HDFS-14403. Cost-based extension to the RPC Fair Call Queue. Contributed by Christopher Gregorian.
This commit is contained in:
Christopher Gregorian
Erik Krogen
parent
d023f1f864
commit
129576f628
@ -106,6 +106,7 @@ public class CommonConfigurationKeys extends CommonConfigurationKeysPublic {
|
||||
public static final String IPC_CALLQUEUE_IMPL_KEY = "callqueue.impl";
|
||||
public static final String IPC_SCHEDULER_IMPL_KEY = "scheduler.impl";
|
||||
public static final String IPC_IDENTITY_PROVIDER_KEY = "identity-provider.impl";
|
||||
public static final String IPC_COST_PROVIDER_KEY = "cost-provider.impl";
|
||||
public static final String IPC_BACKOFF_ENABLE = "backoff.enable";
|
||||
public static final boolean IPC_BACKOFF_ENABLE_DEFAULT = false;
|
||||
|
||||
|
||||
@ -198,7 +198,6 @@ void addResponseTime(String name, Schedulable e, ProcessingDetails details) {
|
||||
}
|
||||
|
||||
// This should be only called once per call and cached in the call object
|
||||
// each getPriorityLevel call will increment the counter for the caller
|
||||
int getPriorityLevel(Schedulable e) {
|
||||
return scheduler.getPriorityLevel(e);
|
||||
}
|
||||
|
||||
@ -0,0 +1,46 @@
|
||||
/**
|
||||
* 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.ipc;
|
||||
|
||||
import org.apache.hadoop.conf.Configuration;
|
||||
|
||||
/**
|
||||
* Used by {@link DecayRpcScheduler} to get the cost of users' operations. This
|
||||
* is configurable using
|
||||
* {@link org.apache.hadoop.fs.CommonConfigurationKeys#IPC_COST_PROVIDER_KEY}.
|
||||
*/
|
||||
public interface CostProvider {
|
||||
|
||||
/**
|
||||
* Initialize this provider using the given configuration, examining only
|
||||
* ones which fall within the provided namespace.
|
||||
*
|
||||
* @param namespace The namespace to use when looking up configurations.
|
||||
* @param conf The configuration
|
||||
*/
|
||||
void init(String namespace, Configuration conf);
|
||||
|
||||
/**
|
||||
* Get cost from {@link ProcessingDetails} which will be used in scheduler.
|
||||
*
|
||||
* @param details Process details
|
||||
* @return The cost of the call
|
||||
*/
|
||||
long getCost(ProcessingDetails details);
|
||||
}
|
||||
@ -58,8 +58,8 @@
|
||||
import static org.apache.hadoop.ipc.ProcessingDetails.Timing;
|
||||
|
||||
/**
|
||||
* The decay RPC scheduler counts incoming requests in a map, then
|
||||
* decays the counts at a fixed time interval. The scheduler is optimized
|
||||
* The decay RPC scheduler tracks the cost of incoming requests in a map, then
|
||||
* decays the costs at a fixed time interval. The scheduler is optimized
|
||||
* for large periods (on the order of seconds), as it offloads work to the
|
||||
* decay sweep.
|
||||
*/
|
||||
@ -77,7 +77,7 @@ public class DecayRpcScheduler implements RpcScheduler,
|
||||
"faircallqueue.decay-scheduler.period-ms";
|
||||
|
||||
/**
|
||||
* Decay factor controls how much each count is suppressed by on each sweep.
|
||||
* Decay factor controls how much each cost is suppressed by on each sweep.
|
||||
* Valid numbers are > 0 and < 1. Decay factor works in tandem with
|
||||
* period
|
||||
* to control how long the scheduler remembers an identity.
|
||||
@ -135,15 +135,15 @@ public class DecayRpcScheduler implements RpcScheduler,
|
||||
private static final ObjectWriter WRITER = new ObjectMapper().writer();
|
||||
|
||||
// Track the decayed and raw (no decay) number of calls for each schedulable
|
||||
// identity from all previous decay windows: idx 0 for decayed call count and
|
||||
// idx 1 for the raw call count
|
||||
private final ConcurrentHashMap<Object, List<AtomicLong>> callCounts =
|
||||
// identity from all previous decay windows: idx 0 for decayed call cost and
|
||||
// idx 1 for the raw call cost
|
||||
private final ConcurrentHashMap<Object, List<AtomicLong>> callCosts =
|
||||
new ConcurrentHashMap<Object, List<AtomicLong>>();
|
||||
|
||||
// Should be the sum of all AtomicLongs in decayed callCounts
|
||||
private final AtomicLong totalDecayedCallCount = new AtomicLong();
|
||||
// The sum of all AtomicLongs in raw callCounts
|
||||
private final AtomicLong totalRawCallCount = new AtomicLong();
|
||||
// Should be the sum of all AtomicLongs in decayed callCosts
|
||||
private final AtomicLong totalDecayedCallCost = new AtomicLong();
|
||||
// The sum of all AtomicLongs in raw callCosts
|
||||
private final AtomicLong totalRawCallCost = new AtomicLong();
|
||||
|
||||
|
||||
// Track total call count and response time in current decay window
|
||||
@ -161,7 +161,7 @@ public class DecayRpcScheduler implements RpcScheduler,
|
||||
|
||||
// Tune the behavior of the scheduler
|
||||
private final long decayPeriodMillis; // How long between each tick
|
||||
private final double decayFactor; // nextCount = currentCount * decayFactor
|
||||
private final double decayFactor; // nextCost = currentCost * decayFactor
|
||||
private final int numLevels;
|
||||
private final double[] thresholds;
|
||||
private final IdentityProvider identityProvider;
|
||||
@ -171,9 +171,10 @@ public class DecayRpcScheduler implements RpcScheduler,
|
||||
private final int topUsersCount; // e.g., report top 10 users' metrics
|
||||
private static final double PRECISION = 0.0001;
|
||||
private MetricsProxy metricsProxy;
|
||||
private final CostProvider costProvider;
|
||||
|
||||
/**
|
||||
* This TimerTask will call decayCurrentCounts until
|
||||
* This TimerTask will call decayCurrentCosts until
|
||||
* the scheduler has been garbage collected.
|
||||
*/
|
||||
public static class DecayTask extends TimerTask {
|
||||
@ -189,7 +190,7 @@ public DecayTask(DecayRpcScheduler scheduler, Timer timer) {
|
||||
public void run() {
|
||||
DecayRpcScheduler sched = schedulerRef.get();
|
||||
if (sched != null) {
|
||||
sched.decayCurrentCounts();
|
||||
sched.decayCurrentCosts();
|
||||
} else {
|
||||
// Our scheduler was garbage collected since it is no longer in use,
|
||||
// so we should terminate the timer as well
|
||||
@ -216,6 +217,7 @@ public DecayRpcScheduler(int numLevels, String ns, Configuration conf) {
|
||||
this.decayFactor = parseDecayFactor(ns, conf);
|
||||
this.decayPeriodMillis = parseDecayPeriodMillis(ns, conf);
|
||||
this.identityProvider = this.parseIdentityProvider(ns, conf);
|
||||
this.costProvider = this.parseCostProvider(ns, conf);
|
||||
this.thresholds = parseThresholds(ns, conf, numLevels);
|
||||
this.backOffByResponseTimeEnabled = parseBackOffByResponseTimeEnabled(ns,
|
||||
conf);
|
||||
@ -243,6 +245,24 @@ public DecayRpcScheduler(int numLevels, String ns, Configuration conf) {
|
||||
recomputeScheduleCache();
|
||||
}
|
||||
|
||||
private CostProvider parseCostProvider(String ns, Configuration conf) {
|
||||
List<CostProvider> providers = conf.getInstances(
|
||||
ns + "." + CommonConfigurationKeys.IPC_COST_PROVIDER_KEY,
|
||||
CostProvider.class);
|
||||
|
||||
if (providers.size() < 1) {
|
||||
LOG.info("CostProvider not specified, defaulting to DefaultCostProvider");
|
||||
return new DefaultCostProvider();
|
||||
} else if (providers.size() > 1) {
|
||||
LOG.warn("Found multiple CostProviders; using: {}",
|
||||
providers.get(0).getClass());
|
||||
}
|
||||
|
||||
CostProvider provider = providers.get(0); // use the first
|
||||
provider.init(ns, conf);
|
||||
return provider;
|
||||
}
|
||||
|
||||
// Load configs
|
||||
private IdentityProvider parseIdentityProvider(String ns,
|
||||
Configuration conf) {
|
||||
@ -389,69 +409,69 @@ private static Boolean parseBackOffByResponseTimeEnabled(String ns,
|
||||
}
|
||||
|
||||
/**
|
||||
* Decay the stored counts for each user and clean as necessary.
