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 com.google.common.base.Preconditions.checkState;
  11 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfBaseOps.getSourceNode;
  12 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_GET_CONFIG_NODEID;
  13 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_GET_CONFIG_QNAME;
  14 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_RUNNING_QNAME;
  15
  16 import com.google.common.util.concurrent.FutureCallback;
  17 import com.google.common.util.concurrent.Futures;
  18 import com.google.common.util.concurrent.ListenableFuture;
  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 org.opendaylight.mdsal.dom.api.DOMActionService;
  26 import org.opendaylight.mdsal.dom.api.DOMNotification;
  27 import org.opendaylight.mdsal.dom.api.DOMRpcAvailabilityListener;
  28 import org.opendaylight.mdsal.dom.api.DOMRpcResult;
  29 import org.opendaylight.mdsal.dom.api.DOMRpcService;
  30 import org.opendaylight.netconf.sal.connect.api.RemoteDeviceHandler;
  31 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfDeviceCommunicator;
  32 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfSessionPreferences;
  33 import org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil;
  34 import org.opendaylight.netconf.sal.connect.util.RemoteDeviceId;
  35 import org.opendaylight.yangtools.concepts.ListenerRegistration;
  36 import org.opendaylight.yangtools.rfc8528.data.api.MountPointContext;
  37 import org.opendaylight.yangtools.yang.common.QName;
  38 import org.opendaylight.yangtools.yang.data.api.schema.ContainerNode;
  39 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
  40 import org.slf4j.Logger;
  41 import org.slf4j.LoggerFactory;
  42
  43 /**
  44  * SalFacade proxy that invokes keepalive RPCs to prevent session shutdown from remote device
  45  * and to detect incorrect session drops (netconf session is inactive, but TCP/SSH connection is still present).
  46  * The keepalive RPC is a get-config with empty filter.
  47  */
  48 public final class KeepaliveSalFacade implements RemoteDeviceHandler<NetconfSessionPreferences> {
  49
  50     private static final Logger LOG = LoggerFactory.getLogger(KeepaliveSalFacade.class);
  51
  52     // 2 minutes keepalive delay by default
  53     private static final long DEFAULT_DELAY = TimeUnit.MINUTES.toSeconds(2);
  54
  55     // 1 minute transaction timeout by default
  56     private static final long DEFAULT_TRANSACTION_TIMEOUT_MILLI = TimeUnit.MILLISECONDS.toMillis(60000);
  57
  58     private final RemoteDeviceId id;
  59     private final RemoteDeviceHandler<NetconfSessionPreferences> salFacade;
  60     private final ScheduledExecutorService executor;
  61     private final long keepaliveDelaySeconds;
  62     private final ResetKeepalive resetKeepaliveTask;
  63     private final long defaultRequestTimeoutMillis;
  64
  65     private volatile NetconfDeviceCommunicator listener;
  66     private volatile ScheduledFuture<?> currentKeepalive;
  67     private volatile DOMRpcService currentDeviceRpc;
  68     private final AtomicBoolean lastKeepAliveSucceeded = new AtomicBoolean(false);
  69
  70     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
  71                               final ScheduledExecutorService executor, final long keepaliveDelaySeconds,
  72                               final long defaultRequestTimeoutMillis) {
  73         this.id = id;
  74         this.salFacade = salFacade;
  75         this.executor = executor;
  76         this.keepaliveDelaySeconds = keepaliveDelaySeconds;
  77         this.defaultRequestTimeoutMillis = defaultRequestTimeoutMillis;
  78         this.resetKeepaliveTask = new ResetKeepalive();
  79     }
  80
  81     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
  82                               final ScheduledExecutorService executor) {
  83         this(id, salFacade, executor, DEFAULT_DELAY, DEFAULT_TRANSACTION_TIMEOUT_MILLI);
  84     }
  85
  86     /**
  87      * Set the netconf session listener whenever ready.
  88      *
  89      * @param listener netconf session listener
  90      */
  91     public void setListener(final NetconfDeviceCommunicator listener) {
  92         this.listener = listener;
  93     }
  94
  95     /**
  96      * Just cancel current keepalive task if exists.
  97      * If its already started, let it finish ... not such a big deal.
  98      *
  99      * <p>
 100      * Then schedule next keepalive.
 101      */
 102     synchronized void resetKeepalive() {
 103         LOG.trace("{}: Resetting netconf keepalive timer", id);
 104         if (currentKeepalive != null) {
 105             currentKeepalive.cancel(false);
 106         } else {
 107             LOG.trace("{}: Keepalive does not exist", id);
 108         }
 109         scheduleKeepalives();
 110     }
 111
 112     /**
 113      * Cancel current keepalive and also reset current deviceRpc.
