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 java.util.Objects.requireNonNull;
  12 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfBaseOps.getSourceNode;
  13 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_GET_CONFIG_NODEID;
  14 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_GET_CONFIG_QNAME;
  15 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_RUNNING_QNAME;
  16
  17 import com.google.common.util.concurrent.FutureCallback;
  18 import com.google.common.util.concurrent.Futures;
  19 import com.google.common.util.concurrent.ListenableFuture;
  20 import com.google.common.util.concurrent.MoreExecutors;
  21 import com.google.common.util.concurrent.SettableFuture;
  22 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
  23 import java.util.Collection;
  24 import java.util.concurrent.ScheduledExecutorService;
  25 import java.util.concurrent.ScheduledFuture;
  26 import java.util.concurrent.TimeUnit;
  27 import org.checkerframework.checker.lock.qual.GuardedBy;
  28 import org.eclipse.jdt.annotation.NonNull;
  29 import org.opendaylight.mdsal.dom.api.DOMActionService;
  30 import org.opendaylight.mdsal.dom.api.DOMNotification;
  31 import org.opendaylight.mdsal.dom.api.DOMRpcAvailabilityListener;
  32 import org.opendaylight.mdsal.dom.api.DOMRpcResult;
  33 import org.opendaylight.mdsal.dom.api.DOMRpcService;
  34 import org.opendaylight.netconf.sal.connect.api.RemoteDeviceHandler;
  35 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfDeviceCommunicator;
  36 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfSessionPreferences;
  37 import org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil;
  38 import org.opendaylight.netconf.sal.connect.util.RemoteDeviceId;
  39 import org.opendaylight.yangtools.concepts.ListenerRegistration;
  40 import org.opendaylight.yangtools.rfc8528.data.api.MountPointContext;
  41 import org.opendaylight.yangtools.yang.common.QName;
  42 import org.opendaylight.yangtools.yang.data.api.schema.ContainerNode;
  43 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
  44 import org.slf4j.Logger;
  45 import org.slf4j.LoggerFactory;
  46
  47 /**
  48  * SalFacade proxy that invokes keepalive RPCs to prevent session shutdown from remote device
  49  * and to detect incorrect session drops (netconf session is inactive, but TCP/SSH connection is still present).
  50  * The keepalive RPC is a get-config with empty filter.
  51  */
  52 public final class KeepaliveSalFacade implements RemoteDeviceHandler<NetconfSessionPreferences> {
  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 KeepaliveTask keepaliveTask = new KeepaliveTask();
  62     private final RemoteDeviceHandler<NetconfSessionPreferences> salFacade;
  63     private final ScheduledExecutorService executor;
  64
  65     private final long keepaliveDelaySeconds;
  66     private final long timeoutNanos;
  67     private final long delayNanos;
  68
  69     private final RemoteDeviceId id;
  70
  71     private volatile NetconfDeviceCommunicator listener;
  72     private volatile DOMRpcService currentDeviceRpc;
  73
  74     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
  75                               final ScheduledExecutorService executor, final long keepaliveDelaySeconds,
  76                               final long requestTimeoutMillis) {
  77         this.id = id;
  78         this.salFacade = salFacade;
  79         this.executor = requireNonNull(executor);
  80         this.keepaliveDelaySeconds = keepaliveDelaySeconds;
  81         delayNanos = TimeUnit.SECONDS.toNanos(keepaliveDelaySeconds);
  82         timeoutNanos = TimeUnit.MILLISECONDS.toNanos(requestTimeoutMillis);
  83     }
  84
  85     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
  86                               final ScheduledExecutorService executor) {
  87         this(id, salFacade, executor, DEFAULT_DELAY, DEFAULT_TRANSACTION_TIMEOUT_MILLI);
  88     }
  89
  90     /**
  91      * Set the netconf session listener whenever ready.
  92      *
  93      * @param listener netconf session listener
  94      */
  95     public void setListener(final NetconfDeviceCommunicator listener) {
  96         this.listener = listener;
  97     }
  98
  99     /**
 100      * Cancel current keepalive and also reset current deviceRpc.