|
||||
* Decay the stored costs for each user and clean as necessary.
|
||||
* This method should be called periodically in order to keep
|
||||
* counts current.
|
||||
* costs current.
|
||||
*/
|
||||
private void decayCurrentCounts() {
|
||||
LOG.debug("Start to decay current counts.");
|
||||
private void decayCurrentCosts() {
|
||||
LOG.debug("Start to decay current costs.");
|
||||
try {
|
||||
long totalDecayedCount = 0;
|
||||
long totalRawCount = 0;
|
||||
long totalDecayedCost = 0;
|
||||
long totalRawCost = 0;
|
||||
Iterator<Map.Entry<Object, List<AtomicLong>>> it =
|
||||
callCounts.entrySet().iterator();
|
||||
callCosts.entrySet().iterator();
|
||||
|
||||
while (it.hasNext()) {
|
||||
Map.Entry<Object, List<AtomicLong>> entry = it.next();
|
||||
AtomicLong decayedCount = entry.getValue().get(0);
|
||||
AtomicLong rawCount = entry.getValue().get(1);
|
||||
AtomicLong decayedCost = entry.getValue().get(0);
|
||||
AtomicLong rawCost = entry.getValue().get(1);
|
||||
|
||||
|
||||
// Compute the next value by reducing it by the decayFactor
|
||||
totalRawCount += rawCount.get();
|
||||
long currentValue = decayedCount.get();
|
||||
totalRawCost += rawCost.get();
|
||||
long currentValue = decayedCost.get();
|
||||
long nextValue = (long) (currentValue * decayFactor);
|
||||
totalDecayedCount += nextValue;
|
||||
decayedCount.set(nextValue);
|
||||
totalDecayedCost += nextValue;
|
||||
decayedCost.set(nextValue);
|
||||
|
||||
LOG.debug("Decaying counts for the user: {}, " +
|
||||
"its decayedCount: {}, rawCount: {}", entry.getKey(),
|
||||
nextValue, rawCount.get());
|
||||
LOG.debug(
|
||||
"Decaying costs for the user: {}, its decayedCost: {}, rawCost: {}",
|
||||
entry.getKey(), nextValue, rawCost.get());
|
||||
if (nextValue == 0) {
|
||||
LOG.debug("The decayed count for the user {} is zero " +
|
||||
LOG.debug("The decayed cost for the user {} is zero " +
|
||||
"and being cleaned.", entry.getKey());
|
||||
// We will clean up unused keys here. An interesting optimization
|
||||
// might be to have an upper bound on keyspace in callCounts and only
|
||||
// might be to have an upper bound on keyspace in callCosts and only
|
||||
// clean once we pass it.
|
||||
it.remove();
|
||||
}
|
||||
}
|
||||
|
||||
// Update the total so that we remain in sync
|
||||
totalDecayedCallCount.set(totalDecayedCount);
|
||||
totalRawCallCount.set(totalRawCount);
|
||||
totalDecayedCallCost.set(totalDecayedCost);
|
||||
totalRawCallCost.set(totalRawCost);
|
||||
|
||||
LOG.debug("After decaying the stored counts, totalDecayedCount: {}, " +
|
||||
"totalRawCallCount: {}.", totalDecayedCount, totalRawCount);
|
||||
LOG.debug("After decaying the stored costs, totalDecayedCost: {}, " +
|
||||
"totalRawCallCost: {}.", totalDecayedCost, totalRawCost);
|
||||
// Now refresh the cache of scheduling decisions
|
||||
recomputeScheduleCache();
|
||||
|
||||
// Update average response time with decay
|
||||
updateAverageResponseTime(true);
|
||||
} catch (Exception ex) {
|
||||
LOG.error("decayCurrentCounts exception: " +
|
||||
LOG.error("decayCurrentCosts exception: " +
|
||||
ExceptionUtils.getStackTrace(ex));
|
||||
throw ex;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Update the scheduleCache to match current conditions in callCounts.
|
||||
* Update the scheduleCache to match current conditions in callCosts.
|
||||
*/
|
||||
private void recomputeScheduleCache() {
|
||||
Map<Object, Integer> nextCache = new HashMap<Object, Integer>();
|
||||
|
||||
for (Map.Entry<Object, List<AtomicLong>> entry : callCounts.entrySet()) {
|
||||
for (Map.Entry<Object, List<AtomicLong>> entry : callCosts.entrySet()) {
|
||||
Object id = entry.getKey();
|
||||
AtomicLong value = entry.getValue().get(0);
|
||||
|
||||
@ -466,51 +486,52 @@ private void recomputeScheduleCache() {
|
||||
}
|
||||
|
||||
/**
|
||||
* Get the number of occurrences and increment atomically.
|
||||
* @param identity the identity of the user to increment
|
||||
* @return the value before incrementation
|
||||
* Adjust the stored cost for a given identity.
|
||||
*
|
||||
* @param identity the identity of the user whose cost should be adjusted
|
||||
* @param costDelta the cost to add for the given identity
|
||||
*/
|
||||
private long getAndIncrementCallCounts(Object identity)
|
||||
throws InterruptedException {
|
||||
// We will increment the count, or create it if no such count exists
|
||||
List<AtomicLong> count = this.callCounts.get(identity);
|
||||
if (count == null) {
|
||||
// Create the counts since no such count exists.
|
||||
// idx 0 for decayed call count
|
||||
// idx 1 for the raw call count
|
||||
count = new ArrayList<AtomicLong>(2);
|
||||
count.add(new AtomicLong(0));
|
||||
count.add(new AtomicLong(0));
|
||||
private void addCost(Object identity, long costDelta) {
|
||||
// We will increment the cost, or create it if no such cost exists
|
||||
List<AtomicLong> cost = this.callCosts.get(identity);
|
||||
if (cost == null) {
|
||||
// Create the costs since no such cost exists.
|
||||
// idx 0 for decayed call cost
|
||||
// idx 1 for the raw call cost
|
||||
cost = new ArrayList<AtomicLong>(2);
|
||||
cost.add(new AtomicLong(0));
|
||||
cost.add(new AtomicLong(0));
|
||||
|
||||
// Put it in, or get the AtomicInteger that was put in by another thread
|
||||
List<AtomicLong> otherCount = callCounts.putIfAbsent(identity, count);
|
||||
if (otherCount != null) {
|
||||
count = otherCount;
|
||||
List<AtomicLong> otherCost = callCosts.putIfAbsent(identity, cost);
|
||||
if (otherCost != null) {
|
||||
cost = otherCost;
|
||||
}
|
||||
}
|
||||
|
||||
// Update the total
|
||||
totalDecayedCallCount.getAndIncrement();
|
||||
totalRawCallCount.getAndIncrement();
|
||||
totalDecayedCallCost.getAndAdd(costDelta);
|
||||
totalRawCallCost.getAndAdd(costDelta);
|
||||
|
||||
// At this point value is guaranteed to be not null. It may however have
|
||||
// been clobbered from callCounts. Nonetheless, we return what
|
||||
// been clobbered from callCosts. Nonetheless, we return what
|
||||
// we have.
