn.exceptions.YarnRuntimeException;
import org.apache.hadoop.yarn.server.resourcemanager.RMContext;
import org.apache.hadoop.yarn.server.resourcemanager.monitor.SchedulingEditPolicy;
import org.apache.hadoop.yarn.server.resourcemanager.nodelabels.RMNodeLabelsManager;
import org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainer;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.ContainerPreemptEvent;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.PreemptableResourceScheduler;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CSQueue;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacitySchedulerConfiguration;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.LeafQueue;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.QueueCapacities;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.SchedulerEventType;
import org.apache.hadoop.yarn.util.Clock;
import org.apache.hadoop.yarn.util.SystemClock;
import org.apache.hadoop.yarn.util.resource.ResourceCalculator;
import org.apache.hadoop.yarn.util.resource.Resources;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.collect.ImmutableSet;
/**
* This class implement a {@link SchedulingEditPolicy} that is designed to be
* paired with the {@code CapacityScheduler}. At every invocation of {@code
* editSchedule()} it computes the ideal amount of resources assigned to each
* queue (for each queue in the hierarchy), and determines whether preemption
* is needed. Overcapacity is distributed among queues in a weighted fair manner,
* where the weight is the amount of guaranteed capacity for the queue.
* Based on this ideal assignment it determines whether preemption is required
* and select a set of containers from each application that would be killed if
* the corresponding amount of resources is not freed up by the application.
*
* If not in {@code observeOnly} mode, it triggers preemption requests via a
* {@link ContainerPreemptEvent} that the {@code ResourceManager} will ensure
* to deliver to the application (or to execute).
*
* If the deficit of resources is persistent over a long enough period of time
* this policy will trigger forced termination of containers (again by generating
* {@link ContainerPreemptEvent}).
*/
public class ProportionalCapacityPreemptionPolicy implements SchedulingEditPolicy {
private static final Log LOG =
LogFactory.getLog(ProportionalCapacityPreemptionPolicy.class);
/** If true, run the policy but do not affect the cluster with preemption and
* kill events. */
public static final String OBSERVE_ONLY =
"yarn.resourcemanager.monitor.capacity.preemption.observe_only";
/** Time in milliseconds between invocations of this policy */
public static final String MONITORING_INTERVAL =
"yarn.resourcemanager.monitor.capacity.preemption.monitoring_interval";
/** Time in milliseconds between requesting a preemption from an application
* and killing the container. */
public static final String WAIT_TIME_BEFORE_KILL =
"yarn.resourcemanager.monitor.capacity.preemption.max_wait_before_kill";
/** Maximum percentage of resources preempted in a single round. By
* controlling this value one can throttle the pace at which containers are
* reclaimed from the cluster. After computing the total desired preemption,
* the policy scales it back within this limit. */
public static final String TOTAL_PREEMPTION_PER_ROUND =
"yarn.resourcemanager.monitor.capacity.preemption.total_preemption_per_round";
/** Maximum amount of resources above the target capacity ignored for
* preemption. This defines a deadzone around the target capacity that helps
* prevent thrashing and oscillations around the computed target balance.
* High values would slow the time to capacity and (absent natural
* completions) it might prevent convergence to guaranteed capacity. */
public static final String MAX_IGNORED_OVER_CAPACITY =
"yarn.resourcemanager.monitor.capacity.preemption.max_ignored_over_capacity";
/**
* Given a computed preemption target, account for containers naturally
* expiring and preempt only this percentage of the delta. This determines
* the rate of geometric convergence into the deadzone ({@link
* #MAX_IGNORED_OVER_CAPACITY}). For example, a termination factor of 0.5
* will reclaim almost 95% of resources within 5 * {@link
* #WAIT_TIME_BEFORE_KILL}, even absent natural termination. */
public static final String NATURAL_TERMINATION_FACTOR =
"yarn.resourcemanager.monitor.capacity.preemption.natural_termination_factor";
private RMContext rmContext;
private final Clock clock;
private double maxIgnoredOverCapacity;
private long maxWaitTime;
private CapacityScheduler scheduler;
private long monitoringInterval;
private final Map<RMContainer,Long> preempted =
new HashMap<RMContainer,Long>();
private ResourceCalculator rc;
private float percentageClusterPreemptionAllowed;
private double naturalTerminationFactor;
private boolean observeOnly;
private Map<String, Map<String, TempQueuePerPartition>> queueToPartitions =
new HashMap<>();
private RMNodeLabelsManager nlm;
public ProportionalCapacityPreemptionPolicy() {
clock = new SystemClock();
}
public ProportionalCapacityPreemptionPolicy(Configuration config,
RMContext context, CapacityScheduler scheduler) {
this(config, context, scheduler, new SystemClock());
}
public ProportionalCapacityPreemptionPolicy(Configuration config,
RMContext context, CapacityScheduler scheduler, Clock clock) {
init(config, context, scheduler);
this.clock = clock;
}
public void init(Configuration config, RMContext context,
PreemptableResourceScheduler sched) {
LOG.info("Preemption monitor:" + this.getClass().getCanonicalName());
assert null == scheduler : "Unexpected duplicate call to init";
if (!(sched instanceof CapacityScheduler)) {
throw new YarnRuntimeException("Class " +
sched.getClass().getCanonicalName() + " not instance of " +
CapacityScheduler.class.getCanonicalName());
}
rmContext = context;
scheduler = (CapacityScheduler) sched;
maxIgnoredOverCapacity = config.getDouble(MAX_IGNORED_OVER_CAPACITY, 0.1);
naturalTerminationFactor =
config.getDouble(NATURAL_TERMINATION_FACTOR, 0.2);
maxWaitTime = config.getLong(WAIT_TIME_BEFORE_KILL, 15000);
monitoringInterval = config.getLong(MONITORING_INTERVAL, 3000);
percentageClusterPreemptionAllowed =
config.getFloat(TOTAL_PREEMPTION_PER_ROUND, (float) 0.1);
observeOnly = config.getBoolean(OBSERVE_ONLY, false);
rc = scheduler.getResourceCalculator();
nlm = scheduler.getRMContext().getNodeLabelManager();
}
@VisibleForTesting
public ResourceCalculator getResourceCalculator() {
return rc;
}
@Override
public void editSchedule() {
CSQueue root = scheduler.getRootQueue();
Resource clusterResources = Resources.clone(scheduler.getClusterResource());
containerBasedPreemptOrKill(root, clusterResources);
}
/**
* This method selects and tracks containers to be preempted. If a container
* is in the target list for more than maxWaitTime it is killed.
