plugins/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.concurrent.ScheduledExecutorService;
  24 import java.util.concurrent.TimeUnit;
  25 import javax.xml.transform.dom.DOMSource;
  26 import org.checkerframework.checker.lock.qual.GuardedBy;
  27 import org.eclipse.jdt.annotation.NonNull;
  28 import org.opendaylight.mdsal.dom.api.DOMNotification;
  29 import org.opendaylight.mdsal.dom.api.DOMRpcAvailabilityListener;
  30 import org.opendaylight.mdsal.dom.api.DOMRpcResult;
  31 import org.opendaylight.netconf.sal.connect.api.RemoteDeviceHandler;
  32 import org.opendaylight.netconf.sal.connect.api.RemoteDeviceId;
  33 import org.opendaylight.netconf.sal.connect.api.RemoteDeviceServices;
  34 import org.opendaylight.netconf.sal.connect.api.RemoteDeviceServices.Rpcs;
  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.yangtools.concepts.ListenerRegistration;
  40 import org.opendaylight.yangtools.yang.common.QName;
  41 import org.opendaylight.yangtools.yang.data.api.schema.ContainerNode;
  42 import org.slf4j.Logger;
  43 import org.slf4j.LoggerFactory;
  44
  45 /**
  46  * SalFacade proxy that invokes keepalive RPCs to prevent session shutdown from remote device
  47  * and to detect incorrect session drops (netconf session is inactive, but TCP/SSH connection is still present).
  48  * The keepalive RPC is a get-config with empty filter.
  49  */
  50 public final class KeepaliveSalFacade implements RemoteDeviceHandler {
  51     private static final Logger LOG = LoggerFactory.getLogger(KeepaliveSalFacade.class);
  52
  53     // 2 minutes keepalive delay by default
  54     private static final long DEFAULT_DELAY = TimeUnit.MINUTES.toSeconds(2);
  55
  56     // 1 minute transaction timeout by default
  57     private static final long DEFAULT_TRANSACTION_TIMEOUT_MILLI = TimeUnit.MILLISECONDS.toMillis(60000);
  58
  59     private final RemoteDeviceHandler salFacade;
  60     private final ScheduledExecutorService executor;
  61
  62     private final long keepaliveDelaySeconds;
  63     private final long timeoutNanos;
  64     private final long delayNanos;
  65
  66     private final RemoteDeviceId id;
  67
  68     private volatile NetconfDeviceCommunicator listener;
  69     private volatile KeepaliveTask task;
  70
  71     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler salFacade,
  72             final ScheduledExecutorService executor, final long keepaliveDelaySeconds,
  73             final long requestTimeoutMillis) {
  74         this.id = id;
  75         this.salFacade = salFacade;
  76         this.executor = requireNonNull(executor);
  77         this.keepaliveDelaySeconds = keepaliveDelaySeconds;
  78         delayNanos = TimeUnit.SECONDS.toNanos(keepaliveDelaySeconds);
  79         timeoutNanos = TimeUnit.MILLISECONDS.toNanos(requestTimeoutMillis);
  80     }
  81
  82     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler salFacade,
  83             final ScheduledExecutorService executor) {
  84         this(id, salFacade, executor, DEFAULT_DELAY, DEFAULT_TRANSACTION_TIMEOUT_MILLI);
  85     }
  86
  87     /**
  88      * Set the netconf session listener whenever ready.
  89      *
  90      * @param listener netconf session listener
  91      */
  92     public void setListener(final NetconfDeviceCommunicator listener) {
  93         this.listener = listener;
  94     }
  95
  96     /**
  97      * Cancel current keepalive and free it.