|
||||
count.get(1).getAndIncrement();
|
||||
return count.get(0).getAndIncrement();
|
||||
cost.get(1).getAndAdd(costDelta);
|
||||
cost.get(0).getAndAdd(costDelta);
|
||||
}
|
||||
|
||||
/**
|
||||
* Given the number of occurrences, compute a scheduling decision.
|
||||
* @param occurrences how many occurrences
|
||||
* Given the cost for an identity, compute a scheduling decision.
|
||||
*
|
||||
* @param cost the cost for an identity
|
||||
* @return scheduling decision from 0 to numLevels - 1
|
||||
*/
|
||||
private int computePriorityLevel(long occurrences) {
|
||||
long totalCallSnapshot = totalDecayedCallCount.get();
|
||||
private int computePriorityLevel(long cost) {
|
||||
long totalCallSnapshot = totalDecayedCallCost.get();
|
||||
|
||||
double proportion = 0;
|
||||
if (totalCallSnapshot > 0) {
|
||||
proportion = (double) occurrences / totalCallSnapshot;
|
||||
proportion = (double) cost / totalCallSnapshot;
|
||||
}
|
||||
|
||||
// Start with low priority levels, since they will be most common
|
||||
@ -531,31 +552,23 @@ private int computePriorityLevel(long occurrences) {
|
||||
* @return integer scheduling decision from 0 to numLevels - 1
|
||||
*/
|
||||
private int cachedOrComputedPriorityLevel(Object identity) {
|
||||
try {
|
||||
long occurrences = this.getAndIncrementCallCounts(identity);
|
||||
|
||||
// Try the cache
|
||||
Map<Object, Integer> scheduleCache = scheduleCacheRef.get();
|
||||
if (scheduleCache != null) {
|
||||
Integer priority = scheduleCache.get(identity);
|
||||
if (priority != null) {
|
||||
LOG.debug("Cache priority for: {} with priority: {}", identity,
|
||||
priority);
|
||||
return priority;
|
||||
}
|
||||
// Try the cache
|
||||
Map<Object, Integer> scheduleCache = scheduleCacheRef.get();
|
||||
if (scheduleCache != null) {
|
||||
Integer priority = scheduleCache.get(identity);
|
||||
if (priority != null) {
|
||||
LOG.debug("Cache priority for: {} with priority: {}", identity,
|
||||
priority);
|
||||
return priority;
|
||||
}
|
||||
|
||||
// Cache was no good, compute it
|
||||
int priority = computePriorityLevel(occurrences);
|
||||
LOG.debug("compute priority for " + identity + " priority " + priority);
|
||||
return priority;
|
||||
|
||||
} catch (InterruptedException ie) {
|
||||
LOG.warn("Caught InterruptedException, returning low priority level");
|
||||
LOG.debug("Fallback priority for: {} with priority: {}", identity,
|
||||
numLevels - 1);
|
||||
return numLevels - 1;
|
||||
}
|
||||
|
||||
// Cache was no good, compute it
|
||||
List<AtomicLong> costList = callCosts.get(identity);
|
||||
long currentCost = costList == null ? 0 : costList.get(0).get();
|
||||
int priority = computePriorityLevel(currentCost);
|
||||
LOG.debug("compute priority for {} priority {}", identity, priority);
|
||||
return priority;
|
||||
}
|
||||
|
||||
/**
|
||||
@ -605,6 +618,10 @@ public boolean shouldBackOff(Schedulable obj) {
|
||||
@Override
|
||||
public void addResponseTime(String callName, Schedulable schedulable,
|
||||
ProcessingDetails details) {
|
||||
String user = identityProvider.makeIdentity(schedulable);
|
||||
long processingCost = costProvider.getCost(details);
|
||||
addCost(user, processingCost);
|
||||
|
||||
int priorityLevel = schedulable.getPriorityLevel();
|
||||
long queueTime = details.get(Timing.QUEUE, TimeUnit.MILLISECONDS);
|
||||
long processingTime = details.get(Timing.PROCESSING, TimeUnit.MILLISECONDS);
|
||||
@ -652,22 +669,30 @@ void updateAverageResponseTime(boolean enableDecay) {
|
||||
|
||||
// For testing
|
||||
@VisibleForTesting
|
||||
public double getDecayFactor() { return decayFactor; }
|
||||
double getDecayFactor() {
|
||||
return decayFactor;
|
||||
}
|
||||
|
||||
@VisibleForTesting
|
||||
public long getDecayPeriodMillis() { return decayPeriodMillis; }
|
||||
long getDecayPeriodMillis() {
|
||||
return decayPeriodMillis;
|
||||
}
|
||||
|
||||
@VisibleForTesting
|
||||
public double[] getThresholds() { return thresholds; }
|
||||
double[] getThresholds() {
|
||||
return thresholds;
|
||||
}
|
||||
|
||||
@VisibleForTesting
|
||||
public void forceDecay() { decayCurrentCounts(); }
|
||||
void forceDecay() {
|
||||
decayCurrentCosts();
|
||||
}
|
||||
|
||||
@VisibleForTesting
|
||||
public Map<Object, Long> getCallCountSnapshot() {
|
||||
Map<Object, Long> getCallCostSnapshot() {
|
||||
HashMap<Object, Long> snapshot = new HashMap<Object, Long>();
|
||||
|
||||
for (Map.Entry<Object, List<AtomicLong>> entry : callCounts.entrySet()) {
|
||||
for (Map.Entry<Object, List<AtomicLong>> entry : callCosts.entrySet()) {
|
||||
snapshot.put(entry.getKey(), entry.getValue().get(0).get());
|
||||
}
|
||||
|
||||
@ -675,8 +700,8 @@ public Map<Object, Long> getCallCountSnapshot() {
|
||||
}
|
||||
|
||||
@VisibleForTesting
|
||||
public long getTotalCallSnapshot() {
|
||||
return totalDecayedCallCount.get();
|
||||
long getTotalCallSnapshot() {
|
||||
return totalDecayedCallCost.get();
|
||||
}
|
||||
|
||||
/**
|
||||
@ -809,15 +834,15 @@ public void getMetrics(MetricsCollector collector, boolean all) {
|
||||
}
|
||||
|
||||
public int getUniqueIdentityCount() {
|
||||
return callCounts.size();
|
||||
return callCosts.size();
|
||||
}
|
||||
|
||||
public long getTotalCallVolume() {
|
||||
return totalDecayedCallCount.get();
|
||||
return totalDecayedCallCost.get();
|
||||
}
|
||||
|
||||
public long getTotalRawCallVolume() {
|
||||
return totalRawCallCount.get();
|
||||
return totalRawCallCost.get();
|
||||
}
|
||||
|
||||
public long[] getResponseTimeCountInLastWindow() {
|
||||
@ -910,17 +935,17 @@ private void addTopNCallerSummary(MetricsRecordBuilder rb) {
|
||||
}
|
||||
}
|
||||
|
||||
// Get the top N callers' raw call count and scheduler decision
|
||||
// Get the top N callers' raw call cost and scheduler decision
|
||||
private TopN getTopCallers(int n) {
|
||||
TopN topNCallers = new TopN(n);
|
||||
Iterator<Map.Entry<Object, List<AtomicLong>>> it =
|
||||
callCounts.entrySet().iterator();
|
||||
callCosts.entrySet().iterator();
|
||||
while (it.hasNext()) {
|
||||
Map.Entry<Object, List<AtomicLong>> entry = it.next();
|
||||
String caller = entry.getKey().toString();
|
||||
Long count = entry.getValue().get(1).get();
|
||||
if (count > 0) {
|
||||
topNCallers.offer(new NameValuePair(caller, count));
|
||||
Long cost = entry.getValue().get(1).get();
|
||||
if (cost > 0) {
|
||||
topNCallers.offer(new NameValuePair(caller, cost));
|
||||
}
|
||||
}
|
||||
return topNCallers;
|
||||
@ -941,25 +966,25 @@ public String getSchedulingDecisionSummary() {
|
||||
|
||||
public String getCallVolumeSummary() {
|
||||
try {
|
||||
return WRITER.writeValueAsString(getDecayedCallCounts());
|
||||
return WRITER.writeValueAsString(getDecayedCallCosts());
|
||||
} catch (Exception e) {
|
||||
return "Error: " + e.getMessage();
|
||||
}
|
||||
}
|
||||
|
||||
private Map<Object, Long> getDecayedCallCounts() {
|
||||
Map<Object, Long> decayedCallCounts = new HashMap<>(callCounts.size());
|
||||
private Map<Object, Long> getDecayedCallCosts() {
|
||||
Map<Object, Long> decayedCallCosts = new HashMap<>(callCosts.size());
|
||||
Iterator<Map.Entry<Object, List<AtomicLong>>> it =
|
||||
callCounts.entrySet().iterator();
|
||||
callCosts.entrySet().iterator();
|
||||
while (it.hasNext()) {
|
||||
Map.Entry<Object, List<AtomicLong>> entry = it.next();
|
||||
Object user = entry.getKey();
|
||||
Long decayedCount = entry.getValue().get(0).get();
|
||||
if (decayedCount > 0) {
|
||||
decayedCallCounts.put(user, decayedCount);
|
||||
Long decayedCost = entry.getValue().get(0).get();
|
||||
if (decayedCost > 0) {
|
||||
decayedCallCosts.put(user, decayedCost);
|
||||
}
|
||||
}
|
||||
return decayedCallCounts;
|
||||
return decayedCallCosts;
|
||||
}
|
||||
|
||||
@Override
|
||||
|
||||
@ -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.ipc;
|
||||
|
||||
import org.apache.hadoop.conf.Configuration;
|
||||
|
||||
/**
|
||||
* Ignores process details and returns a constant value for each call.
