Bug 5596 Created lifecycle service
[openflowplugin.git] / openflowplugin-impl / src / main / java / org / opendaylight / openflowplugin / impl / statistics / StatisticsManagerImpl.java
1 /*
2  * Copyright (c) 2015 Cisco Systems, Inc. and others.  All rights reserved.
3  *
4  * This program and the accompanying materials are made available under the
5  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
6  * and is available at http://www.eclipse.org/legal/epl-v10.html
7  */
8
9 package org.opendaylight.openflowplugin.impl.statistics;
10
11 import com.google.common.annotations.VisibleForTesting;
12 import com.google.common.base.Preconditions;
13 import com.google.common.base.Verify;
14 import com.google.common.collect.Iterators;
15 import com.google.common.util.concurrent.FutureCallback;
16 import com.google.common.util.concurrent.Futures;
17 import com.google.common.util.concurrent.ListenableFuture;
18 import io.netty.util.Timeout;
19 import io.netty.util.TimerTask;
20 import java.util.Iterator;
21 import java.util.Map;
22 import java.util.Optional;
23 import java.util.concurrent.CancellationException;
24 import java.util.concurrent.ConcurrentHashMap;
25 import java.util.concurrent.ConcurrentMap;
26 import java.util.concurrent.Future;
27 import java.util.concurrent.Semaphore;
28 import java.util.concurrent.TimeUnit;
29 import javax.annotation.CheckForNull;
30 import javax.annotation.Nonnull;
31 import org.opendaylight.controller.sal.binding.api.BindingAwareBroker;
32 import org.opendaylight.controller.sal.binding.api.RpcProviderRegistry;
33 import org.opendaylight.openflowplugin.api.openflow.OFPContext;
34 import org.opendaylight.openflowplugin.api.openflow.device.DeviceContext;
35 import org.opendaylight.openflowplugin.api.openflow.device.DeviceInfo;
36 import org.opendaylight.openflowplugin.api.openflow.device.DeviceState;
37 import org.opendaylight.openflowplugin.api.openflow.device.handlers.DeviceInitializationPhaseHandler;
38 import org.opendaylight.openflowplugin.api.openflow.device.handlers.DeviceTerminationPhaseHandler;
39 import org.opendaylight.openflowplugin.api.openflow.lifecycle.LifecycleConductor;
40 import org.opendaylight.openflowplugin.api.openflow.rpc.ItemLifeCycleSource;
41 import org.opendaylight.openflowplugin.api.openflow.statistics.StatisticsContext;
42 import org.opendaylight.openflowplugin.api.openflow.statistics.StatisticsManager;
43 import org.opendaylight.openflowplugin.openflow.md.core.sal.convertor.ConvertorExecutor;
44 import org.opendaylight.yang.gen.v1.urn.opendaylight.params.xml.ns.yang.openflowplugin.sm.control.rev150812.ChangeStatisticsWorkModeInput;
45 import org.opendaylight.yang.gen.v1.urn.opendaylight.params.xml.ns.yang.openflowplugin.sm.control.rev150812.GetStatisticsWorkModeOutput;
46 import org.opendaylight.yang.gen.v1.urn.opendaylight.params.xml.ns.yang.openflowplugin.sm.control.rev150812.GetStatisticsWorkModeOutputBuilder;
47 import org.opendaylight.yang.gen.v1.urn.opendaylight.params.xml.ns.yang.openflowplugin.sm.control.rev150812.StatisticsManagerControlService;
48 import org.opendaylight.yang.gen.v1.urn.opendaylight.params.xml.ns.yang.openflowplugin.sm.control.rev150812.StatisticsWorkMode;
49 import org.opendaylight.yangtools.yang.common.RpcError;
50 import org.opendaylight.yangtools.yang.common.RpcResult;
51 import org.opendaylight.yangtools.yang.common.RpcResultBuilder;
52 import org.slf4j.Logger;
53 import org.slf4j.LoggerFactory;
54
55 public class StatisticsManagerImpl implements StatisticsManager, StatisticsManagerControlService {
56
57     private static final Logger LOG = LoggerFactory.getLogger(StatisticsManagerImpl.class);
58
59     private static final long DEFAULT_STATS_TIMEOUT_SEC = 50L;
60     private final ConvertorExecutor convertorExecutor;
61
62     private DeviceInitializationPhaseHandler deviceInitPhaseHandler;
63     private DeviceTerminationPhaseHandler deviceTerminPhaseHandler;
64
65     private final ConcurrentMap<DeviceInfo, StatisticsContext> contexts = new ConcurrentHashMap<>();
66
67     private static final long basicTimerDelay = 3000;
68     private static long currentTimerDelay = basicTimerDelay;