  98      */
  99     private synchronized void stopKeepalives() {
 100         final var localTask = task;
 101         if (localTask != null) {
 102             localTask.disableKeepalive();
 103             task = null;
 104         }
 105     }
 106
 107     private void disableKeepalive() {
 108         final var localTask = task;
 109         if (localTask != null) {
 110             localTask.disableKeepalive();
 111         }
 112     }
 113
 114     private void enableKeepalive() {
 115         final var localTask = task;
 116         if (localTask != null) {
 117             localTask.enableKeepalive();
 118         }
 119     }
 120
 121     void reconnect() {
 122         checkState(listener != null, "%s: Unable to reconnect, session listener is missing", id);
 123         stopKeepalives();
 124         LOG.info("{}: Reconnecting inactive netconf session", id);
 125         listener.disconnect();
 126     }
 127
 128     @Override
 129     public void onDeviceConnected(final NetconfDeviceSchema deviceSchema,
 130             final NetconfSessionPreferences sessionPreferences, final RemoteDeviceServices services) {
 131         final var devRpc = services.rpcs();
 132         task = new KeepaliveTask(devRpc);
 133
 134         final Rpcs keepaliveRpcs;
 135         if (devRpc instanceof Rpcs.Normalized normalized) {
 136             keepaliveRpcs = new NormalizedKeepaliveRpcs(normalized);
 137         } else if (devRpc instanceof Rpcs.Schemaless schemaless) {
 138             keepaliveRpcs = new SchemalessKeepaliveRpcs(schemaless);
 139         } else {
 140             throw new IllegalStateException("Unhandled " + devRpc);
 141         }
 142
 143         salFacade.onDeviceConnected(deviceSchema, sessionPreferences, new RemoteDeviceServices(keepaliveRpcs,
 144             // FIXME: wrap with keepalive
 145             services.actions()));
 146
 147         // We have performed a callback, which might have termined keepalives
 148         final var localTask = task;
 149         if (localTask != null) {
 150             LOG.debug("{}: Netconf session initiated, starting keepalives", id);
 151             LOG.trace("{}: Scheduling keepalives every {}s", id, keepaliveDelaySeconds);
 152             localTask.enableKeepalive();
 153         }
 154     }
 155
 156     @Override
 157     public void onDeviceDisconnected() {
 158         stopKeepalives();
 159         salFacade.onDeviceDisconnected();
 160     }
 161
 162     @Override
 163     public void onDeviceFailed(final Throwable throwable) {
 164         stopKeepalives();
 165         salFacade.onDeviceFailed(throwable);
 166     }
 167
 168     @Override
 169     public void onNotification(final DOMNotification domNotification) {
 170         final var localTask = task;
 171         if (localTask != null) {
 172             localTask.recordActivity();
 173         }
 174         salFacade.onNotification(domNotification);
 175     }
 176
 177     @Override
 178     public void close() {
 179         stopKeepalives();
 180         salFacade.close();
 181     }
 182
 183     private <T> @NonNull ListenableFuture<T> scheduleTimeout(final ListenableFuture<T> invokeFuture) {
 184         final var timeout = new RequestTimeoutTask<>(invokeFuture);
 185         final var timeoutFuture = executor.schedule(timeout, timeoutNanos, TimeUnit.NANOSECONDS);
 186         invokeFuture.addListener(() -> timeoutFuture.cancel(false), MoreExecutors.directExecutor());
 187         return timeout.userFuture;
 188     }
 189
 190     /**
 191      * Invoke keepalive RPC and check the response. In case of any received response the keepalive
 192      * is considered successful and schedules next keepalive with a fixed delay. If the response is unsuccessful (no
 193      * response received, or the rcp could not even be sent) immediate reconnect is triggered as netconf session
 194      * is considered inactive/failed.