|
||||
*/
|
||||
public class DefaultCostProvider implements CostProvider {
|
||||
|
||||
@Override
|
||||
public void init(String namespace, Configuration conf) {
|
||||
// No-op
|
||||
}
|
||||
|
||||
/**
|
||||
* Returns 1, regardless of the processing details.
|
||||
*
|
||||
* @param details Process details (ignored)
|
||||
* @return 1
|
||||
*/
|
||||
@Override
|
||||
public long getCost(ProcessingDetails details) {
|
||||
return 1;
|
||||
}
|
||||
}
|
||||
@ -0,0 +1,110 @@
|
||||
/**
|
||||
* 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.ipc;
|
||||
|
||||
import java.util.Locale;
|
||||
import org.apache.hadoop.conf.Configuration;
|
||||
|
||||
import static org.apache.hadoop.ipc.ProcessingDetails.Timing;
|
||||
|
||||
/**
|
||||
* A {@link CostProvider} that calculates the cost for an operation
|
||||
* as a weighted sum of its processing time values (see
|
||||
* {@link ProcessingDetails}). This can be used by specifying the
|
||||
* {@link org.apache.hadoop.fs.CommonConfigurationKeys#IPC_COST_PROVIDER_KEY}
|
||||
* configuration key.
|
||||
*
|
||||
* <p/>This allows for configuration of how heavily each of the operations
|
||||
* within {@link ProcessingDetails} is weighted. By default,
|
||||
* {@link ProcessingDetails.Timing#LOCKFREE},
|
||||
* {@link ProcessingDetails.Timing#RESPONSE}, and
|
||||
* {@link ProcessingDetails.Timing#HANDLER} times have a weight of
|
||||
* {@value #DEFAULT_LOCKFREE_WEIGHT},
|
||||
* {@link ProcessingDetails.Timing#LOCKSHARED} has a weight of
|
||||
* {@value #DEFAULT_LOCKSHARED_WEIGHT},
|
||||
* {@link ProcessingDetails.Timing#LOCKEXCLUSIVE} has a weight of
|
||||
* {@value #DEFAULT_LOCKEXCLUSIVE_WEIGHT}, and others are ignored.
|
||||
* These values can all be configured using the {@link #WEIGHT_CONFIG_PREFIX}
|
||||
* key, prefixed with the IPC namespace, and suffixed with the name of the
|
||||
* timing measurement from {@link ProcessingDetails} (all lowercase).
|
||||
* For example, to set the lock exclusive weight to be 1000, set:
|
||||
* <pre>
|
||||
* ipc.8020.cost-provider.impl=org.apache.hadoop.ipc.WeightedTimeCostProvider
|
||||
* ipc.8020.weighted-cost.lockexclusive=1000
|
||||
* </pre>
|
||||
*/
|
||||
public class WeightedTimeCostProvider implements CostProvider {
|
||||
|
||||
/**
|
||||
* The prefix used in configuration values specifying the weight to use when
|
||||
* determining the cost of an operation. See the class Javadoc for more info.
|
||||
*/
|
||||
public static final String WEIGHT_CONFIG_PREFIX = ".weighted-cost.";
|
||||
static final int DEFAULT_LOCKFREE_WEIGHT = 1;
|
||||
static final int DEFAULT_LOCKSHARED_WEIGHT = 10;
|
||||
static final int DEFAULT_LOCKEXCLUSIVE_WEIGHT = 100;
|
||||
|
||||
private long[] weights;
|
||||
|
||||
@Override
|
||||
public void init(String namespace, Configuration conf) {
|
||||
weights = new long[Timing.values().length];
|
||||
for (Timing timing : ProcessingDetails.Timing.values()) {
|
||||
final int defaultValue;
|
||||
switch (timing) {
|
||||
case LOCKFREE:
|
||||
case RESPONSE:
|
||||
case HANDLER:
|
||||
defaultValue = DEFAULT_LOCKFREE_WEIGHT;
|
||||
break;
|
||||
case LOCKSHARED:
|
||||
defaultValue = DEFAULT_LOCKSHARED_WEIGHT;
|
||||
break;
|
||||
case LOCKEXCLUSIVE:
|
||||
defaultValue = DEFAULT_LOCKEXCLUSIVE_WEIGHT;
|
||||
break;
|
||||
default:
|
||||
// by default don't bill for queueing or lock wait time
|
||||
defaultValue = 0;
|
||||
}
|
||||
String key = namespace + WEIGHT_CONFIG_PREFIX
|
||||
+ timing.name().toLowerCase(Locale.ENGLISH);
|
||||
weights[timing.ordinal()] = conf.getInt(key, defaultValue);
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Calculates a weighted sum of the times stored on the provided processing
|
||||
* details to be used as the cost in {@link DecayRpcScheduler}.
|
||||
*
|
||||
* @param details Processing details
|
||||
* @return The weighted sum of the times. The returned unit is the same
|
||||
* as the default unit used by the provided processing details.
|
||||
*/
|
||||
@Override
|
||||
public long getCost(ProcessingDetails details) {
|
||||
assert weights != null : "Cost provider must be initialized before use";
|
||||
long cost = 0;
|
||||
// weights was initialized to the same length as Timing.values()
|
||||
for (int i = 0; i < Timing.values().length; i++) {
|
||||
cost += details.get(Timing.values()[i]) * weights[i];
|
||||
}
|
||||
return cost;
|
||||
}
|
||||
}
|
||||
@ -91,6 +91,21 @@ This is configurable via the **identity provider**, which defaults to the **User
|
||||
provider simply uses the username of the client submitting the request. However, a custom identity provider can be used
|
||||
to performing throttling based on other groupings, or using an external identity provider.