*
* @param root the root of the CapacityScheduler queue hierarchy
* @param clusterResources the total amount of resources in the cluster
*/
@SuppressWarnings("unchecked")
private void containerBasedPreemptOrKill(CSQueue root,
Resource clusterResources) {
// All partitions to look at
Set<String> allPartitions = new HashSet<>();
allPartitions.addAll(scheduler.getRMContext()
.getNodeLabelManager().getClusterNodeLabelNames());
allPartitions.add(RMNodeLabelsManager.NO_LABEL);
// extract a summary of the queues from scheduler
synchronized (scheduler) {
queueToPartitions.clear();
for (String partitionToLookAt : allPartitions) {
cloneQueues(root,
nlm.getResourceByLabel(partitionToLookAt, clusterResources),
partitionToLookAt);
}
}
// compute total preemption allowed
Resource totalPreemptionAllowed = Resources.multiply(clusterResources,
percentageClusterPreemptionAllowed);
Set<String> leafQueueNames = null;
for (String partition : allPartitions) {
TempQueuePerPartition tRoot =
getQueueByPartition(CapacitySchedulerConfiguration.ROOT, partition);
// compute the ideal distribution of resources among queues
// updates cloned queues state accordingly
tRoot.idealAssigned = tRoot.guaranteed;
leafQueueNames =
recursivelyComputeIdealAssignment(tRoot, totalPreemptionAllowed);
}
// based on ideal allocation select containers to be preempted from each
// queue and each application
Map<ApplicationAttemptId,Set<RMContainer>> toPreempt =
getContainersToPreempt(leafQueueNames, clusterResources);
if (LOG.isDebugEnabled()) {
logToCSV(new ArrayList<String>(leafQueueNames));
}
// if we are in observeOnly mode return before any action is taken
if (observeOnly) {
return;
}
// preempt (or kill) the selected containers
for (Map.Entry<ApplicationAttemptId,Set<RMContainer>> e
: toPreempt.entrySet()) {
ApplicationAttemptId appAttemptId = e.getKey();
if (LOG.isDebugEnabled()) {
LOG.debug("Send to scheduler: in app=" + appAttemptId
+ " #containers-to-be-preempted=" + e.getValue().size());
}
for (RMContainer container : e.getValue()) {
// if we tried to preempt this for more than maxWaitTime
if (preempted.get(container) != null &&
preempted.get(container) + maxWaitTime < clock.getTime()) {
// kill it
rmContext.getDispatcher().getEventHandler().handle(
new ContainerPreemptEvent(appAttemptId, container,
SchedulerEventType.KILL_CONTAINER));
preempted.remove(container);
} else {
if (preempted.get(container) != null) {
// We already updated the information to scheduler earlier, we need
// not have to raise another event.
continue;
}
//otherwise just send preemption events
rmContext.getDispatcher().getEventHandler().handle(
new ContainerPreemptEvent(appAttemptId, container,
SchedulerEventType.PREEMPT_CONTAINER));
preempted.put(container, clock.getTime());
}
}
}
// Keep the preempted list clean
for (Iterator<RMContainer> i = preempted.keySet().iterator(); i.hasNext();){
RMContainer id = i.next();
// garbage collect containers that are irrelevant for preemption
if (preempted.get(id) + 2 * maxWaitTime < clock.getTime()) {
i.remove();
}
}
}
/**
* This method recursively computes the ideal assignment of resources to each
* level of the hierarchy. This ensures that leafs that are over-capacity but
* with parents within capacity will not be preempted. Preemptions are allowed
* within each subtree according to local over/under capacity.
