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_NODEID;
  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.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.api.RemoteDeviceServices;
  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 RemoteDeviceServices currentServices;
  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         currentServices = 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 sessionPreferences, final RemoteDeviceServices services) {
 117         currentServices = requireNonNull(services);
 118
 119         final var devAction = services.actions();
 120         // FIXME: wrap with keepalive
 121         final var kaAction = devAction;
 122
 123         salFacade.onDeviceConnected(deviceSchema, sessionPreferences,
 124             new RemoteDeviceServices(new KeepaliveDOMRpcService(services.rpcs()), kaAction));
 125
 126         LOG.debug("{}: Netconf session initiated, starting keepalives", id);
 127         LOG.trace("{}: Scheduling keepalives every {}s", id, keepaliveDelaySeconds);
 128         keepaliveTask.enableKeepalive();
 129     }
 130
 131     @Override
 132     public void onDeviceDisconnected() {
 133         stopKeepalives();
 134         salFacade.onDeviceDisconnected();
 135     }
 136
 137     @Override
 138     public void onDeviceFailed(final Throwable throwable) {
 139         stopKeepalives();
 140         salFacade.onDeviceFailed(throwable);
 141     }
 142
 143     @Override
 144     public void onNotification(final DOMNotification domNotification) {
 145         keepaliveTask.recordActivity();
 146         salFacade.onNotification(domNotification);
 147     }
 148
 149     @Override
 150     public void close() {
 151         stopKeepalives();
 152         salFacade.close();
 153     }
 154
 155     // Keepalive RPC static resources
 156     private static final @NonNull ContainerNode KEEPALIVE_PAYLOAD =
 157         NetconfMessageTransformUtil.wrap(NETCONF_GET_CONFIG_NODEID,
 158             getSourceNode(NETCONF_RUNNING_NODEID), NetconfMessageTransformUtil.EMPTY_FILTER);
 159
 160     /**
 161      * Invoke keepalive RPC and check the response. In case of any received response the keepalive
 162      * is considered successful and schedules next keepalive with a fixed delay. If the response is unsuccessful (no
 163      * response received, or the rcp could not even be sent) immediate reconnect is triggered as netconf session
 164      * is considered inactive/failed.
 165      */
 166     private final class KeepaliveTask implements Runnable, FutureCallback<DOMRpcResult> {
 167         private volatile long lastActivity;
 168         @GuardedBy("this")
 169         private boolean suppressed;
 170
 171         KeepaliveTask() {
 172             suppressed = false;
 173         }
 174
 175         @Override
 176         public void run() {
 177             final long local = lastActivity;
 178             final long now = System.nanoTime();
 179             final long inFutureNanos = local + delayNanos - now;
 180             if (inFutureNanos > 0) {
 181                 reschedule(inFutureNanos);
 182             } else {
 183                 sendKeepalive(now);
 184             }
 185         }
 186
 187         void recordActivity() {
 188             lastActivity = System.nanoTime();
 189         }
 190
 191         synchronized void disableKeepalive() {
 192             // unsuppressed -> suppressed
 193             suppressed = true;
 194         }
 195
 196         synchronized void enableKeepalive() {
 197             recordActivity();
 198             if (!suppressed) {
 199                 // unscheduled -> unsuppressed
 200                 reschedule();
 201             } else {
 202                 // suppressed -> unsuppressed
 203                 suppressed = false;
 204             }
 205         }
 206
 207         private synchronized void sendKeepalive(final long now) {
 208             if (suppressed) {
 209                 // suppressed -> unscheduled
 210                 suppressed = false;
 211                 return;
 212             }
 213
 214             final var localServices = currentServices;
 215             if (localServices == null) {
 216                 // deviceRpc is null, which means we hit the reconnect window and attempted to send keepalive while
 217                 // we were reconnecting. Next keepalive will be scheduled after reconnect so no action necessary here.