|
||||
|
||||
### Cost-based Fair Call Queue
|
||||
|
||||
Though the fair call queue itself does a good job of mitigating the impact from users who submit a very high _number_
|
||||
of requests, it does not take account into how expensive each request is to process. Thus, when considering the
|
||||
HDFS NameNode, a user who submits 1000 "getFileInfo" requests would be prioritized the same as a user who submits 1000
|
||||
"listStatus" requests on some very large directory, or a user who submits 1000 "mkdir" requests, which are more
|
||||
expensive as they require an exclusive lock on the namesystem. To account for the _cost_ of an operation when
|
||||
considering the prioritization of user requests, there is a "cost-based" extension to the Fair Call Queue which uses
|
||||
the aggregate processing time of a user's operations to determine how that user should be prioritized. By default,
|
||||
queue time (time spent waiting to be processed) and lock wait time (time spent waiting to acquire a lock) is not
|
||||
considered in the cost, time spent processing without a lock is neutrally (1x) weighted, time spent processing with a
|
||||
shared lock is weighted 10x higher, and time spent processing with an exclusive lock is weighted 100x higher.
|
||||
This attempts to prioritize users based on the actual load they place on the server. To enable this feature, set the
|
||||
`costprovder.impl` configuration to `org.apache.hadoop.ipc.WeightedTimeCostProvider` as described below.
|
||||
|
||||
Configuration
|
||||
-------------
|
||||
|
||||
@ -115,12 +130,16 @@ omitted.
|
||||
| scheduler.priority.levels | RpcScheduler, CallQueue | How many priority levels to use within the scheduler and call queue. | 4 |
|
||||
| faircallqueue.multiplexer.weights | WeightedRoundRobinMultiplexer | How much weight to give to each priority queue. This should be a comma-separated list of length equal to the number of priority levels. | Weights descend by a factor of 2 (e.g., for 4 levels: `8,4,2,1`) |
|
||||
| identity-provider.impl | DecayRpcScheduler | The identity provider mapping user requests to their identity. | org.apache.hadoop.ipc.UserIdentityProvider |
|
||||
| cost-provider.impl | DecayRpcScheduler | The cost provider mapping user requests to their cost. To enable determination of cost based on processing time, use `org.apache.hadoop.ipc.WeightedTimeCostProvider`. | org.apache.hadoop.ipc.DefaultCostProvider |
|
||||
| decay-scheduler.period-ms | DecayRpcScheduler | How frequently the decay factor should be applied to the operation counts of users. Higher values have less overhead, but respond less quickly to changes in client behavior. | 5000 |
|
||||
| decay-scheduler.decay-factor | DecayRpcScheduler | When decaying the operation counts of users, the multiplicative decay factor to apply. Higher values will weight older operations more strongly, essentially giving the scheduler a longer memory, and penalizing heavy clients for a longer period of time. | 0.5 |
|
||||
| decay-scheduler.thresholds | DecayRpcScheduler | The client load threshold, as an integer percentage, for each priority queue. Clients producing less load, as a percent of total operations, than specified at position _i_ will be given priority _i_. This should be a comma-separated list of length equal to the number of priority levels minus 1 (the last is implicitly 100). | Thresholds ascend by a factor of 2 (e.g., for 4 levels: `13,25,50`) |
|
||||
| decay-scheduler.backoff.responsetime.enable | DecayRpcScheduler | Whether or not to enable the backoff by response time feature. | false |
|
||||
| decay-scheduler.backoff.responsetime.thresholds | DecayRpcScheduler | The response time thresholds, as time durations, for each priority queue. If the average response time for a queue is above this threshold, backoff will occur in lower priority queues. This should be a comma-separated list of length equal to the number of priority levels. | Threshold increases by 10s per level (e.g., for 4 levels: `10s,20s,30s,40s`) |
|
||||
| decay-scheduler.metrics.top.user.count | DecayRpcScheduler | The number of top (i.e., heaviest) users to emit metric information about. | 10 |
|
||||
| weighted-cost.lockshared | WeightedTimeCostProvider | The weight multiplier to apply to the time spent in the processing phase which holds a shared (read) lock. | 10 |
|
||||
| weighted-cost.lockexclusive | WeightedTimeCostProvider | The weight multiplier to apply to the time spent in the processing phase which holds an exclusive (write) lock. | 100 |
|
||||
| weighted-cost.{handler,lockfree,response} | WeightedTimeCostProvider | The weight multiplier to apply to the time spent in the processing phases which do not involve holding a lock. See `org.apache.hadoop.ipc.ProcessingDetails.Timing` for more details on each phase. | 1 |
|
||||
|
||||
### Example Configuration
|
||||
|
||||
|
||||
@ -26,6 +26,7 @@
|
||||
import static org.mockito.Mockito.mock;
|
||||
import static org.mockito.Mockito.when;
|
||||
|
||||
import org.apache.hadoop.fs.CommonConfigurationKeys;
|
||||
import org.apache.hadoop.metrics2.lib.DefaultMetricsSystem;
|
||||
import org.apache.hadoop.security.UserGroupInformation;
|
||||
import org.apache.hadoop.conf.Configuration;
|
||||
@ -36,6 +37,7 @@
|
||||
import java.io.ByteArrayOutputStream;
|
||||
import java.io.PrintStream;
|
||||
import java.lang.management.ManagementFactory;
|
||||
import java.util.concurrent.TimeUnit;
|
||||
|
||||
public class TestDecayRpcScheduler {
|
||||
private Schedulable mockCall(String id) {
|
||||
@ -131,67 +133,69 @@ public void testAccumulate() {
|
||||
conf.set("ns." + DecayRpcScheduler.IPC_FCQ_DECAYSCHEDULER_PERIOD_KEY, "99999999"); // Never flush
|
||||
scheduler = new DecayRpcScheduler(1, "ns", conf);
|
||||
|
||||
assertEquals(0, scheduler.getCallCountSnapshot().size()); // empty first
|
||||
assertEquals(0, scheduler.getCallCostSnapshot().size()); // empty first
|
||||
|
||||
scheduler.getPriorityLevel(mockCall("A"));
|
||||
assertEquals(1, scheduler.getCallCountSnapshot().get("A").longValue());
|
||||
assertEquals(1, scheduler.getCallCountSnapshot().get("A").longValue());
|
||||
getPriorityIncrementCallCount("A");
|
||||
assertEquals(1, scheduler.getCallCostSnapshot().get("A").longValue());
|
||||
assertEquals(1, scheduler.getCallCostSnapshot().get("A").longValue());
|
||||
|
||||
scheduler.getPriorityLevel(mockCall("A"));
|
||||
scheduler.getPriorityLevel(mockCall("B"));
|
||||
scheduler.getPriorityLevel(mockCall("A"));
|
||||
getPriorityIncrementCallCount("A");
|
||||
getPriorityIncrementCallCount("B");
|
||||
getPriorityIncrementCallCount("A");
|
||||
|
||||
assertEquals(3, scheduler.getCallCountSnapshot().get("A").longValue());
|
||||
assertEquals(1, scheduler.getCallCountSnapshot().get("B").longValue());
|
||||
assertEquals(3, scheduler.getCallCostSnapshot().get("A").longValue());
|
||||
assertEquals(1, scheduler.getCallCostSnapshot().get("B").longValue());
|
||||
}
|
||||
|
||||
@Test
|
||||
@SuppressWarnings("deprecation")
|
||||
public void testDecay() throws Exception {
|
||||
Configuration conf = new Configuration();
|
||||
conf.set("ns." + DecayRpcScheduler.IPC_FCQ_DECAYSCHEDULER_PERIOD_KEY, "999999999"); // Never
|
||||
conf.set("ns." + DecayRpcScheduler.IPC_FCQ_DECAYSCHEDULER_FACTOR_KEY, "0.5");
|
||||
conf.setLong("ns." // Never decay
|
||||
+ DecayRpcScheduler.IPC_SCHEDULER_DECAYSCHEDULER_PERIOD_KEY, 999999999);
|
||||
conf.setDouble("ns."