*
* @param root the root of the cloned queue hierachy
* @param totalPreemptionAllowed maximum amount of preemption allowed
* @return a list of leaf queues updated with preemption targets
*/
private Set<String> recursivelyComputeIdealAssignment(
TempQueuePerPartition root, Resource totalPreemptionAllowed) {
Set<String> leafQueueNames = new HashSet<>();
if (root.getChildren() != null &&
root.getChildren().size() > 0) {
// compute ideal distribution at this level
computeIdealResourceDistribution(rc, root.getChildren(),
totalPreemptionAllowed, root.idealAssigned);
// compute recursively for lower levels and build list of leafs
for(TempQueuePerPartition t : root.getChildren()) {
leafQueueNames.addAll(recursivelyComputeIdealAssignment(t,
totalPreemptionAllowed));
}
} else {
// we are in a leaf nothing to do, just return yourself
return ImmutableSet.of(root.queueName);
}
return leafQueueNames;
}
/**
* This method computes (for a single level in the tree, passed as a {@code
* List<TempQueue>}) the ideal assignment of resources. This is done
* recursively to allocate capacity fairly across all queues with pending
* demands. It terminates when no resources are left to assign, or when all
* demand is satisfied.
*
* @param rc resource calculator
* @param queues a list of cloned queues to be assigned capacity to (this is
* an out param)
* @param totalPreemptionAllowed total amount of preemption we allow
* @param tot_guarant the amount of capacity assigned to this pool of queues
*/
private void computeIdealResourceDistribution(ResourceCalculator rc,
List<TempQueuePerPartition> queues, Resource totalPreemptionAllowed,
Resource tot_guarant) {
// qAlloc tracks currently active queues (will decrease progressively as
// demand is met)
List<TempQueuePerPartition> qAlloc = new ArrayList<TempQueuePerPartition>(queues);
// unassigned tracks how much resources are still to assign, initialized
// with the total capacity for this set of queues
Resource unassigned = Resources.clone(tot_guarant);
// group queues based on whether they have non-zero guaranteed capacity
Set<TempQueuePerPartition> nonZeroGuarQueues = new HashSet<TempQueuePerPartition>();
Set<TempQueuePerPartition> zeroGuarQueues = new HashSet<TempQueuePerPartition>();
for (TempQueuePerPartition q : qAlloc) {
if (Resources
.greaterThan(rc, tot_guarant, q.guaranteed, Resources.none())) {
nonZeroGuarQueues.add(q);
} else {
zeroGuarQueues.add(q);
}
}
// first compute the allocation as a fixpoint based on guaranteed capacity
computeFixpointAllocation(rc, tot_guarant, nonZeroGuarQueues, unassigned,
false);
// if any capacity is left unassigned, distributed among zero-guarantee
// queues uniformly (i.e., not based on guaranteed capacity, as this is zero)
if (!zeroGuarQueues.isEmpty()
&& Resources.greaterThan(rc, tot_guarant, unassigned, Resources.none())) {
computeFixpointAllocation(rc, tot_guarant, zeroGuarQueues, unassigned,
true);
}
// based on ideal assignment computed above and current assignment we derive
// how much preemption is required overall
Resource totPreemptionNeeded = Resource.newInstance(0, 0);
for (TempQueuePerPartition t:queues) {
if (Resources.greaterThan(rc, tot_guarant, t.current, t.idealAssigned)) {
Resources.addTo(totPreemptionNeeded,
Resources.subtract(t.current, t.idealAssigned));
}
}
// if we need to preempt more than is allowed, compute a factor (0<f<1)
// that is used to scale down how much we ask back from each queue
float scalingFactor = 1.0F;
if (Resources.greaterThan(rc, tot_guarant,
totPreemptionNeeded, totalPreemptionAllowed)) {
scalingFactor = Resources.divide(rc, tot_guarant,
totalPreemptionAllowed, totPreemptionNeeded);
}
// assign to each queue the amount of actual preemption based on local
// information of ideal preemption and scaling factor
for (TempQueuePerPartition t : queues) {
t.assignPreemption(scalingFactor, rc, tot_guarant);
}
if (LOG.isDebugEnabled()) {
long time = clock.getTime();
for (TempQueuePerPartition t : queues) {
LOG.debug(time + ": " + t);
}
}
}
/**
* Given a set of queues compute the fix-point distribution of unassigned
* resources among them. As pending request of a queue are exhausted, the
* queue is removed from the set and remaining capacity redistributed among
* remaining queues. The distribution is weighted based on guaranteed
* capacity, unless asked to ignoreGuarantee, in which case resources are
* distributed uniformly.
*/
private void computeFixpointAllocation(ResourceCalculator rc,
Resource tot_guarant, Collection<TempQueuePerPartition> qAlloc,
Resource unassigned, boolean ignoreGuarantee) {
// Prior to assigning the unused resources, process each queue as follows:
// If current > guaranteed, idealAssigned = guaranteed + untouchable extra
// Else idealAssigned = current;
// Subtract idealAssigned resources from unassigned.
// If the queue has all of its needs met (that is, if
// idealAssigned >= current + pending), remove it from consideration.
// Sort queues from most under-guaranteed to most over-guaranteed.