 101      */
 102     private synchronized void stopKeepalives() {
 103         keepaliveTask.disableKeepalive();
 104         currentDeviceRpc = null;
 105     }
 106
 107     void reconnect() {
 108         checkState(listener != null, "%s: Unable to reconnect, session listener is missing", id);
 109         stopKeepalives();
 110         LOG.info("{}: Reconnecting inactive netconf session", id);
 111         listener.disconnect();
 112     }
 113
 114     @Override
 115     public void onDeviceConnected(final MountPointContext remoteSchemaContext,
 116                           final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc) {
 117         onDeviceConnected(remoteSchemaContext, netconfSessionPreferences, deviceRpc, null);
 118     }
 119
 120     @Override
 121     public void onDeviceConnected(final MountPointContext remoteSchemaContext,
 122             final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc,
 123             final DOMActionService deviceAction) {
 124         this.currentDeviceRpc = requireNonNull(deviceRpc);
 125         salFacade.onDeviceConnected(remoteSchemaContext, netconfSessionPreferences,
 126             new KeepaliveDOMRpcService(deviceRpc), deviceAction);
 127
 128         LOG.debug("{}: Netconf session initiated, starting keepalives", id);
 129         LOG.trace("{}: Scheduling keepalives every {}s", id, keepaliveDelaySeconds);
 130         keepaliveTask.enableKeepalive();
 131     }
 132
 133     @Override
 134     public void onDeviceDisconnected() {
 135         stopKeepalives();
 136         salFacade.onDeviceDisconnected();
 137     }
 138
 139     @Override
 140     public void onDeviceFailed(final Throwable throwable) {
 141         stopKeepalives();
 142         salFacade.onDeviceFailed(throwable);
 143     }
 144
 145     @Override
 146     public void onNotification(final DOMNotification domNotification) {
 147         keepaliveTask.recordActivity();
 148         salFacade.onNotification(domNotification);
 149     }
 150
 151     @Override
 152     public void close() {
 153         stopKeepalives();
 154         salFacade.close();
 155     }
 156
 157     // Keepalive RPC static resources
 158     private static final @NonNull ContainerNode KEEPALIVE_PAYLOAD =
 159         NetconfMessageTransformUtil.wrap(NETCONF_GET_CONFIG_NODEID,
 160             getSourceNode(NETCONF_RUNNING_QNAME), NetconfMessageTransformUtil.EMPTY_FILTER);
 161
 162     /**
 163      * Invoke keepalive RPC and check the response. In case of any received response the keepalive
 164      * is considered successful and schedules next keepalive with a fixed delay. If the response is unsuccessful (no
 165      * response received, or the rcp could not even be sent) immediate reconnect is triggered as netconf session
 166      * is considered inactive/failed.
 167      */
 168     private final class KeepaliveTask implements Runnable, FutureCallback<DOMRpcResult> {
 169         private volatile long lastActivity;
 170         @GuardedBy("this")
 171         private boolean suppressed;
 172
 173         KeepaliveTask() {
 174             suppressed = false;
 175         }
 176
 177         @Override
 178         public void run() {
 179             final long local = lastActivity;
 180             final long now = System.nanoTime();
 181             final long inFutureNanos = local + delayNanos - now;
 182             if (inFutureNanos > 0) {
 183                 reschedule(inFutureNanos);
 184             } else {
 185                 sendKeepalive(now);
 186             }
 187         }
 188
 189         void recordActivity() {
 190             lastActivity = System.nanoTime();
 191         }
 192
 193         synchronized void disableKeepalive() {
 194             // unsuppressed -> suppressed
 195             suppressed = true;
 196         }
 197
 198         synchronized void enableKeepalive() {
 199             recordActivity();
 200             if (!suppressed) {
 201                 // unscheduled -> unsuppressed
 202                 reschedule();
 203             } else {
 204                 // suppressed -> unsuppressed
 205                 suppressed = false;
 206             }
 207         }
 208
 209         private synchronized void sendKeepalive(final long now) {
 210             if (suppressed) {
 211                 // suppressed -> unscheduled
 212                 suppressed = false;
 213                 return;
 214             }
 215
 216             final DOMRpcService deviceRpc = currentDeviceRpc;
 217             if (deviceRpc == null) {
 218                 // deviceRpc is null, which means we hit the reconnect window and attempted to send keepalive while
 219                 // we were reconnecting. Next keepalive will be scheduled after reconnect so no action necessary here.