|
||||
+ DecayRpcScheduler.IPC_SCHEDULER_DECAYSCHEDULER_FACTOR_KEY, 0.5);
|
||||
scheduler = new DecayRpcScheduler(1, "ns", conf);
|
||||
|
||||
assertEquals(0, scheduler.getTotalCallSnapshot());
|
||||
|
||||
for (int i = 0; i < 4; i++) {
|
||||
scheduler.getPriorityLevel(mockCall("A"));
|
||||
getPriorityIncrementCallCount("A");
|
||||
}
|
||||
|
||||
sleep(1000);
|
||||
|
||||
for (int i = 0; i < 8; i++) {
|
||||
scheduler.getPriorityLevel(mockCall("B"));
|
||||
getPriorityIncrementCallCount("B");
|
||||
}
|
||||
|
||||
assertEquals(12, scheduler.getTotalCallSnapshot());
|
||||
assertEquals(4, scheduler.getCallCountSnapshot().get("A").longValue());
|
||||
assertEquals(8, scheduler.getCallCountSnapshot().get("B").longValue());
|
||||
assertEquals(4, scheduler.getCallCostSnapshot().get("A").longValue());
|
||||
assertEquals(8, scheduler.getCallCostSnapshot().get("B").longValue());
|
||||
|
||||
scheduler.forceDecay();
|
||||
|
||||
assertEquals(6, scheduler.getTotalCallSnapshot());
|
||||
assertEquals(2, scheduler.getCallCountSnapshot().get("A").longValue());
|
||||
assertEquals(4, scheduler.getCallCountSnapshot().get("B").longValue());
|
||||
assertEquals(2, scheduler.getCallCostSnapshot().get("A").longValue());
|
||||
assertEquals(4, scheduler.getCallCostSnapshot().get("B").longValue());
|
||||
|
||||
scheduler.forceDecay();
|
||||
|
||||
assertEquals(3, scheduler.getTotalCallSnapshot());
|
||||
assertEquals(1, scheduler.getCallCountSnapshot().get("A").longValue());
|
||||
assertEquals(2, scheduler.getCallCountSnapshot().get("B").longValue());
|
||||
assertEquals(1, scheduler.getCallCostSnapshot().get("A").longValue());
|
||||
assertEquals(2, scheduler.getCallCostSnapshot().get("B").longValue());
|
||||
|
||||
scheduler.forceDecay();
|
||||
|
||||
assertEquals(1, scheduler.getTotalCallSnapshot());
|
||||
assertEquals(null, scheduler.getCallCountSnapshot().get("A"));
|
||||
assertEquals(1, scheduler.getCallCountSnapshot().get("B").longValue());
|
||||
assertEquals(null, scheduler.getCallCostSnapshot().get("A"));
|
||||
assertEquals(1, scheduler.getCallCostSnapshot().get("B").longValue());
|
||||
|
||||
scheduler.forceDecay();
|
||||
|
||||
assertEquals(0, scheduler.getTotalCallSnapshot());
|
||||
assertEquals(null, scheduler.getCallCountSnapshot().get("A"));
|
||||
assertEquals(null, scheduler.getCallCountSnapshot().get("B"));
|
||||
assertEquals(null, scheduler.getCallCostSnapshot().get("A"));
|
||||
assertEquals(null, scheduler.getCallCostSnapshot().get("B"));
|
||||
}
|
||||
|
||||
@Test
|
||||
@ -205,16 +209,16 @@ public void testPriority() throws Exception {
|
||||
.IPC_FCQ_DECAYSCHEDULER_THRESHOLDS_KEY, "25, 50, 75");
|
||||
scheduler = new DecayRpcScheduler(4, namespace, conf);
|
||||
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("A")));
|
||||
assertEquals(2, scheduler.getPriorityLevel(mockCall("A")));
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("B")));
|
||||
assertEquals(1, scheduler.getPriorityLevel(mockCall("B")));
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("C")));
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("C")));
|
||||
assertEquals(1, scheduler.getPriorityLevel(mockCall("A")));
|
||||
assertEquals(1, scheduler.getPriorityLevel(mockCall("A")));
|
||||
assertEquals(1, scheduler.getPriorityLevel(mockCall("A")));
|
||||
assertEquals(2, scheduler.getPriorityLevel(mockCall("A")));
|
||||
assertEquals(0, getPriorityIncrementCallCount("A")); // 0 out of 0 calls
|
||||
assertEquals(3, getPriorityIncrementCallCount("A")); // 1 out of 1 calls
|
||||
assertEquals(0, getPriorityIncrementCallCount("B")); // 0 out of 2 calls
|
||||
assertEquals(1, getPriorityIncrementCallCount("B")); // 1 out of 3 calls
|
||||
assertEquals(0, getPriorityIncrementCallCount("C")); // 0 out of 4 calls
|
||||
assertEquals(0, getPriorityIncrementCallCount("C")); // 1 out of 5 calls
|
||||
assertEquals(1, getPriorityIncrementCallCount("A")); // 2 out of 6 calls
|
||||
assertEquals(1, getPriorityIncrementCallCount("A")); // 3 out of 7 calls
|
||||
assertEquals(2, getPriorityIncrementCallCount("A")); // 4 out of 8 calls
|
||||
assertEquals(2, getPriorityIncrementCallCount("A")); // 5 out of 9 calls
|
||||
|
||||
MBeanServer mbs = ManagementFactory.getPlatformMBeanServer();
|
||||
ObjectName mxbeanName = new ObjectName(
|
||||
@ -243,7 +247,7 @@ public void testPeriodic() throws InterruptedException {
|
||||
assertEquals(0, scheduler.getTotalCallSnapshot());
|
||||
|
||||
for (int i = 0; i < 64; i++) {
|
||||
scheduler.getPriorityLevel(mockCall("A"));
|
||||
getPriorityIncrementCallCount("A");
|
||||
}
|
||||
|
||||
// It should eventually decay to zero
|
||||
@ -272,6 +276,108 @@ public void testNPEatInitialization() throws InterruptedException {
|
||||
//set systout back
|
||||
System.setOut(output);
|
||||
}
|
||||
}
|
||||
|
||||
@Test
|
||||
public void testUsingWeightedTimeCostProvider() {
|
||||
scheduler = getSchedulerWithWeightedTimeCostProvider(3);
|
||||
|
||||
// 3 details in increasing order of cost. Although medium has a longer
|
||||
// duration, the shared lock is weighted less than the exclusive lock
|
||||
ProcessingDetails callDetailsLow =
|
||||
new ProcessingDetails(TimeUnit.MILLISECONDS);
|
||||
callDetailsLow.set(ProcessingDetails.Timing.LOCKFREE, 1);
|
||||
ProcessingDetails callDetailsMedium =
|
||||
new ProcessingDetails(TimeUnit.MILLISECONDS);
|
||||
callDetailsMedium.set(ProcessingDetails.Timing.LOCKSHARED, 500);
|
||||
ProcessingDetails callDetailsHigh =
|
||||
new ProcessingDetails(TimeUnit.MILLISECONDS);
|
||||
callDetailsHigh.set(ProcessingDetails.Timing.LOCKEXCLUSIVE, 100);
|
||||
|
||||
for (int i = 0; i < 10; i++) {
|
||||
scheduler.addResponseTime("ignored", mockCall("LOW"), callDetailsLow);
|
||||
}
|
||||
scheduler.addResponseTime("ignored", mockCall("MED"), callDetailsMedium);
|
||||
scheduler.addResponseTime("ignored", mockCall("HIGH"), callDetailsHigh);
|
||||
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("LOW")));
|
||||