TQComparator tqComparator = new TQComparator(rc, tot_guarant);
PriorityQueue<TempQueuePerPartition> orderedByNeed =
new PriorityQueue<TempQueuePerPartition>(10, tqComparator);
for (Iterator<TempQueuePerPartition> i = qAlloc.iterator(); i.hasNext();) {
TempQueuePerPartition q = i.next();
if (Resources.greaterThan(rc, tot_guarant, q.current, q.guaranteed)) {
q.idealAssigned = Resources.add(q.guaranteed, q.untouchableExtra);
} else {
q.idealAssigned = Resources.clone(q.current);
}
Resources.subtractFrom(unassigned, q.idealAssigned);
// If idealAssigned < (current + pending), q needs more resources, so
// add it to the list of underserved queues, ordered by need.
Resource curPlusPend = Resources.add(q.current, q.pending);
if (Resources.lessThan(rc, tot_guarant, q.idealAssigned, curPlusPend)) {
orderedByNeed.add(q);
}
}
//assign all cluster resources until no more demand, or no resources are left
while (!orderedByNeed.isEmpty()
&& Resources.greaterThan(rc,tot_guarant, unassigned,Resources.none())) {
Resource wQassigned = Resource.newInstance(0, 0);
// we compute normalizedGuarantees capacity based on currently active
// queues
resetCapacity(rc, unassigned, orderedByNeed, ignoreGuarantee);
// For each underserved queue (or set of queues if multiple are equally
// underserved), offer its share of the unassigned resources based on its
// normalized guarantee. After the offer, if the queue is not satisfied,
// place it back in the ordered list of queues, recalculating its place
// in the order of most under-guaranteed to most over-guaranteed. In this
// way, the most underserved queue(s) are always given resources first.
Collection<TempQueuePerPartition> underserved =
getMostUnderservedQueues(orderedByNeed, tqComparator);
for (Iterator<TempQueuePerPartition> i = underserved.iterator(); i
.hasNext();) {
TempQueuePerPartition sub = i.next();
Resource wQavail = Resources.multiplyAndNormalizeUp(rc,
unassigned, sub.normalizedGuarantee, Resource.newInstance(1, 1));
Resource wQidle = sub.offer(wQavail, rc, tot_guarant);
Resource wQdone = Resources.subtract(wQavail, wQidle);
if (Resources.greaterThan(rc, tot_guarant,
wQdone, Resources.none())) {
// The queue is still asking for more. Put it back in the priority
// queue, recalculating its order based on need.
orderedByNeed.add(sub);
}
Resources.addTo(wQassigned, wQdone);
}
Resources.subtractFrom(unassigned, wQassigned);
}
}
// Take the most underserved TempQueue (the one on the head). Collect and
// return the list of all queues that have the same idealAssigned
// percentage of guaranteed.
protected Collection<TempQueuePerPartition> getMostUnderservedQueues(
PriorityQueue<TempQueuePerPartition> orderedByNeed, TQComparator tqComparator) {
ArrayList<TempQueuePerPartition> underserved = new ArrayList<TempQueuePerPartition>();
while (!orderedByNeed.isEmpty()) {
TempQueuePerPartition q1 = orderedByNeed.remove();
underserved.add(q1);
TempQueuePerPartition q2 = orderedByNeed.peek();
// q1's pct of guaranteed won't be larger than q2's. If it's less, then
// return what has already been collected. Otherwise, q1's pct of
// guaranteed == that of q2, so add q2 to underserved list during the
// next pass.
if (q2 == null || tqComparator.compare(q1,q2) < 0) {
return underserved;
}
}
return underserved;
}
/**
* Computes a normalizedGuaranteed capacity based on active queues
* @param rc resource calculator
* @param clusterResource the total amount of resources in the cluster
* @param queues the list of queues to consider
*/
private void resetCapacity(ResourceCalculator rc, Resource clusterResource,
Collection<TempQueuePerPartition> queues, boolean ignoreGuar) {
Resource activeCap = Resource.newInstance(0, 0);
if (ignoreGuar) {
for (TempQueuePerPartition q : queues) {
q.normalizedGuarantee = (float) 1.0f / ((float) queues.size());
}
} else {
for (TempQueuePerPartition q : queues) {
Resources.addTo(activeCap, q.guaranteed);
}
for (TempQueuePerPartition q : queues) {
q.normalizedGuarantee = Resources.divide(rc, clusterResource,
q.guaranteed, activeCap);
}
}
}
private String getPartitionByNodeId(NodeId nodeId) {
return scheduler.getSchedulerNode(nodeId).getPartition();
}
/**
* Return should we preempt rmContainer. If we should, deduct from
* <code>resourceToObtainByPartition</code>
*/
private boolean tryPreemptContainerAndDeductResToObtain(
Map<String, Resource> resourceToObtainByPartitions,
RMContainer rmContainer, Resource clusterResource,
Map<ApplicationAttemptId, Set<RMContainer>> preemptMap) {
ApplicationAttemptId attemptId = rmContainer.getApplicationAttemptId();
// We will not account resource of a container twice or more
if (preemptMapContains(preemptMap, attemptId, rmContainer)) {
return false;
}
String nodePartition = getPartitionByNodeId(rmContainer.getAllocatedNode());
Resource toObtainByPartition =
resourceToObtainByPartitions.get(nodePartition);
if (null != toObtainByPartition
&& Resources.greaterThan(rc, clusterResource, toObtainByPartition,
Resources.none())) {
Resources.subtractFrom(toObtainByPartition,
rmContainer.getAllocatedResource());
// When we have no more resource need to obtain, remove from map.