 220                 LOG.debug("{}: Skipping keepalive while reconnecting", id);
 221                 return;
 222             }
 223
 224             LOG.trace("{}: Invoking keepalive RPC", id);
 225             final ListenableFuture<? extends DOMRpcResult> deviceFuture =
 226                 currentDeviceRpc.invokeRpc(NETCONF_GET_CONFIG_QNAME, KEEPALIVE_PAYLOAD);
 227
 228             lastActivity = now;
 229             Futures.addCallback(deviceFuture, this, MoreExecutors.directExecutor());
 230         }
 231
 232         @SuppressFBWarnings(value = "RCN_REDUNDANT_NULLCHECK_OF_NONNULL_VALUE",
 233                 justification = "Unrecognised NullableDecl")
 234         @Override
 235         public void onSuccess(final DOMRpcResult result) {
 236             // No matter what response we got, rpc-reply or rpc-error,
 237             // we got it from device so the netconf session is OK
 238             if (result == null) {
 239                 LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 240                 reconnect();
 241                 return;
 242             }
 243
 244             if (result.getResult() != null) {
 245                 reschedule();
 246             } else {
 247                 final Collection<?> errors = result.getErrors();
 248                 if (!errors.isEmpty()) {
 249                     LOG.warn("{}: Keepalive RPC failed with error: {}", id, errors);
 250                     reschedule();
 251                 } else {
 252                     LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 253                     reconnect();
 254                 }
 255             }
 256         }
 257
 258         @Override
 259         public void onFailure(final Throwable throwable) {
 260             LOG.warn("{}: Keepalive RPC failed. Reconnecting netconf session.", id, throwable);
 261             reconnect();
 262         }
 263
 264         private void reschedule() {
 265             reschedule(delayNanos);
 266         }
 267
 268         private void reschedule(final long delay) {
 269             executor.schedule(this, delay, TimeUnit.NANOSECONDS);
 270         }
 271     }
 272
 273     /*
 274      * Request timeout task is called once the requestTimeoutMillis is reached. At that moment, if the request is not
 275      * yet finished, we cancel it.
 276      */
 277     private final class RequestTimeoutTask implements FutureCallback<DOMRpcResult>, Runnable {
 278         private final @NonNull SettableFuture<DOMRpcResult> userFuture = SettableFuture.create();
 279         private final @NonNull ListenableFuture<? extends DOMRpcResult> deviceFuture;
 280
 281         RequestTimeoutTask(final ListenableFuture<? extends DOMRpcResult> rpcResultFuture) {
 282             this.deviceFuture = requireNonNull(rpcResultFuture);
 283             Futures.addCallback(deviceFuture, this, MoreExecutors.directExecutor());
 284         }
 285
 286         @Override
 287         public void run() {
 288             deviceFuture.cancel(true);
 289             userFuture.cancel(false);
 290             keepaliveTask.enableKeepalive();
 291         }
 292
 293         @Override
 294         public void onSuccess(final DOMRpcResult result) {
 295             // No matter what response we got,
 296             // rpc-reply or rpc-error, we got it from device so the netconf session is OK.
 297             userFuture.set(result);
 298             keepaliveTask.enableKeepalive();
 299         }
 300
 301         @Override
 302         public void onFailure(final Throwable throwable) {
 303             // User/Application RPC failed (The RPC did not reach the remote device or ...)
 304             // FIXME: what other reasons could cause this ?)
 305             LOG.warn("{}: Rpc failure detected. Reconnecting netconf session", id, throwable);
 306             userFuture.setException(throwable);
 307             // There is no point in keeping this session. Reconnect.
 308             reconnect();
 309         }
 310     }
 311
 312     /**
 313      * DOMRpcService proxy that attaches reset-keepalive-task and schedule
 314      * request-timeout-task to each RPC invocation.
 315      */
 316     public final class KeepaliveDOMRpcService implements DOMRpcService {
 317         private final @NonNull DOMRpcService deviceRpc;
 318
 319         KeepaliveDOMRpcService(final DOMRpcService deviceRpc) {
 320             this.deviceRpc = requireNonNull(deviceRpc);
 321         }
 322
 323         public @NonNull DOMRpcService getDeviceRpc() {
 324             return deviceRpc;
 325         }
 326
 327         @Override
 328         public ListenableFuture<? extends DOMRpcResult> invokeRpc(final QName type, final NormalizedNode<?, ?> input) {
 329             keepaliveTask.disableKeepalive();
 330             final ListenableFuture<? extends DOMRpcResult> deviceFuture = deviceRpc.invokeRpc(type, input);
 331
 332             final RequestTimeoutTask timeout = new RequestTimeoutTask(deviceFuture);
 333             final ScheduledFuture<?> timeoutFuture = executor.schedule(timeout, timeoutNanos, TimeUnit.NANOSECONDS);
 334             deviceFuture.addListener(() -> timeoutFuture.cancel(false), MoreExecutors.directExecutor());
 335
 336             return timeout.userFuture;
 337         }
 338
 339         @Override
 340         public <T extends DOMRpcAvailabilityListener> ListenerRegistration<T> registerRpcListener(final T rpcListener) {
 341             // There is no real communication with the device (yet), hence recordActivity() or anything
 342             return deviceRpc.registerRpcListener(rpcListener);
 343         }
 344     }
 345 }