69     private static final long maximumTimerDelay = 900000; //wait max 15 minutes for next statistics
70
71     private StatisticsWorkMode workMode = StatisticsWorkMode.COLLECTALL;
72     private final Semaphore workModeGuard = new Semaphore(1, true);
73     private boolean shuttingDownStatisticsPolling;
74     private BindingAwareBroker.RpcRegistration<StatisticsManagerControlService> controlServiceRegistration;
75
76     private final LifecycleConductor conductor;
77
78     @Override
79     public void setDeviceInitializationPhaseHandler(final DeviceInitializationPhaseHandler handler) {
80         deviceInitPhaseHandler = handler;
81     }
82
83     public StatisticsManagerImpl(@CheckForNull final RpcProviderRegistry rpcProviderRegistry,
84                                  final boolean shuttingDownStatisticsPolling,
85                                  final LifecycleConductor lifecycleConductor,
86                                  final ConvertorExecutor convertorExecutor) {
87         this.convertorExecutor = convertorExecutor;
88         Preconditions.checkArgument(rpcProviderRegistry != null);
89         this.controlServiceRegistration = Preconditions.checkNotNull(rpcProviderRegistry.addRpcImplementation(
90                 StatisticsManagerControlService.class, this));
91         this.shuttingDownStatisticsPolling = shuttingDownStatisticsPolling;
92         this.conductor = lifecycleConductor;
93     }
94
95     @Override
96     public void onDeviceContextLevelUp(final DeviceInfo deviceInfo) throws Exception {
97
98         final StatisticsContext statisticsContext = new StatisticsContextImpl(deviceInfo, shuttingDownStatisticsPolling, conductor, convertorExecutor);
99         Verify.verify(contexts.putIfAbsent(deviceInfo, statisticsContext) == null, "StatisticsCtx still not closed for Node {}", deviceInfo.getNodeId());
100
101         deviceInitPhaseHandler.onDeviceContextLevelUp(deviceInfo);
102     }
103
104     @VisibleForTesting
105     void pollStatistics(final DeviceState deviceState,
106                         final StatisticsContext statisticsContext,
107                         final TimeCounter timeCounter,
108                         final DeviceInfo deviceInfo) {
109
110         if (!statisticsContext.isSchedulingEnabled()) {
111             LOG.debug("Disabling statistics scheduling for device: {}", deviceInfo.getNodeId());
112             return;
113         }
114         
115         if (!deviceState.isValid()) {
116             LOG.debug("Session is not valid for device: {}", deviceInfo.getNodeId());
117             return;
118         }
119
120         if (!deviceState.isStatisticsPollingEnabled()) {
121             LOG.debug("Statistics polling is currently disabled for device: {}", deviceInfo.getNodeId());
122             scheduleNextPolling(deviceState, deviceInfo, statisticsContext, timeCounter);
123             return;
124         }
125
126         LOG.debug("POLLING ALL STATISTICS for device: {}", deviceInfo.getNodeId());
127         timeCounter.markStart();
128         final ListenableFuture<Boolean> deviceStatisticsCollectionFuture = statisticsContext.gatherDynamicData();
129         Futures.addCallback(deviceStatisticsCollectionFuture, new FutureCallback<Boolean>() {
130             @Override
131             public void onSuccess(final Boolean o) {
132                 timeCounter.addTimeMark();
133                 calculateTimerDelay(timeCounter);
134                 scheduleNextPolling(deviceState, deviceInfo, statisticsContext, timeCounter);
135             }
136
137             @Override
138             public void onFailure(@Nonnull final Throwable throwable) {
139                 timeCounter.addTimeMark();
140                 LOG.warn("Statistics gathering for single node was not successful: {}", throwable.getMessage());
141                 LOG.trace("Statistics gathering for single node was not successful.. ", throwable);
142                 calculateTimerDelay(timeCounter);
143                 if (throwable instanceof CancellationException) {
144                     /* This often happens when something wrong with akka or DS, so closing connection will help to restart device **/