if (Resources.lessThanOrEqual(rc, clusterResource, toObtainByPartition,
Resources.none())) {
resourceToObtainByPartitions.remove(nodePartition);
}
if (LOG.isDebugEnabled()) {
LOG.debug("Marked container=" + rmContainer.getContainerId()
+ " in partition=" + nodePartition + " will be preempted");
}
// Add to preemptMap
addToPreemptMap(preemptMap, attemptId, rmContainer);
return true;
}
return false;
}
private boolean preemptMapContains(
Map<ApplicationAttemptId, Set<RMContainer>> preemptMap,
ApplicationAttemptId attemptId, RMContainer rmContainer) {
Set<RMContainer> rmContainers;
if (null == (rmContainers = preemptMap.get(attemptId))) {
return false;
}
return rmContainers.contains(rmContainer);
}
private void addToPreemptMap(
Map<ApplicationAttemptId, Set<RMContainer>> preemptMap,
ApplicationAttemptId appAttemptId, RMContainer containerToPreempt) {
Set<RMContainer> set;
if (null == (set = preemptMap.get(appAttemptId))) {
set = new HashSet<RMContainer>();
preemptMap.put(appAttemptId, set);
}
set.add(containerToPreempt);
}
/**
* Based a resource preemption target drop reservations of containers and
* if necessary select containers for preemption from applications in each
* over-capacity queue. It uses {@link #NATURAL_TERMINATION_FACTOR} to
* account for containers that will naturally complete.
*
* @param queues set of leaf queues to preempt from
* @param clusterResource total amount of cluster resources
* @return a map of applciationID to set of containers to preempt
*/
private Map<ApplicationAttemptId,Set<RMContainer>> getContainersToPreempt(
Set<String> leafQueueNames, Resource clusterResource) {
Map<ApplicationAttemptId, Set<RMContainer>> preemptMap =
new HashMap<ApplicationAttemptId, Set<RMContainer>>();
List<RMContainer> skippedAMContainerlist = new ArrayList<RMContainer>();
// Loop all leaf queues
for (String queueName : leafQueueNames) {
// check if preemption disabled for the queue
if (getQueueByPartition(queueName,
RMNodeLabelsManager.NO_LABEL).preemptionDisabled) {
if (LOG.isDebugEnabled()) {
LOG.debug("skipping from queue=" + queueName
+ " because it's a non-preemptable queue");
}
continue;
}
// compute resToObtainByPartition considered inter-queue preemption
LeafQueue leafQueue = null;
Map<String, Resource> resToObtainByPartition =
new HashMap<String, Resource>();
for (TempQueuePerPartition qT : getQueuePartitions(queueName)) {
leafQueue = qT.leafQueue;
// we act only if we are violating balance by more than
// maxIgnoredOverCapacity
if (Resources.greaterThan(rc, clusterResource, qT.current,
Resources.multiply(qT.guaranteed, 1.0 + maxIgnoredOverCapacity))) {
// we introduce a dampening factor naturalTerminationFactor that
// accounts for natural termination of containers
Resource resToObtain =
Resources.multiply(qT.toBePreempted, naturalTerminationFactor);
// Only add resToObtain when it >= 0
if (Resources.greaterThan(rc, clusterResource, resToObtain,
Resources.none())) {
resToObtainByPartition.put(qT.partition, resToObtain);
if (LOG.isDebugEnabled()) {
LOG.debug("Queue=" + queueName + " partition=" + qT.partition
+ " resource-to-obtain=" + resToObtain);
}
}
qT.actuallyPreempted = Resources.clone(resToObtain);
} else {
qT.actuallyPreempted = Resources.none();
}
}
synchronized (leafQueue) {
// go through all ignore-partition-exclusivity containers first to make
// sure such containers will be preempted first
Map<String, TreeSet<RMContainer>> ignorePartitionExclusivityContainers =
leafQueue.getIgnoreExclusivityRMContainers();
for (String partition : resToObtainByPartition.keySet()) {
if (ignorePartitionExclusivityContainers.containsKey(partition)) {
TreeSet<RMContainer> rmContainers =
ignorePartitionExclusivityContainers.get(partition);
// We will check container from reverse order, so latter submitted
// application's containers will be preempted first.
for (RMContainer c : rmContainers.descendingSet()) {
boolean preempted =
tryPreemptContainerAndDeductResToObtain(
resToObtainByPartition, c, clusterResource, preemptMap);
if (!preempted) {
break;
}
}
}
}
// preempt other containers
Resource skippedAMSize = Resource.newInstance(0, 0);
Iterator<FiCaSchedulerApp> desc =
leafQueue.getOrderingPolicy().getPreemptionIterator();
while (desc.hasNext()) {
FiCaSchedulerApp fc = desc.next();
// When we complete preempt from one partition, we will remove from
// resToObtainByPartition, so when it becomes empty, we can get no
// more preemption is needed
if (resToObtainByPartition.isEmpty()) {
break;
}
preemptFrom(fc, clusterResource, resToObtainByPartition,
skippedAMContainerlist, skippedAMSize, preemptMap);
}
// Can try preempting AMContainers (still saving atmost
// maxAMCapacityForThisQueue AMResource's) if more resources are
// required to be preempted from this Queue.
