netconf/sal-netconf-connector/src/main/java/org/opendaylight/netconf/sal/connect/netconf/sal/KeepaliveSalFacade.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 package org.opendaylight.netconf.sal.connect.netconf.sal;
   9
  10 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfBaseOps.getSourceNode;
  11 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_GET_CONFIG_NODEID;
  12 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_GET_CONFIG_PATH;
  13 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_RUNNING_QNAME;
  14
  15 import com.google.common.base.Preconditions;
  16 import com.google.common.util.concurrent.FluentFuture;
  17 import com.google.common.util.concurrent.FutureCallback;
  18 import com.google.common.util.concurrent.Futures;
  19 import com.google.common.util.concurrent.MoreExecutors;
  20 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
  21 import java.util.concurrent.ScheduledExecutorService;
  22 import java.util.concurrent.ScheduledFuture;
  23 import java.util.concurrent.TimeUnit;
  24 import java.util.concurrent.atomic.AtomicBoolean;
  25 import javax.annotation.Nonnull;
  26 import javax.annotation.Nullable;
  27 import org.eclipse.jdt.annotation.NonNull;
  28 import org.opendaylight.mdsal.dom.api.DOMActionService;
  29 import org.opendaylight.mdsal.dom.api.DOMNotification;
  30 import org.opendaylight.mdsal.dom.api.DOMRpcAvailabilityListener;
  31 import org.opendaylight.mdsal.dom.api.DOMRpcResult;
  32 import org.opendaylight.mdsal.dom.api.DOMRpcService;
  33 import org.opendaylight.netconf.sal.connect.api.RemoteDeviceHandler;
  34 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfDeviceCommunicator;
  35 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfSessionPreferences;
  36 import org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil;
  37 import org.opendaylight.netconf.sal.connect.util.RemoteDeviceId;
  38 import org.opendaylight.yangtools.concepts.ListenerRegistration;
  39 import org.opendaylight.yangtools.yang.data.api.schema.ContainerNode;
  40 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
  41 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  42 import org.opendaylight.yangtools.yang.model.api.SchemaPath;
  43 import org.slf4j.Logger;
  44 import org.slf4j.LoggerFactory;
  45
  46 /**
  47  * SalFacade proxy that invokes keepalive RPCs to prevent session shutdown from remote device
  48  * and to detect incorrect session drops (netconf session is inactive, but TCP/SSH connection is still present).
  49  * The keepalive RPC is a get-config with empty filter.
  50  */
  51 public final class KeepaliveSalFacade implements RemoteDeviceHandler<NetconfSessionPreferences> {
  52
  53     private static final Logger LOG = LoggerFactory.getLogger(KeepaliveSalFacade.class);
  54
  55     // 2 minutes keepalive delay by default
  56     private static final long DEFAULT_DELAY = TimeUnit.MINUTES.toSeconds(2);
  57
  58     // 1 minute transaction timeout by default
  59     private static final long DEFAULT_TRANSACTION_TIMEOUT_MILLI = TimeUnit.MILLISECONDS.toMillis(60000);
  60
  61     private final RemoteDeviceId id;
  62     private final RemoteDeviceHandler<NetconfSessionPreferences> salFacade;
  63     private final ScheduledExecutorService executor;
  64     private final long keepaliveDelaySeconds;
  65     private final ResetKeepalive resetKeepaliveTask;
  66     private final long defaultRequestTimeoutMillis;
  67
  68     private volatile NetconfDeviceCommunicator listener;
  69     private volatile ScheduledFuture<?> currentKeepalive;
  70     private volatile DOMRpcService currentDeviceRpc;
  71     private final AtomicBoolean lastKeepAliveSucceeded = new AtomicBoolean(false);
  72
  73     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
  74                               final ScheduledExecutorService executor, final long keepaliveDelaySeconds,
  75                               final long defaultRequestTimeoutMillis) {
  76         this.id = id;
  77         this.salFacade = salFacade;
  78         this.executor = executor;
  79         this.keepaliveDelaySeconds = keepaliveDelaySeconds;
  80         this.defaultRequestTimeoutMillis = defaultRequestTimeoutMillis;
  81         this.resetKeepaliveTask = new ResetKeepalive();
  82     }
  83
  84     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
  85                               final ScheduledExecutorService executor) {
  86         this(id, salFacade, executor, DEFAULT_DELAY, DEFAULT_TRANSACTION_TIMEOUT_MILLI);
  87     }
  88
  89     /**
  90      * Set the netconf session listener whenever ready.