 114      */
 115     private synchronized void stopKeepalives() {
 116         if (currentKeepalive != null) {
 117             currentKeepalive.cancel(false);
 118         }
 119         currentDeviceRpc = null;
 120     }
 121
 122     void reconnect() {
 123         checkState(listener != null, "%s: Unable to reconnect, session listener is missing", id);
 124         stopKeepalives();
 125         LOG.info("{}: Reconnecting inactive netconf session", id);
 126         listener.disconnect();
 127     }
 128
 129     @Override
 130     public void onDeviceConnected(final MountPointContext remoteSchemaContext,
 131                           final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc) {
 132         onDeviceConnected(remoteSchemaContext, netconfSessionPreferences, deviceRpc, null);
 133     }
 134
 135     @Override
 136     public void onDeviceConnected(final MountPointContext remoteSchemaContext,
 137             final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc,
 138             final DOMActionService deviceAction) {
 139         this.currentDeviceRpc = deviceRpc;
 140         final DOMRpcService deviceRpc1 =
 141                 new KeepaliveDOMRpcService(deviceRpc, resetKeepaliveTask, defaultRequestTimeoutMillis, executor,
 142                         new ResponseWaitingScheduler());
 143
 144         salFacade.onDeviceConnected(remoteSchemaContext, netconfSessionPreferences, deviceRpc1, deviceAction);
 145
 146         LOG.debug("{}: Netconf session initiated, starting keepalives", id);
 147         resetKeepalive();
 148     }
 149
 150     private void scheduleKeepalives() {
 151         lastKeepAliveSucceeded.set(true);
 152         checkState(currentDeviceRpc != null);
 153         LOG.trace("{}: Scheduling keepalives every  {} {}", id, keepaliveDelaySeconds, TimeUnit.SECONDS);
 154         currentKeepalive = executor.scheduleWithFixedDelay(new Keepalive(),
 155           keepaliveDelaySeconds, keepaliveDelaySeconds, TimeUnit.SECONDS);
 156     }
 157
 158     @Override
 159     public void onDeviceDisconnected() {
 160         stopKeepalives();
 161         salFacade.onDeviceDisconnected();
 162     }
 163
 164     @Override
 165     public void onDeviceFailed(final Throwable throwable) {
 166         stopKeepalives();
 167         salFacade.onDeviceFailed(throwable);
 168     }
 169
 170     @Override
 171     public void onNotification(final DOMNotification domNotification) {
 172         resetKeepalive();
 173         salFacade.onNotification(domNotification);
 174     }
 175
 176     @Override
 177     public void close() {
 178         stopKeepalives();
 179         salFacade.close();
 180     }
 181
 182     // Keepalive RPC static resources
 183     private static final ContainerNode KEEPALIVE_PAYLOAD = NetconfMessageTransformUtil.wrap(NETCONF_GET_CONFIG_NODEID,
 184             getSourceNode(NETCONF_RUNNING_QNAME), NetconfMessageTransformUtil.EMPTY_FILTER);
 185
 186     /**
 187      * Invoke keepalive RPC and check the response. In case of any received response the keepalive
 188      * is considered successful and schedules next keepalive with a fixed delay. If the response is unsuccessful (no
 189      * response received, or the rcp could not even be sent) immediate reconnect is triggered as netconf session
 190      * is considered inactive/failed.
 191      */
 192     private class Keepalive implements Runnable, FutureCallback<DOMRpcResult> {
 193
 194         @Override
 195         public void run() {
 196             LOG.trace("{}: Invoking keepalive RPC", id);
 197
 198             try {
 199                 final boolean lastJobSucceeded = lastKeepAliveSucceeded.getAndSet(false);
 200                 if (!lastJobSucceeded) {
 201                     onFailure(new IllegalStateException("Previous keepalive timed out"));
 202                 } else {
 203                     Futures.addCallback(currentDeviceRpc.invokeRpc(NETCONF_GET_CONFIG_QNAME, KEEPALIVE_PAYLOAD), this,
 204                         MoreExecutors.directExecutor());
 205                 }
 206             } catch (final NullPointerException e) {
 207                 LOG.debug("{}: Skipping keepalive while reconnecting", id);
 208                 // Empty catch block intentional
 209                 // Do nothing. The currentDeviceRpc was null and it means we hit the reconnect window and
 210                 // attempted to send keepalive while we were reconnecting. Next keepalive will be scheduled
 211                 // after reconnect so no action necessary here.