assertEquals(1, scheduler.getPriorityLevel(mockCall("MED")));
|
||||
assertEquals(2, scheduler.getPriorityLevel(mockCall("HIGH")));
|
||||
|
||||
assertEquals(3, scheduler.getUniqueIdentityCount());
|
||||
long totalCallInitial = scheduler.getTotalRawCallVolume();
|
||||
assertEquals(totalCallInitial, scheduler.getTotalCallVolume());
|
||||
|
||||
scheduler.forceDecay();
|
||||
|
||||
// Relative priorities should stay the same after a single decay
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("LOW")));
|
||||
assertEquals(1, scheduler.getPriorityLevel(mockCall("MED")));
|
||||
assertEquals(2, scheduler.getPriorityLevel(mockCall("HIGH")));
|
||||
|
||||
assertEquals(3, scheduler.getUniqueIdentityCount());
|
||||
assertEquals(totalCallInitial, scheduler.getTotalRawCallVolume());
|
||||
assertTrue(scheduler.getTotalCallVolume() < totalCallInitial);
|
||||
|
||||
for (int i = 0; i < 100; i++) {
|
||||
scheduler.forceDecay();
|
||||
}
|
||||
// After enough decay cycles, all callers should be high priority again
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("LOW")));
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("MED")));
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("HIGH")));
|
||||
}
|
||||
|
||||
@Test
|
||||
public void testUsingWeightedTimeCostProviderWithZeroCostCalls() {
|
||||
scheduler = getSchedulerWithWeightedTimeCostProvider(2);
|
||||
|
||||
ProcessingDetails emptyDetails =
|
||||
new ProcessingDetails(TimeUnit.MILLISECONDS);
|
||||
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
scheduler.addResponseTime("ignored", mockCall("MANY"), emptyDetails);
|
||||
}
|
||||
scheduler.addResponseTime("ignored", mockCall("FEW"), emptyDetails);
|
||||
|
||||
// Since the calls are all "free", they should have the same priority
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("MANY")));
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("FEW")));
|
||||
}
|
||||
|
||||
@Test
|
||||
public void testUsingWeightedTimeCostProviderNoRequests() {
|
||||
scheduler = getSchedulerWithWeightedTimeCostProvider(2);
|
||||
|
||||
assertEquals(0, scheduler.getPriorityLevel(mockCall("A")));
|
||||
}
|
||||
|
||||
/**
|
||||
* Get a scheduler that uses {@link WeightedTimeCostProvider} and has
|
||||
* normal decaying disabled.
|
||||
*/
|
||||
private static DecayRpcScheduler getSchedulerWithWeightedTimeCostProvider(
|
||||
int priorityLevels) {
|
||||
Configuration conf = new Configuration();
|
||||
conf.setClass("ns." + CommonConfigurationKeys.IPC_COST_PROVIDER_KEY,
|
||||
WeightedTimeCostProvider.class, CostProvider.class);
|
||||
conf.setLong("ns."
|
||||
+ DecayRpcScheduler.IPC_SCHEDULER_DECAYSCHEDULER_PERIOD_KEY, 999999);
|
||||
return new DecayRpcScheduler(priorityLevels, "ns", conf);
|
||||
}
|
||||
|
||||
/**
|
||||
* Get the priority and increment the call count, assuming that
|
||||
* {@link DefaultCostProvider} is in use.
|
||||
*/
|
||||
private int getPriorityIncrementCallCount(String callId) {
|
||||
Schedulable mockCall = mockCall(callId);
|
||||
int priority = scheduler.getPriorityLevel(mockCall);
|
||||
// The DefaultCostProvider uses a cost of 1 for all calls, ignoring
|
||||
// the processing details, so an empty one is fine
|
||||
ProcessingDetails emptyProcessingDetails =
|
||||
new ProcessingDetails(TimeUnit.MILLISECONDS);
|
||||
scheduler.addResponseTime("ignored", mockCall, emptyProcessingDetails);
|
||||
return priority;
|
||||
}
|
||||
}
|
||||
@ -18,7 +18,6 @@
|
||||
|
||||
package org.apache.hadoop.ipc;
|
||||
|
||||
import com.google.common.base.Supplier;
|
||||
import com.google.protobuf.ServiceException;
|
||||
import org.apache.hadoop.HadoopIllegalArgumentException;
|
||||
import org.apache.hadoop.conf.Configuration;
|
||||
@ -1195,15 +1194,6 @@ public void testClientBackOffByResponseTime() throws Exception {
|
||||
Exception lastException = null;
|
||||
proxy = getClient(addr, conf);
|
||||
|
||||
MetricsRecordBuilder rb1 =
|
||||
getMetrics("DecayRpcSchedulerMetrics2." + ns);
|
||||
final long beginDecayedCallVolume = MetricsAsserts.getLongCounter(
|
||||
"DecayedCallVolume", rb1);
|
||||
final long beginRawCallVolume = MetricsAsserts.getLongCounter(
|
||||
"CallVolume", rb1);
|
||||
final int beginUniqueCaller = MetricsAsserts.getIntCounter("UniqueCallers",
|
||||
rb1);
|
||||
|
||||
try {
|
||||
// start a sleep RPC call that sleeps 3s.
|
||||
for (int i = 0; i < numClients; i++) {
|
||||
@ -1231,41 +1221,6 @@ public Void call() throws ServiceException, InterruptedException {
|
||||
} else {
|
||||
lastException = unwrapExeption;
|
||||
}
|
||||
|
||||
// Lets Metric system update latest metrics
|
||||
GenericTestUtils.waitFor(new Supplier<Boolean>() {
|
||||
@Override
|
||||
public Boolean get() {
|
||||
MetricsRecordBuilder rb2 =
|
||||
getMetrics("DecayRpcSchedulerMetrics2." + ns);
|
||||
long decayedCallVolume1 = MetricsAsserts.getLongCounter(
|
||||
"DecayedCallVolume", rb2);
|
||||
long rawCallVolume1 = MetricsAsserts.getLongCounter(
|
||||
"CallVolume", rb2);
|
||||
int uniqueCaller1 = MetricsAsserts.getIntCounter(
|
||||
"UniqueCallers", rb2);
|
||||
long callVolumePriority0 = MetricsAsserts.getLongGauge(
|
||||
"Priority.0.CompletedCallVolume", rb2);
|
||||
long callVolumePriority1 = MetricsAsserts.getLongGauge(
|
||||
"Priority.1.CompletedCallVolume", rb2);
|
||||
double avgRespTimePriority0 = MetricsAsserts.getDoubleGauge(
|
||||
"Priority.0.AvgResponseTime", rb2);
|
||||
double avgRespTimePriority1 = MetricsAsserts.getDoubleGauge(
|
||||
"Priority.1.AvgResponseTime", rb2);
|
||||
|
||||
LOG.info("DecayedCallVolume: " + decayedCallVolume1);
|
||||
LOG.info("CallVolume: " + rawCallVolume1);
|
||||
LOG.info("UniqueCaller: " + uniqueCaller1);
|
||||
LOG.info("Priority.0.CompletedCallVolume: " + callVolumePriority0);
|
||||
LOG.info("Priority.1.CompletedCallVolume: " + callVolumePriority1);
|
||||
LOG.info("Priority.0.AvgResponseTime: " + avgRespTimePriority0);
|
||||
LOG.info("Priority.1.AvgResponseTime: " + avgRespTimePriority1);
|
||||
|
||||