  91      *
  92      * @param listener netconf session listener
  93      */
  94     public void setListener(final NetconfDeviceCommunicator listener) {
  95         this.listener = listener;
  96     }
  97
  98     /**
  99      * Just cancel current keepalive task.
 100      * If its already started, let it finish ... not such a big deal.
 101      *
 102      * <p>
 103      * Then schedule next keepalive.
 104      */
 105     void resetKeepalive() {
 106         LOG.trace("{}: Resetting netconf keepalive timer", id);
 107         if (currentKeepalive != null) {
 108             currentKeepalive.cancel(false);
 109         }
 110         scheduleKeepalives();
 111     }
 112
 113     /**
 114      * Cancel current keepalive and also reset current deviceRpc.
 115      */
 116     private void stopKeepalives() {
 117         if (currentKeepalive != null) {
 118             currentKeepalive.cancel(false);
 119         }
 120         currentDeviceRpc = null;
 121     }
 122
 123     void reconnect() {
 124         Preconditions.checkState(listener != null, "%s: Unable to reconnect, session listener is missing", id);
 125         stopKeepalives();
 126         LOG.info("{}: Reconnecting inactive netconf session", id);
 127         listener.disconnect();
 128     }
 129
 130     @Override
 131     public void onDeviceConnected(final SchemaContext remoteSchemaContext,
 132                           final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc) {
 133         onDeviceConnected(remoteSchemaContext, netconfSessionPreferences, deviceRpc, null);
 134     }
 135
 136     @Override
 137     public void onDeviceConnected(final SchemaContext remoteSchemaContext,
 138             final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc,
 139             final DOMActionService deviceAction) {
 140         this.currentDeviceRpc = deviceRpc;
 141         final DOMRpcService deviceRpc1 =
 142                 new KeepaliveDOMRpcService(deviceRpc, resetKeepaliveTask, defaultRequestTimeoutMillis, executor,
 143                         new ResponseWaitingScheduler());
 144
 145         salFacade.onDeviceConnected(remoteSchemaContext, netconfSessionPreferences, deviceRpc1, deviceAction);
 146
 147         LOG.debug("{}: Netconf session initiated, starting keepalives", id);
 148         scheduleKeepalives();
 149     }
 150
 151     private void scheduleKeepalives() {
 152         lastKeepAliveSucceeded.set(true);
 153         Preconditions.checkState(currentDeviceRpc != null);
 154         LOG.trace("{}: Scheduling keepalives every  {} {}", id, keepaliveDelaySeconds, TimeUnit.SECONDS);
 155         currentKeepalive = executor.scheduleWithFixedDelay(new Keepalive(),
 156           keepaliveDelaySeconds, keepaliveDelaySeconds, TimeUnit.SECONDS);
 157     }
 158
 159     @Override
 160     public void onDeviceDisconnected() {
 161         stopKeepalives();
 162         salFacade.onDeviceDisconnected();
 163     }
 164
 165     @Override
 166     public void onDeviceFailed(final Throwable throwable) {
 167         stopKeepalives();
 168         salFacade.onDeviceFailed(throwable);
 169     }
 170
 171     @Override
 172     public void onNotification(final DOMNotification domNotification) {
 173         resetKeepalive();
 174         salFacade.onNotification(domNotification);
 175     }
 176
 177     @Override
 178     public void close() {
 179         stopKeepalives();
 180         salFacade.close();
 181     }
 182
 183     // Keepalive RPC static resources
 184     private static final ContainerNode KEEPALIVE_PAYLOAD = NetconfMessageTransformUtil.wrap(NETCONF_GET_CONFIG_NODEID,
 185             getSourceNode(NETCONF_RUNNING_QNAME), NetconfMessageTransformUtil.EMPTY_FILTER);
 186
 187     /**
 188      * Invoke keepalive RPC and check the response. In case of any received response the keepalive
 189      * is considered successful and schedules next keepalive with a fixed delay. If the response is unsuccessful (no
 190      * response received, or the rcp could not even be sent) immediate reconnect is triggered as netconf session
 191      * is considered inactive/failed.