145                     conductor.closeConnection(deviceInfo);
146                 } else {
147                     scheduleNextPolling(deviceState, deviceInfo, statisticsContext, timeCounter);
148                 }
149             }
150         });
151
152         final long averageTime = TimeUnit.MILLISECONDS.toSeconds(timeCounter.getAverageTimeBetweenMarks());
153         final long STATS_TIMEOUT_SEC = averageTime > 0 ? 3 * averageTime : DEFAULT_STATS_TIMEOUT_SEC;
154         final TimerTask timerTask = timeout -> {
155             if (!deviceStatisticsCollectionFuture.isDone()) {
156                 LOG.info("Statistics collection for node {} still in progress even after {} secs", deviceInfo.getNodeId(), STATS_TIMEOUT_SEC);
157                 deviceStatisticsCollectionFuture.cancel(true);
158             }
159         };
160
161         conductor.newTimeout(timerTask, STATS_TIMEOUT_SEC, TimeUnit.SECONDS);
162     }
163
164     private void scheduleNextPolling(final DeviceState deviceState,
165                                      final DeviceInfo deviceInfo,
166                                      final StatisticsContext statisticsContext,
167                                      final TimeCounter timeCounter) {
168         LOG.debug("SCHEDULING NEXT STATISTICS POLLING for device: {}", deviceInfo.getNodeId());
169         if (!shuttingDownStatisticsPolling) {
170             final Timeout pollTimeout = conductor.newTimeout(timeout -> pollStatistics(deviceState, statisticsContext, timeCounter, deviceInfo), currentTimerDelay, TimeUnit.MILLISECONDS);
171             statisticsContext.setPollTimeout(pollTimeout);
172         }
173     }
174
175     @VisibleForTesting
176     void calculateTimerDelay(final TimeCounter timeCounter) {
177         final long averageStatisticsGatheringTime = timeCounter.getAverageTimeBetweenMarks();
178         if (averageStatisticsGatheringTime > currentTimerDelay) {
179             currentTimerDelay *= 2;
180             if (currentTimerDelay > maximumTimerDelay) {
181                 currentTimerDelay = maximumTimerDelay;
182             }
183         } else {
184             if (currentTimerDelay > basicTimerDelay) {
185                 currentTimerDelay /= 2;
186             } else {
187                 currentTimerDelay = basicTimerDelay;
188             }
189         }
190     }
191
192     @VisibleForTesting
193     static long getCurrentTimerDelay() {
194         return currentTimerDelay;
195     }
196
197     @Override
198     public void onDeviceContextLevelDown(final DeviceInfo deviceInfo) {
199         final StatisticsContext statisticsContext = contexts.remove(deviceInfo);
200         if (null != statisticsContext) {
201             LOG.trace("Removing device context from stack. No more statistics gathering for device: {}", deviceInfo.getNodeId());
202             statisticsContext.close();
203         }
204         deviceTerminPhaseHandler.onDeviceContextLevelDown(deviceInfo);
205     }
206
207     @Override
208     public Future<RpcResult<GetStatisticsWorkModeOutput>> getStatisticsWorkMode() {
209         final GetStatisticsWorkModeOutputBuilder smModeOutputBld = new GetStatisticsWorkModeOutputBuilder();
210         smModeOutputBld.setMode(workMode);
211         return RpcResultBuilder.success(smModeOutputBld.build()).buildFuture();
212     }
213
214     @Override
215     public Future<RpcResult<Void>> changeStatisticsWorkMode(ChangeStatisticsWorkModeInput input) {
216         final Future<RpcResult<Void>> result;
217         // acquire exclusive access
218         if (workModeGuard.tryAcquire()) {
219             final StatisticsWorkMode targetWorkMode = input.getMode();
220             if (!workMode.equals(targetWorkMode)) {
221                 shuttingDownStatisticsPolling = StatisticsWorkMode.FULLYDISABLED.equals(targetWorkMode);
222                 // iterate through stats-ctx: propagate mode
223                 for (Map.Entry<DeviceInfo, StatisticsContext> entry : contexts.entrySet()) {
224                     switch (targetWorkMode) {
225                         case COLLECTALL:
226                             scheduleNextPolling(conductor.getDeviceContext(entry.getKey()).getDeviceState(), entry.getKey(), entry.getValue(), new TimeCounter());
227                             for (final ItemLifeCycleSource lifeCycleSource : conductor.getDeviceContext(entry.getKey()).getItemLifeCycleSourceRegistry().getLifeCycleSources()) {
228                                 lifeCycleSource.setItemLifecycleListener(null);
229                             }
230                             break;
231                         case FULLYDISABLED:
232                             final Optional<Timeout> pollTimeout = entry.getValue().getPollTimeout();
233                             if (pollTimeout.isPresent()) {
234                                 pollTimeout.get().cancel();
235                             }
236                             for (final ItemLifeCycleSource lifeCycleSource : conductor.getDeviceContext(entry.getKey()).getItemLifeCycleSourceRegistry().getLifeCycleSources()) {
237                                 lifeCycleSource.setItemLifecycleListener(entry.getValue().getItemLifeCycleListener());
238                             }
239                             break;
240                         default:
241                             LOG.warn("Statistics work mode not supported: {}", targetWorkMode);
242                     }
243                 }
244                 workMode = targetWorkMode;
245             }
246             workModeGuard.release();
247             result = RpcResultBuilder.<Void>success().buildFuture();
248         } else {
249             result = RpcResultBuilder.<Void>failed()
250                     .withError(RpcError.ErrorType.APPLICATION, "mode change already in progress")
251                     .buildFuture();
252         }
253         return result;
254     }
255
256     @Override
257     public void startScheduling(final DeviceInfo deviceInfo) {
258         if (shuttingDownStatisticsPolling) {
259             LOG.info("Statistics are shut down for device: {}", deviceInfo.getNodeId());
260             return;
261         }
262
263         final StatisticsContext statisticsContext = contexts.get(deviceInfo);
264
265         if (statisticsContext == null) {
266             LOG.warn("Statistics context not found for device: {}", deviceInfo.getNodeId());
267             return;
268         }
269
270         if (statisticsContext.isSchedulingEnabled()) {
271             LOG.debug("Statistics scheduling is already enabled for device: {}", deviceInfo.getNodeId());
272             return;
273         }
274
275         LOG.info("Scheduling statistics poll for device: {}", deviceInfo.getNodeId());
276
277         statisticsContext.setSchedulingEnabled(true);
278         scheduleNextPolling(conductor.getDeviceContext(deviceInfo).getDeviceState(), deviceInfo, statisticsContext, new TimeCounter());
279     }
280
281     @Override
282     public void stopScheduling(final DeviceInfo deviceInfo) {
283         LOG.debug("Stopping statistics scheduling for device: {}", deviceInfo.getNodeId());
284         final StatisticsContext statisticsContext = contexts.get(deviceInfo);
285
286         if (statisticsContext == null) {
287             LOG.warn("Statistics context not found for device: {}", deviceInfo.getNodeId());
288             return;
289         }
290
291         statisticsContext.setSchedulingEnabled(false);
292     }
293
294     @Override
295     public void close() {
296         if (controlServiceRegistration != null) {
297             controlServiceRegistration.close();
298             controlServiceRegistration = null;
299         }
300         for (final Iterator<StatisticsContext> iterator = Iterators.consumingIterator(contexts.values().iterator());
301                 iterator.hasNext();) {
302             iterator.next().close();
303         }
304     }
305
306     @Override
307     public void setDeviceTerminationPhaseHandler(final DeviceTerminationPhaseHandler handler) {
308         this.deviceTerminPhaseHandler = handler;
309     }
310
311     @Override
312     public <T extends OFPContext> T gainContext(DeviceInfo deviceInfo) {
313         return (T) contexts.get(deviceInfo);
314     }
315 }