 212             }
 213         }
 214
 215         @SuppressFBWarnings(value = "RCN_REDUNDANT_NULLCHECK_OF_NONNULL_VALUE",
 216                 justification = "Unrecognised NullableDecl")
 217         @Override
 218         public void onSuccess(final DOMRpcResult result) {
 219             // No matter what response we got, rpc-reply or rpc-error,
 220             // we got it from device so the netconf session is OK
 221             if (result == null) {
 222                 LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 223                 reconnect();
 224                 return;
 225             }
 226
 227             if (result.getResult() != null) {
 228                 lastKeepAliveSucceeded.set(true);
 229             }  else if (result.getErrors() != null) {
 230                 LOG.warn("{}: Keepalive RPC failed with error: {}", id, result.getErrors());
 231                 lastKeepAliveSucceeded.set(true);
 232             } else {
 233                 LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 234                 reconnect();
 235             }
 236         }
 237
 238         @Override
 239         public void onFailure(final Throwable throwable) {
 240             LOG.warn("{}: Keepalive RPC failed. Reconnecting netconf session.", id, throwable);
 241             reconnect();
 242         }
 243     }
 244
 245     /**
 246      * Reset keepalive after each RPC response received.
 247      */
 248     private class ResetKeepalive implements FutureCallback<DOMRpcResult> {
 249         @Override
 250         public void onSuccess(final DOMRpcResult result) {
 251             // No matter what response we got,
 252             // rpc-reply or rpc-error, we got it from device so the netconf session is OK.
 253             resetKeepalive();
 254         }
 255
 256         @Override
 257         public void onFailure(final Throwable throwable) {
 258             // User/Application RPC failed (The RPC did not reach the remote device or ..
 259             // TODO what other reasons could cause this ?)
 260             // There is no point in keeping this session. Reconnect.
 261             LOG.warn("{}: Rpc failure detected. Reconnecting netconf session", id, throwable);
 262             reconnect();
 263         }
 264     }
 265
 266     private final class ResponseWaitingScheduler {
 267
 268         private ScheduledFuture<?> schedule;
 269
 270         public void initScheduler(final Runnable runnable) {
 271             resetKeepalive();
 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 ListenableFuture<? extends DOMRpcResult> rpcResultFuture;
 289         private final ResponseWaitingScheduler responseWaitingScheduler;
 290
 291         ResponseWaiting(final ResponseWaitingScheduler responseWaitingScheduler,
 292                 final ListenableFuture<? extends 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         private final ListenableFuture<? extends DOMRpcResult> rpcResultFuture;
 325         private final ResponseWaiting responseWaiting;
 326
 327         RequestTimeoutTask(final ListenableFuture<? extends DOMRpcResult> rpcResultFuture,
 328                 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         private final DOMRpcService deviceRpc;
 350         private final ResetKeepalive resetKeepaliveTask;
 351         private final long defaultRequestTimeoutMillis;
 352         private final ScheduledExecutorService executor;
 353         private final ResponseWaitingScheduler responseWaitingScheduler;
 354
 355         KeepaliveDOMRpcService(final DOMRpcService deviceRpc, final ResetKeepalive resetKeepaliveTask,
 356                 final long defaultRequestTimeoutMillis, final ScheduledExecutorService executor,
 357                 final ResponseWaitingScheduler responseWaitingScheduler) {
 358             this.deviceRpc = deviceRpc;
 359             this.resetKeepaliveTask = resetKeepaliveTask;
 360             this.defaultRequestTimeoutMillis = defaultRequestTimeoutMillis;
 361             this.executor = executor;
 362             this.responseWaitingScheduler = responseWaitingScheduler;
 363         }
 364
 365         public DOMRpcService getDeviceRpc() {
 366             return deviceRpc;
 367         }
 368
 369         @Override
 370         public ListenableFuture<? extends DOMRpcResult> invokeRpc(final QName type, final NormalizedNode<?, ?> input) {
 371             final ListenableFuture<? extends DOMRpcResult> rpcResultFuture = deviceRpc.invokeRpc(type, input);
 372             final ResponseWaiting responseWaiting = new ResponseWaiting(responseWaitingScheduler, rpcResultFuture);
 373             responseWaiting.start();
 374             Futures.addCallback(rpcResultFuture, resetKeepaliveTask, MoreExecutors.directExecutor());
 375
 376             final RequestTimeoutTask timeoutTask = new RequestTimeoutTask(rpcResultFuture, responseWaiting);
 377             executor.schedule(timeoutTask, defaultRequestTimeoutMillis, TimeUnit.MILLISECONDS);
 378
 379             return rpcResultFuture;
 380         }
 381
 382         @Override
 383         public <T extends DOMRpcAvailabilityListener> ListenerRegistration<T> registerRpcListener(final T listener) {
 384             // There is no real communication with the device (yet), no reset here
 385             return deviceRpc.registerRpcListener(listener);
 386         }
 387     }
 388 }