 192      */
 193     private class Keepalive implements Runnable, FutureCallback<DOMRpcResult> {
 194
 195         @Override
 196         public void run() {
 197             LOG.trace("{}: Invoking keepalive RPC", id);
 198
 199             try {
 200                 final boolean lastJobSucceeded = lastKeepAliveSucceeded.getAndSet(false);
 201                 if (!lastJobSucceeded) {
 202                     onFailure(new IllegalStateException("Previous keepalive timed out"));
 203                 } else {
 204                     Futures.addCallback(currentDeviceRpc.invokeRpc(NETCONF_GET_CONFIG_PATH, KEEPALIVE_PAYLOAD), this,
 205                                         MoreExecutors.directExecutor());
 206                 }
 207             } catch (final NullPointerException e) {
 208                 LOG.debug("{}: Skipping keepalive while reconnecting", id);
 209                 // Empty catch block intentional
 210                 // Do nothing. The currentDeviceRpc was null and it means we hit the reconnect window and
 211                 // attempted to send keepalive while we were reconnecting. Next keepalive will be scheduled
 212                 // after reconnect so no action necessary here.
 213             }
 214         }
 215
 216         @SuppressFBWarnings(value = "RCN_REDUNDANT_NULLCHECK_OF_NONNULL_VALUE",
 217                 justification = "Unrecognised NullableDecl")
 218         @Override
 219         public void onSuccess(final DOMRpcResult result) {
 220             // No matter what response we got, rpc-reply or rpc-error,
 221             // we got it from device so the netconf session is OK
 222             if (result != null && result.getResult() != null) {
 223                 lastKeepAliveSucceeded.set(true);
 224             }  else if (result != null && result.getErrors() != null) {
 225                 LOG.warn("{}: Keepalive RPC failed with error: {}", id, result.getErrors());
 226                 lastKeepAliveSucceeded.set(true);
 227             } else {
 228                 LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 229                 reconnect();
 230             }
 231         }
 232
 233         @Override
 234         public void onFailure(@Nonnull final Throwable throwable) {
 235             LOG.warn("{}: Keepalive RPC failed. Reconnecting netconf session.", id, throwable);
 236             reconnect();
 237         }
 238     }
 239
 240     /**
 241      * Reset keepalive after each RPC response received.
 242      */
 243     private class ResetKeepalive implements FutureCallback<DOMRpcResult> {
 244         @Override
 245         public void onSuccess(@Nullable final DOMRpcResult result) {
 246             // No matter what response we got,
 247             // rpc-reply or rpc-error, we got it from device so the netconf session is OK.
 248             resetKeepalive();
 249         }
 250
 251         @Override
 252         public void onFailure(@Nonnull final Throwable throwable) {
 253             // User/Application RPC failed (The RPC did not reach the remote device or ..
 254             // TODO what other reasons could cause this ?)
 255             // There is no point in keeping this session. Reconnect.
 256             LOG.warn("{}: Rpc failure detected. Reconnecting netconf session", id, throwable);
 257             reconnect();
 258         }
 259     }
 260
 261     private final class ResponseWaitingScheduler {
 262
 263         private ScheduledFuture<?> schedule;
 264
 265         public void initScheduler(final Runnable runnable) {
 266             if (currentKeepalive != null) {
 267                 currentKeepalive.cancel(true);
 268             } else {
 269                 LOG.trace("Keepalive does not exist.");
 270             }
 271             scheduleKeepalives();
 272             //Listening on the result should be done before the keepalive rpc will be send
 273             final long delay = (keepaliveDelaySeconds * 1000) - 500;
 274             schedule = executor.schedule(runnable, delay, TimeUnit.MILLISECONDS);
 275         }
 276
 277         public void stopScheduler() {
 278             if (schedule != null) {
 279                 schedule.cancel(true);
 280             } else {
 281                 LOG.trace("Scheduler does not exist.");
 282             }
 283         }
 284     }
 285
 286     private static final class ResponseWaiting implements Runnable {
 287
 288         private final FluentFuture<DOMRpcResult> rpcResultFuture;
 289         private final ResponseWaitingScheduler responseWaitingScheduler;
 290
 291         ResponseWaiting(final ResponseWaitingScheduler responseWaitingScheduler,
 292                 final FluentFuture<DOMRpcResult> rpcResultFuture) {
 293             this.responseWaitingScheduler = responseWaitingScheduler;
 294             this.rpcResultFuture = rpcResultFuture;
 295         }
 296
 297         public void start() {
 298             LOG.trace("Start to waiting for result.");
 299             responseWaitingScheduler.initScheduler(this);
 300         }
 301
 302         public void stop() {
 303             LOG.info("Stop to waiting for result.");
 304             responseWaitingScheduler.stopScheduler();
 305         }
 306
 307         @Override
 308         public void run() {
 309             if (!rpcResultFuture.isCancelled() && !rpcResultFuture.isDone()) {
 310                 LOG.trace("Waiting for result");
 311                 responseWaitingScheduler.initScheduler(this);
 312             } else {
 313                 LOG.trace("Result has been cancelled or done.");
 314             }
 315         }
 316     }
 317
 318     /*
 319      * Request timeout task is called once the defaultRequestTimeoutMillis is
 320      * reached. At this moment, if the request is not yet finished, we cancel
 321      * it.