return decayedCallVolume1 > beginDecayedCallVolume &&
|
||||
rawCallVolume1 > beginRawCallVolume &&
|
||||
uniqueCaller1 > beginUniqueCaller;
|
||||
}
|
||||
}, 30, 60000);
|
||||
}
|
||||
} finally {
|
||||
executorService.shutdown();
|
||||
@ -1277,6 +1232,63 @@ public Boolean get() {
|
||||
assertTrue("RetriableException not received", succeeded);
|
||||
}
|
||||
|
||||
/** Test that the metrics for DecayRpcScheduler are updated. */
|
||||
@Test (timeout=30000)
|
||||
public void testDecayRpcSchedulerMetrics() throws Exception {
|
||||
final String ns = CommonConfigurationKeys.IPC_NAMESPACE + ".0";
|
||||
Server server = setupDecayRpcSchedulerandTestServer(ns + ".");
|
||||
|
||||
MetricsRecordBuilder rb1 =
|
||||
getMetrics("DecayRpcSchedulerMetrics2." + ns);
|
||||
final long beginDecayedCallVolume = MetricsAsserts.getLongCounter(
|
||||
"DecayedCallVolume", rb1);
|
||||
final long beginRawCallVolume = MetricsAsserts.getLongCounter(
|
||||
"CallVolume", rb1);
|
||||
final int beginUniqueCaller = MetricsAsserts.getIntCounter("UniqueCallers",
|
||||
rb1);
|
||||
|
||||
TestRpcService proxy = getClient(addr, conf);
|
||||
try {
|
||||
for (int i = 0; i < 2; i++) {
|
||||
proxy.sleep(null, newSleepRequest(100));
|
||||
}
|
||||
|
||||
// Lets Metric system update latest metrics
|
||||
GenericTestUtils.waitFor(() -> {
|
||||
MetricsRecordBuilder rb2 =
|
||||
getMetrics("DecayRpcSchedulerMetrics2." + ns);
|
||||
long decayedCallVolume1 = MetricsAsserts.getLongCounter(
|
||||
"DecayedCallVolume", rb2);
|
||||
long rawCallVolume1 = MetricsAsserts.getLongCounter(
|
||||
"CallVolume", rb2);
|
||||
int uniqueCaller1 = MetricsAsserts.getIntCounter(
|
||||
"UniqueCallers", rb2);
|
||||
long callVolumePriority0 = MetricsAsserts.getLongGauge(
|
||||
"Priority.0.CompletedCallVolume", rb2);
|
||||
long callVolumePriority1 = MetricsAsserts.getLongGauge(
|
||||
"Priority.1.CompletedCallVolume", rb2);
|
||||
double avgRespTimePriority0 = MetricsAsserts.getDoubleGauge(
|
||||
"Priority.0.AvgResponseTime", rb2);
|
||||
double avgRespTimePriority1 = MetricsAsserts.getDoubleGauge(
|
||||
"Priority.1.AvgResponseTime", rb2);
|
||||
|
||||
LOG.info("DecayedCallVolume: {}", decayedCallVolume1);
|
||||
LOG.info("CallVolume: {}", rawCallVolume1);
|
||||
LOG.info("UniqueCaller: {}", uniqueCaller1);
|
||||
LOG.info("Priority.0.CompletedCallVolume: {}", callVolumePriority0);
|
||||
LOG.info("Priority.1.CompletedCallVolume: {}", callVolumePriority1);
|
||||
LOG.info("Priority.0.AvgResponseTime: {}", avgRespTimePriority0);
|
||||
LOG.info("Priority.1.AvgResponseTime: {}", avgRespTimePriority1);
|
||||
|
||||
return decayedCallVolume1 > beginDecayedCallVolume &&
|
||||
rawCallVolume1 > beginRawCallVolume &&
|
||||
uniqueCaller1 > beginUniqueCaller;
|
||||
}, 30, 60000);
|
||||
} finally {
|
||||
stop(server, proxy);
|
||||
}
|
||||
}
|
||||
|
||||
private Server setupDecayRpcSchedulerandTestServer(String ns)
|
||||
throws Exception {
|
||||
final int queueSizePerHandler = 3;
|
||||
|
||||
@ -0,0 +1,86 @@
|
||||
/**
|
||||
* 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.ipc;
|
||||
|
||||
import java.util.concurrent.TimeUnit;
|
||||
import org.apache.hadoop.conf.Configuration;
|
||||
import org.apache.hadoop.ipc.ProcessingDetails.Timing;
|
||||
import org.junit.Before;
|
||||
import org.junit.Test;
|
||||
|
||||
import static org.apache.hadoop.ipc.WeightedTimeCostProvider.DEFAULT_LOCKEXCLUSIVE_WEIGHT;
|
||||
import static org.apache.hadoop.ipc.WeightedTimeCostProvider.DEFAULT_LOCKFREE_WEIGHT;
|
||||
import static org.apache.hadoop.ipc.WeightedTimeCostProvider.DEFAULT_LOCKSHARED_WEIGHT;
|
||||
import static org.junit.Assert.assertEquals;
|
||||
|
||||
/** Tests for {@link WeightedTimeCostProvider}. */
|
||||
public class TestWeightedTimeCostProvider {
|
||||
|
||||
private static final int QUEUE_TIME = 3;
|
||||
private static final int LOCKFREE_TIME = 5;
|
||||
private static final int LOCKSHARED_TIME = 7;
|
||||
private static final int LOCKEXCLUSIVE_TIME = 11;
|
||||
|
||||
private WeightedTimeCostProvider costProvider;
|
||||
private ProcessingDetails processingDetails;
|
||||
|
||||
@Before
|
||||
public void setup() {
|
||||
costProvider = new WeightedTimeCostProvider();
|
||||
processingDetails = new ProcessingDetails(TimeUnit.MILLISECONDS);
|
||||
processingDetails.set(Timing.QUEUE, QUEUE_TIME);
|
||||
processingDetails.set(Timing.LOCKFREE, LOCKFREE_TIME);
|
||||
processingDetails.set(Timing.LOCKSHARED, LOCKSHARED_TIME);
|
||||
processingDetails.set(Timing.LOCKEXCLUSIVE, LOCKEXCLUSIVE_TIME);
|
||||
}
|
||||
|
||||
@Test(expected = AssertionError.class)
|
||||
public void testGetCostBeforeInit() {
|
||||
costProvider.getCost(null);
|
||||
}
|
||||
|
||||
@Test
|
||||
public void testGetCostDefaultWeights() {
|
||||
costProvider.init("foo", new Configuration());
|
||||
long actualCost = costProvider.getCost(processingDetails);
|
||||
long expectedCost = DEFAULT_LOCKFREE_WEIGHT * LOCKFREE_TIME
|
||||
+ DEFAULT_LOCKSHARED_WEIGHT * LOCKSHARED_TIME
|
||||
+ DEFAULT_LOCKEXCLUSIVE_WEIGHT * LOCKEXCLUSIVE_TIME;
|
||||
assertEquals(expectedCost, actualCost);
|
||||
}
|
||||
|
||||
@Test
|
||||
public void testGetCostConfiguredWeights() {
|
||||
Configuration conf = new Configuration();
|
||||
int queueWeight = 1000;
|
||||
int lockfreeWeight = 10000;
|
||||
int locksharedWeight = 100000;
|
||||
conf.setInt("foo.weighted-cost.queue", queueWeight);
|
||||
conf.setInt("foo.weighted-cost.lockfree", lockfreeWeight);
|
||||
conf.setInt("foo.weighted-cost.lockshared", locksharedWeight);
|
||||
conf.setInt("bar.weighted-cost.lockexclusive", 0); // should not apply
|
||||
costProvider.init("foo", conf);
|
||||
long actualCost = costProvider.getCost(processingDetails);
|
||||
long expectedCost = queueWeight * QUEUE_TIME
|
||||
+ lockfreeWeight * LOCKFREE_TIME
|
||||
+ locksharedWeight * LOCKSHARED_TIME
|
||||
+ DEFAULT_LOCKEXCLUSIVE_WEIGHT * LOCKEXCLUSIVE_TIME;
|
||||
assertEquals(expectedCost, actualCost);
|
||||
}
|
||||
}
|
||||
Loading…
Reference in New Issue
Block a user