 218                 LOG.debug("{}: Skipping keepalive while reconnecting", id);
 219                 return;
 220             }
 221
 222             LOG.trace("{}: Invoking keepalive RPC", id);
 223             final var deviceFuture = localServices.rpcs().invokeRpc(NETCONF_GET_CONFIG_QNAME, KEEPALIVE_PAYLOAD);
 224
 225             lastActivity = now;
 226             Futures.addCallback(deviceFuture, this, MoreExecutors.directExecutor());
 227         }
 228
 229         @SuppressFBWarnings(value = "RCN_REDUNDANT_NULLCHECK_OF_NONNULL_VALUE",
 230                 justification = "Unrecognised NullableDecl")
 231         @Override
 232         public void onSuccess(final DOMRpcResult result) {
 233             // No matter what response we got, rpc-reply or rpc-error,
 234             // we got it from device so the netconf session is OK
 235             if (result == null) {
 236                 LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 237                 reconnect();
 238                 return;
 239             }
 240
 241             if (result.getResult() != null) {
 242                 reschedule();
 243             } else {
 244                 final Collection<?> errors = result.getErrors();
 245                 if (!errors.isEmpty()) {
 246                     LOG.warn("{}: Keepalive RPC failed with error: {}", id, errors);
 247                     reschedule();
 248                 } else {
 249                     LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 250                     reconnect();
 251                 }
 252             }
 253         }
 254
 255         @Override
 256         public void onFailure(final Throwable throwable) {
 257             LOG.warn("{}: Keepalive RPC failed. Reconnecting netconf session.", id, throwable);
 258             reconnect();
 259         }
 260
 261         private void reschedule() {
 262             reschedule(delayNanos);
 263         }
 264
 265         private void reschedule(final long delay) {
 266             executor.schedule(this, delay, TimeUnit.NANOSECONDS);
 267         }
 268     }
 269
 270     /*
 271      * Request timeout task is called once the requestTimeoutMillis is reached. At that moment, if the request is not
 272      * yet finished, we cancel it.
 273      */
 274     private final class RequestTimeoutTask implements FutureCallback<DOMRpcResult>, Runnable {
 275         private final @NonNull SettableFuture<DOMRpcResult> userFuture = SettableFuture.create();
 276         private final @NonNull ListenableFuture<? extends DOMRpcResult> deviceFuture;
 277
 278         RequestTimeoutTask(final ListenableFuture<? extends DOMRpcResult> rpcResultFuture) {
 279             deviceFuture = requireNonNull(rpcResultFuture);
 280             Futures.addCallback(deviceFuture, this, MoreExecutors.directExecutor());
 281         }
 282
 283         @Override
 284         public void run() {
 285             deviceFuture.cancel(true);
 286             userFuture.cancel(false);
 287             keepaliveTask.enableKeepalive();
 288         }
 289
 290         @Override
 291         public void onSuccess(final DOMRpcResult result) {
 292             // No matter what response we got,
 293             // rpc-reply or rpc-error, we got it from device so the netconf session is OK.
 294             userFuture.set(result);
 295             keepaliveTask.enableKeepalive();
 296         }
 297
 298         @Override
 299         public void onFailure(final Throwable throwable) {
 300             // User/Application RPC failed (The RPC did not reach the remote device or ...)
 301             // FIXME: what other reasons could cause this ?)
 302             LOG.warn("{}: Rpc failure detected. Reconnecting netconf session", id, throwable);
 303             userFuture.setException(throwable);
 304             // There is no point in keeping this session. Reconnect.
 305             reconnect();
 306         }
 307     }
 308
 309     /**
 310      * DOMRpcService proxy that attaches reset-keepalive-task and schedule
 311      * request-timeout-task to each RPC invocation.
 312      */
 313     public final class KeepaliveDOMRpcService implements DOMRpcService {
 314         private final @NonNull DOMRpcService deviceRpc;
 315
 316         KeepaliveDOMRpcService(final DOMRpcService deviceRpc) {
 317             this.deviceRpc = requireNonNull(deviceRpc);
 318         }
 319
 320         public @NonNull DOMRpcService getDeviceRpc() {
 321             return deviceRpc;
 322         }
 323
 324         @Override
 325         public ListenableFuture<? extends DOMRpcResult> invokeRpc(final QName type, final NormalizedNode input) {
 326             keepaliveTask.disableKeepalive();
 327             final ListenableFuture<? extends DOMRpcResult> deviceFuture = deviceRpc.invokeRpc(type, input);
 328
 329             final RequestTimeoutTask timeout = new RequestTimeoutTask(deviceFuture);
 330             final ScheduledFuture<?> timeoutFuture = executor.schedule(timeout, timeoutNanos, TimeUnit.NANOSECONDS);
 331             deviceFuture.addListener(() -> timeoutFuture.cancel(false), MoreExecutors.directExecutor());
 332
 333             return timeout.userFuture;
 334         }
 335
 336         @Override
 337         public <T extends DOMRpcAvailabilityListener> ListenerRegistration<T> registerRpcListener(final T rpcListener) {
 338             // There is no real communication with the device (yet), hence recordActivity() or anything
 339             return deviceRpc.registerRpcListener(rpcListener);
 340         }
 341     }
 342 }