 322      */
 323     private static final class RequestTimeoutTask implements Runnable {
 324
 325         private final FluentFuture<DOMRpcResult> rpcResultFuture;
 326         private final ResponseWaiting responseWaiting;
 327
 328         RequestTimeoutTask(final FluentFuture<DOMRpcResult> rpcResultFuture, final ResponseWaiting responseWaiting) {
 329             this.rpcResultFuture = rpcResultFuture;
 330             this.responseWaiting = responseWaiting;
 331         }
 332
 333         @Override
 334         public void run() {
 335             if (!rpcResultFuture.isDone()) {
 336                 rpcResultFuture.cancel(true);
 337             }
 338             if (responseWaiting != null) {
 339                 responseWaiting.stop();
 340             }
 341         }
 342     }
 343
 344     /**
 345      * DOMRpcService proxy that attaches reset-keepalive-task and schedule
 346      * request-timeout-task to each RPC invocation.
 347      */
 348     public static final class KeepaliveDOMRpcService implements DOMRpcService {
 349
 350         private final DOMRpcService deviceRpc;
 351         private final ResetKeepalive resetKeepaliveTask;
 352         private final long defaultRequestTimeoutMillis;
 353         private final ScheduledExecutorService executor;
 354         private final ResponseWaitingScheduler responseWaitingScheduler;
 355
 356         KeepaliveDOMRpcService(final DOMRpcService deviceRpc, final ResetKeepalive resetKeepaliveTask,
 357                 final long defaultRequestTimeoutMillis, final ScheduledExecutorService executor,
 358                 final ResponseWaitingScheduler responseWaitingScheduler) {
 359             this.deviceRpc = deviceRpc;
 360             this.resetKeepaliveTask = resetKeepaliveTask;
 361             this.defaultRequestTimeoutMillis = defaultRequestTimeoutMillis;
 362             this.executor = executor;
 363             this.responseWaitingScheduler = responseWaitingScheduler;
 364         }
 365
 366         public DOMRpcService getDeviceRpc() {
 367             return deviceRpc;
 368         }
 369
 370         @Nonnull
 371         @Override
 372         public @NonNull FluentFuture<DOMRpcResult> invokeRpc(@Nonnull final SchemaPath type,
 373                                                                       final NormalizedNode<?, ?> input) {
 374             final FluentFuture<DOMRpcResult> rpcResultFuture = deviceRpc.invokeRpc(type, input);
 375             final ResponseWaiting responseWaiting = new ResponseWaiting(responseWaitingScheduler, rpcResultFuture);
 376             responseWaiting.start();
 377             rpcResultFuture.addCallback(resetKeepaliveTask, MoreExecutors.directExecutor());
 378
 379             final RequestTimeoutTask timeoutTask = new RequestTimeoutTask(rpcResultFuture, responseWaiting);
 380             executor.schedule(timeoutTask, defaultRequestTimeoutMillis, TimeUnit.MILLISECONDS);
 381
 382             return rpcResultFuture;
 383         }
 384
 385         @Override
 386         public <T extends DOMRpcAvailabilityListener> ListenerRegistration<T> registerRpcListener(
 387                 @Nonnull final T listener) {
 388             // There is no real communication with the device (yet), no reset here
 389             return deviceRpc.registerRpcListener(listener);
 390         }
 391     }
 392 }