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.NetconfDeviceSchema;
  36 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfDeviceCommunicator;
  37 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfSessionPreferences;
  38 import org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil;
  39 import org.opendaylight.netconf.sal.connect.util.RemoteDeviceId;
  40 import org.opendaylight.yangtools.concepts.ListenerRegistration;
  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 {
  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 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 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 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 NetconfDeviceSchema deviceSchema,
 116             final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc) {
 117         onDeviceConnected(deviceSchema, netconfSessionPreferences, deviceRpc, null);
 118     }
 119
 120     @Override
 121     public void onDeviceConnected(final NetconfDeviceSchema deviceSchema,
 122             final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc,
 123             final DOMActionService deviceAction) {
 124         currentDeviceRpc = requireNonNull(deviceRpc);
 125         salFacade.onDeviceConnected(deviceSchema, 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 var 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 var deviceFuture = deviceRpc.invokeRpc(NETCONF_GET_CONFIG_QNAME, KEEPALIVE_PAYLOAD);
 226
 227             lastActivity = now;
 228             Futures.addCallback(deviceFuture, this, MoreExecutors.directExecutor());
 229         }
 230
 231         @SuppressFBWarnings(value = "RCN_REDUNDANT_NULLCHECK_OF_NONNULL_VALUE",
 232                 justification = "Unrecognised NullableDecl")
 233         @Override
 234         public void onSuccess(final DOMRpcResult result) {
 235             // No matter what response we got, rpc-reply or rpc-error,
 236             // we got it from device so the netconf session is OK
 237             if (result == null) {
 238                 LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 239                 reconnect();
 240                 return;
 241             }
 242
 243             if (result.getResult() != null) {
 244                 reschedule();
 245             } else {
 246                 final Collection<?> errors = result.getErrors();
 247                 if (!errors.isEmpty()) {
 248                     LOG.warn("{}: Keepalive RPC failed with error: {}", id, errors);
 249                     reschedule();
 250                 } else {
 251                     LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 252                     reconnect();
 253                 }
 254             }
 255         }
 256
 257         @Override
 258         public void onFailure(final Throwable throwable) {
 259             LOG.warn("{}: Keepalive RPC failed. Reconnecting netconf session.", id, throwable);
 260             reconnect();
 261         }
 262
 263         private void reschedule() {
 264             reschedule(delayNanos);
 265         }
 266
 267         private void reschedule(final long delay) {
 268             executor.schedule(this, delay, TimeUnit.NANOSECONDS);
 269         }
 270     }
 271
 272     /*
 273      * Request timeout task is called once the requestTimeoutMillis is reached. At that moment, if the request is not
 274      * yet finished, we cancel it.
 275      */
 276     private final class RequestTimeoutTask implements FutureCallback<DOMRpcResult>, Runnable {
 277         private final @NonNull SettableFuture<DOMRpcResult> userFuture = SettableFuture.create();
 278         private final @NonNull ListenableFuture<? extends DOMRpcResult> deviceFuture;
 279
 280         RequestTimeoutTask(final ListenableFuture<? extends DOMRpcResult> rpcResultFuture) {
 281             deviceFuture = requireNonNull(rpcResultFuture);
 282             Futures.addCallback(deviceFuture, this, MoreExecutors.directExecutor());
 283         }
 284
 285         @Override
 286         public void run() {
 287             deviceFuture.cancel(true);
 288             userFuture.cancel(false);
 289             keepaliveTask.enableKeepalive();
 290         }
 291
 292         @Override
 293         public void onSuccess(final DOMRpcResult result) {
 294             // No matter what response we got,
 295             // rpc-reply or rpc-error, we got it from device so the netconf session is OK.
 296             userFuture.set(result);
 297             keepaliveTask.enableKeepalive();
 298         }
 299
 300         @Override
 301         public void onFailure(final Throwable throwable) {
 302             // User/Application RPC failed (The RPC did not reach the remote device or ...)
 303             // FIXME: what other reasons could cause this ?)
 304             LOG.warn("{}: Rpc failure detected. Reconnecting netconf session", id, throwable);
 305             userFuture.setException(throwable);
 306             // There is no point in keeping this session. Reconnect.
 307             reconnect();
 308         }
 309     }
 310
 311     /**
 312      * DOMRpcService proxy that attaches reset-keepalive-task and schedule
 313      * request-timeout-task to each RPC invocation.
 314      */
 315     public final class KeepaliveDOMRpcService implements DOMRpcService {
 316         private final @NonNull DOMRpcService deviceRpc;
 317
 318         KeepaliveDOMRpcService(final DOMRpcService deviceRpc) {
 319             this.deviceRpc = requireNonNull(deviceRpc);
 320         }
 321
 322         public @NonNull DOMRpcService getDeviceRpc() {
 323             return deviceRpc;
 324         }
 325
 326         @Override
 327         public ListenableFuture<? extends DOMRpcResult> invokeRpc(final QName type, final NormalizedNode input) {
 328             keepaliveTask.disableKeepalive();
 329             final ListenableFuture<? extends DOMRpcResult> deviceFuture = deviceRpc.invokeRpc(type, input);
 330
 331             final RequestTimeoutTask timeout = new RequestTimeoutTask(deviceFuture);
 332             final ScheduledFuture<?> timeoutFuture = executor.schedule(timeout, timeoutNanos, TimeUnit.NANOSECONDS);
 333             deviceFuture.addListener(() -> timeoutFuture.cancel(false), MoreExecutors.directExecutor());
 334
 335             return timeout.userFuture;
 336         }
 337
 338         @Override
 339         public <T extends DOMRpcAvailabilityListener> ListenerRegistration<T> registerRpcListener(final T rpcListener) {
 340             // There is no real communication with the device (yet), hence recordActivity() or anything
 341             return deviceRpc.registerRpcListener(rpcListener);
 342         }
 343     }
 344 }