Resource maxAMCapacityForThisQueue = Resources.multiply(
Resources.multiply(clusterResource,
leafQueue.getAbsoluteCapacity()),
leafQueue.getMaxAMResourcePerQueuePercent());
preemptAMContainers(clusterResource, preemptMap, skippedAMContainerlist,
resToObtainByPartition, skippedAMSize, maxAMCapacityForThisQueue);
}
}
return preemptMap;
}
/**
* As more resources are needed for preemption, saved AMContainers has to be
* rescanned. Such AMContainers can be preempted based on resToObtain, but
* maxAMCapacityForThisQueue resources will be still retained.
*
* @param clusterResource
* @param preemptMap
* @param skippedAMContainerlist
* @param resToObtain
* @param skippedAMSize
* @param maxAMCapacityForThisQueue
*/
private void preemptAMContainers(Resource clusterResource,
Map<ApplicationAttemptId, Set<RMContainer>> preemptMap,
List<RMContainer> skippedAMContainerlist,
Map<String, Resource> resToObtainByPartition, Resource skippedAMSize,
Resource maxAMCapacityForThisQueue) {
for (RMContainer c : skippedAMContainerlist) {
// Got required amount of resources for preemption, can stop now
if (resToObtainByPartition.isEmpty()) {
break;
}
// Once skippedAMSize reaches down to maxAMCapacityForThisQueue,
// container selection iteration for preemption will be stopped.
if (Resources.lessThanOrEqual(rc, clusterResource, skippedAMSize,
maxAMCapacityForThisQueue)) {
break;
}
boolean preempted =
tryPreemptContainerAndDeductResToObtain(resToObtainByPartition, c,
clusterResource, preemptMap);
if (preempted) {
Resources.subtractFrom(skippedAMSize, c.getAllocatedResource());
}
}
skippedAMContainerlist.clear();
}
/**
* Given a target preemption for a specific application, select containers
* to preempt (after unreserving all reservation for that app).
*/
@SuppressWarnings("unchecked")
private void preemptFrom(FiCaSchedulerApp app,
Resource clusterResource, Map<String, Resource> resToObtainByPartition,
List<RMContainer> skippedAMContainerlist, Resource skippedAMSize,
Map<ApplicationAttemptId, Set<RMContainer>> preemptMap) {
ApplicationAttemptId appId = app.getApplicationAttemptId();
if (LOG.isDebugEnabled()) {
LOG.debug("Looking at application=" + app.getApplicationAttemptId()
+ " resourceToObtain=" + resToObtainByPartition);
}
// first drop reserved containers towards rsrcPreempt
List<RMContainer> reservedContainers =
new ArrayList<RMContainer>(app.getReservedContainers());
for (RMContainer c : reservedContainers) {
if (resToObtainByPartition.isEmpty()) {
return;
}
// Try to preempt this container
tryPreemptContainerAndDeductResToObtain(resToObtainByPartition, c,
clusterResource, preemptMap);
if (!observeOnly) {
rmContext.getDispatcher().getEventHandler().handle(
new ContainerPreemptEvent(
appId, c, SchedulerEventType.DROP_RESERVATION));
}
}
// if more resources are to be freed go through all live containers in
// reverse priority and reverse allocation order and mark them for
// preemption
List<RMContainer> liveContainers =
new ArrayList<RMContainer>(app.getLiveContainers());
sortContainers(liveContainers);
for (RMContainer c : liveContainers) {
if (resToObtainByPartition.isEmpty()) {
return;
}
// Skip AM Container from preemption for now.
if (c.isAMContainer()) {
skippedAMContainerlist.add(c);
Resources.addTo(skippedAMSize, c.getAllocatedResource());
continue;
}
// Try to preempt this container
tryPreemptContainerAndDeductResToObtain(resToObtainByPartition, c,
clusterResource, preemptMap);
}
}
/**
* Compare by reversed priority order first, and then reversed containerId
* order
* @param containers
*/
@VisibleForTesting
static void sortContainers(List<RMContainer> containers){
Collections.sort(containers, new Comparator<RMContainer>() {
@Override
public int compare(RMContainer a, RMContainer b) {
Comparator<Priority> c = new org.apache.hadoop.yarn.server
.resourcemanager.resource.Priority.Comparator();
int priorityComp = c.compare(b.getContainer().getPriority(),
a.getContainer().getPriority());
if (priorityComp != 0) {
return priorityComp;
}
return b.getContainerId().compareTo(a.getContainerId());
}
});
}
@Override
public long getMonitoringInterval() {
return monitoringInterval;
}
@Override
public String getPolicyName() {
return "ProportionalCapacityPreemptionPolicy";
}
/**
* This method walks a tree of CSQueue and clones the portion of the state
* relevant for preemption in TempQueue(s). It also maintains a pointer to
* the leaves. Finally it aggregates pending resources in each queue and rolls
* it up to higher levels.
*
* @param curQueue current queue which I'm looking at now
* @param partitionResource the total amount of resources in the cluster
* @return the root of the cloned queue hierarchy
*/
private TempQueuePerPartition cloneQueues(CSQueue curQueue,
Resource partitionResource, String partitionToLookAt) {
TempQueuePerPartition ret;
synchronized (curQueue) {
String queueName = curQueue.getQueueName();
QueueCapacities qc = curQueue.getQueueCapacities();
float absCap = qc.getAbsoluteCapacity(partitionToLookAt);
float absMaxCap = qc.getAbsoluteMaximumCapacity(partitionToLookAt);
boolean preemptionDisabled = curQueue.getPreemptionDisabled();
Resource current = curQueue.getQueueResourceUsage().getUsed(
partitionToLookAt);
Resource guaranteed = Resources.multiply(partitionResource, absCap);
Resource maxCapacity = Resources.multiply(partitionResource, absMaxCap);
// when partition is a non-exclusive partition, the actual maxCapacity
// could more than specified maxCapacity
try {
if (!scheduler.getRMContext().getNodeLabelManager()
.isExclusiveNodeLabel(partitionToLookAt)) {
maxCapacity =
Resources.max(rc, partitionResource, maxCapacity, current);
}
} catch (IOException e) {
// This may cause by partition removed when running capacity monitor,
// just ignore the error, this will be corrected when doing next check.
}
Resource extra = Resource.newInstance(0, 0);
if (Resources.greaterThan(rc, partitionResource, current, guaranteed)) {
extra = Resources.subtract(current, guaranteed);
}
if (curQueue instanceof LeafQueue) {
LeafQueue l = (LeafQueue) curQueue;
Resource pending =
l.getTotalPendingResourcesConsideringUserLimit(
partitionResource, partitionToLookAt);
ret = new TempQueuePerPartition(queueName, current, pending, guaranteed,
maxCapacity, preemptionDisabled, partitionToLookAt);
if (preemptionDisabled) {
ret.untouchableExtra = extra;
} else {
ret.preemptableExtra = extra;
}
ret.setLeafQueue(l);
} else {
Resource pending = Resource.newInstance(0, 0);
ret =
new TempQueuePerPartition(curQueue.getQueueName(), current, pending,
guaranteed, maxCapacity, false, partitionToLookAt);
Resource childrensPreemptable = Resource.newInstance(0, 0);
for (CSQueue c : curQueue.getChildQueues()) {
TempQueuePerPartition subq =
cloneQueues(c, partitionResource, partitionToLookAt);
Resources.addTo(childrensPreemptable, subq.preemptableExtra);
ret.addChild(subq);
}
// untouchableExtra = max(extra - childrenPreemptable, 0)
if (Resources.greaterThanOrEqual(
rc, partitionResource, childrensPreemptable, extra)) {
ret.untouchableExtra = Resource.newInstance(0, 0);
} else {
ret.untouchableExtra =
Resources.subtract(extra, childrensPreemptable);
}
ret.preemptableExtra = Resources.min(
rc, partitionResource, childrensPreemptable, extra);
}
}
addTempQueuePartition(ret);
return ret;
}
// simple printout function that reports internal queue state (useful for
// plotting)
private void logToCSV(List<String> leafQueueNames){
Collections.sort(leafQueueNames);
String queueState = " QUEUESTATE: " + clock.getTime();
StringBuilder sb = new StringBuilder();
sb.append(queueState);
for (String queueName : leafQueueNames) {
TempQueuePerPartition tq =
getQueueByPartition(queueName, RMNodeLabelsManager.NO_LABEL);
sb.append(", ");
tq.appendLogString(sb);
}
LOG.debug(sb.toString());
}
private void addTempQueuePartition(TempQueuePerPartition queuePartition) {
String queueName = queuePartition.queueName;
Map<String, TempQueuePerPartition> queuePartitions;
if (null == (queuePartitions = queueToPartitions.get(queueName))) {
queuePartitions = new HashMap<String, TempQueuePerPartition>();
queueToPartitions.put(queueName, queuePartitions);
}
queuePartitions.put(queuePartition.partition, queuePartition);
}
/**
* Get queue partition by given queueName and partitionName
*/
private TempQueuePerPartition getQueueByPartition(String queueName,
String partition) {
Map<String, TempQueuePerPartition> partitionToQueues = null;
if (null == (partitionToQueues = queueToPartitions.get(queueName))) {
return null;
}
return partitionToQueues.get(partition);
}
/**
* Get all queue partitions by given queueName
*/
private Collection<TempQueuePerPartition> getQueuePartitions(String queueName) {
if (!queueToPartitions.containsKey(queueName)) {
return null;
}
return queueToPartitions.get(queueName).values();
}
/**
* Temporary data-structure tracking resource availability, pending resource
* need, current utilization. This is per-queue-per-partition data structure
*/
static class TempQueuePerPartition {
final String queueName;
final Resource current;
final Resource pending;
final Resource guaranteed;
final Resource maxCapacity;
final String partition;
Resource idealAssigned;
Resource toBePreempted;
// For logging purpose
Resource actuallyPreempted;
Resource untouchableExtra;
Resource preemptableExtra;
double normalizedGuarantee;
final ArrayList<TempQueuePerPartition> children;
LeafQueue leafQueue;
boolean preemptionDisabled;
TempQueuePerPartition(String queueName, Resource current, Resource pending,
Resource guaranteed, Resource maxCapacity, boolean preemptionDisabled,
String partition) {
this.queueName = queueName;
this.current = current;
this.pending = pending;
this.guaranteed = guaranteed;
this.maxCapacity = maxCapacity;
this.idealAssigned = Resource.newInstance(0, 0);
this.actuallyPreempted = Resource.newInstance(0, 0);
this.toBePreempted = Resource.newInstance(0, 0);
this.normalizedGuarantee = Float.NaN;
this.children = new ArrayList<TempQueuePerPartition>();
this.untouchableExtra = Resource.newInstance(0, 0);
this.preemptableExtra = Resource.newInstance(0, 0);
this.preemptionDisabled = preemptionDisabled;
this.partition = partition;
}
public void setLeafQueue(LeafQueue l){
assert children.size() == 0;
this.leafQueue = l;
}
/**
* When adding a child we also aggregate its pending resource needs.
* @param q the child queue to add to this queue
*/
public void addChild(TempQueuePerPartition q) {
assert leafQueue == null;
children.add(q);
Resources.addTo(pending, q.pending);
}
public void addChildren(ArrayList<TempQueuePerPartition> queues) {
assert leafQueue == null;
children.addAll(queues);
}
public ArrayList<TempQueuePerPartition> getChildren(){
return children;
}
// This function "accepts" all the resources it can (pending) and return
// the unused ones
Resource offer(Resource avail, ResourceCalculator rc,
Resource clusterResource) {
Resource absMaxCapIdealAssignedDelta = Resources.componentwiseMax(
Resources.subtract(maxCapacity, idealAssigned),
Resource.newInstance(0, 0));
// remain = avail - min(avail, (max - assigned), (current + pending - assigned))
Resource accepted =
Resources.min(rc, clusterResource,
absMaxCapIdealAssignedDelta,
Resources.min(rc, clusterResource, avail, Resources.subtract(
Resources.add(current, pending), idealAssigned)));
Resource remain = Resources.subtract(avail, accepted);
Resources.addTo(idealAssigned, accepted);
return remain;
}
@Override
public String toString() {
StringBuilder sb = new StringBuilder();
sb.append(" NAME: " + queueName)
.append(" CUR: ").append(current)
.append(" PEN: ").append(pending)
.append(" GAR: ").append(guaranteed)
.append(" NORM: ").append(normalizedGuarantee)
.append(" IDEAL_ASSIGNED: ").append(idealAssigned)
.append(" IDEAL_PREEMPT: ").append(toBePreempted)
.append(" ACTUAL_PREEMPT: ").append(actuallyPreempted)
.append(" UNTOUCHABLE: ").append(untouchableExtra)
.append(" PREEMPTABLE: ").append(preemptableExtra)
.append("\n");
return sb.toString();
}
public void printAll() {
LOG.info(this.toString());
for (TempQueuePerPartition sub : this.getChildren()) {
sub.printAll();
}
}
public void assignPreemption(float scalingFactor,
ResourceCalculator rc, Resource clusterResource) {
if (Resources.greaterThan(rc, clusterResource, current, idealAssigned)) {
toBePreempted = Resources.multiply(
Resources.subtract(current, idealAssigned), scalingFactor);
} else {
toBePreempted = Resource.newInstance(0, 0);
}
}
void appendLogString(StringBuilder sb) {
sb.append(queueName).append(", ")
.append(current.getMemory()).append(", ")
.append(current.getVirtualCores()).append(", ")
.append(pending.getMemory()).append(", ")
.append(pending.getVirtualCores()).append(", ")
.append(guaranteed.getMemory()).append(", ")
.append(guaranteed.getVirtualCores()).append(", ")
.append(idealAssigned.getMemory()).append(", ")
.append(idealAssigned.getVirtualCores()).append(", ")
.append(toBePreempted.getMemory()).append(", ")
.append(toBePreempted.getVirtualCores() ).append(", ")
.append(actuallyPreempted.getMemory()).append(", ")
.append(actuallyPreempted.getVirtualCores());
}
}
static class TQComparator implements Comparator<TempQueuePerPartition> {
private ResourceCalculator rc;
private Resource clusterRes;
TQComparator(ResourceCalculator rc, Resource clusterRes) {
this.rc = rc;
this.clusterRes = clusterRes;
}
@Override
public int compare(TempQueuePerPartition tq1, TempQueuePerPartition tq2) {
if (getIdealPctOfGuaranteed(tq1) < getIdealPctOfGuaranteed(tq2)) {
return -1;
}
if (getIdealPctOfGuaranteed(tq1) > getIdealPctOfGuaranteed(tq2)) {
return 1;
}
return 0;
}
// Calculates idealAssigned / guaranteed
// TempQueues with 0 guarantees are always considered the most over
// capacity and therefore considered last for resources.
private double getIdealPctOfGuaranteed(TempQueuePerPartition q) {
double pctOver = Integer.MAX_VALUE;
if (q != null && Resources.greaterThan(
rc, clusterRes, q.guaranteed, Resources.none())) {
pctOver =
Resources.divide(rc, clusterRes, q.idealAssigned, q.guaranteed);
}
return (pctOver);
}
}
@VisibleForTesting
public Map<String, Map<String, TempQueuePerPartition>> getQueuePartitions() {
return queueToPartitions;
}
}
You can’t perform that action at this time.