 195      */
 196     private final class KeepaliveTask implements Runnable, FutureCallback<DOMRpcResult> {
 197         // Keepalive RPC static resources
 198         static final @NonNull ContainerNode KEEPALIVE_PAYLOAD = NetconfMessageTransformUtil.wrap(
 199             NETCONF_GET_CONFIG_NODEID, getSourceNode(NETCONF_RUNNING_NODEID), NetconfMessageTransformUtil.EMPTY_FILTER);
 200
 201         private final Rpcs devRpc;
 202
 203         @GuardedBy("this")
 204         private boolean suppressed = false;
 205
 206         private volatile long lastActivity;
 207
 208         KeepaliveTask(final Rpcs devRpc) {
 209             this.devRpc = requireNonNull(devRpc);
 210         }
 211
 212         @Override
 213         public void run() {
 214             final long local = lastActivity;
 215             final long now = System.nanoTime();
 216             final long inFutureNanos = local + delayNanos - now;
 217             if (inFutureNanos > 0) {
 218                 reschedule(inFutureNanos);
 219             } else {
 220                 sendKeepalive(now);
 221             }
 222         }
 223
 224         void recordActivity() {
 225             lastActivity = System.nanoTime();
 226         }
 227
 228         synchronized void disableKeepalive() {
 229             // unsuppressed -> suppressed
 230             suppressed = true;
 231         }
 232
 233         synchronized void enableKeepalive() {
 234             recordActivity();
 235             if (!suppressed) {
 236                 // unscheduled -> unsuppressed
 237                 reschedule();
 238             } else {
 239                 // suppressed -> unsuppressed
 240                 suppressed = false;
 241             }
 242         }
 243
 244         private synchronized void sendKeepalive(final long now) {
 245             if (suppressed) {
 246                 LOG.debug("{}: Skipping keepalive while disabled", id);
 247                 // suppressed -> unscheduled
 248                 suppressed = false;
 249                 return;
 250             }
 251
 252             LOG.trace("{}: Invoking keepalive RPC", id);
 253             final var deviceFuture = devRpc.invokeNetconf(NETCONF_GET_CONFIG_QNAME, KEEPALIVE_PAYLOAD);
 254             lastActivity = now;
 255             Futures.addCallback(deviceFuture, this, MoreExecutors.directExecutor());
 256         }
 257
 258         // FIXME: re-examine this suppression
 259         @SuppressFBWarnings(value = "RCN_REDUNDANT_NULLCHECK_OF_NONNULL_VALUE",
 260                 justification = "Unrecognised NullableDecl")
 261         @Override
 262         public void onSuccess(final DOMRpcResult result) {
 263             // No matter what response we got, rpc-reply or rpc-error,
 264             // we got it from device so the netconf session is OK
 265             if (result == null) {
 266                 LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 267                 reconnect();
 268                 return;
 269             }
 270
 271             if (result.value() != null) {
 272                 reschedule();
 273             } else {
 274                 final var errors = result.errors();
 275                 if (!errors.isEmpty()) {
 276                     LOG.warn("{}: Keepalive RPC failed with error: {}", id, errors);
 277                     reschedule();
 278                 } else {
 279                     LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 280                     reconnect();
 281                 }
 282             }
 283         }
 284
 285         @Override
 286         public void onFailure(final Throwable throwable) {
 287             LOG.warn("{}: Keepalive RPC failed. Reconnecting netconf session.", id, throwable);
 288             reconnect();
 289         }
 290
 291         private void reschedule() {
 292             reschedule(delayNanos);
 293         }
 294
 295         private void reschedule(final long delay) {
 296             executor.schedule(this, delay, TimeUnit.NANOSECONDS);
 297         }
 298     }
 299
 300     /*
 301      * Request timeout task is called once the requestTimeoutMillis is reached. At that moment, if the request is not
 302      * yet finished, we cancel it.
 303      */
 304     private final class RequestTimeoutTask<V> implements FutureCallback<V>, Runnable {
 305         private final @NonNull SettableFuture<V> userFuture = SettableFuture.create();
 306         private final @NonNull ListenableFuture<? extends V> rpcResultFuture;
 307
 308         RequestTimeoutTask(final ListenableFuture<V> rpcResultFuture) {
 309             this.rpcResultFuture = requireNonNull(rpcResultFuture);
 310             Futures.addCallback(rpcResultFuture, this, MoreExecutors.directExecutor());
 311         }
 312
 313         @Override
 314         public void run() {
 315             rpcResultFuture.cancel(true);
 316             userFuture.cancel(false);
 317             enableKeepalive();
 318         }
 319
 320         @Override
 321         public void onSuccess(final V result) {
 322             // No matter what response we got,
 323             // rpc-reply or rpc-error, we got it from device so the netconf session is OK.
 324             userFuture.set(result);
 325             enableKeepalive();
 326         }
 327
 328         @Override
 329         public void onFailure(final Throwable throwable) {
 330             // User/Application RPC failed (The RPC did not reach the remote device or ...)
 331             // FIXME: what other reasons could cause this ?)
 332             LOG.warn("{}: Rpc failure detected. Reconnecting netconf session", id, throwable);
 333             userFuture.setException(throwable);
 334             // There is no point in keeping this session. Reconnect.
 335             reconnect();
 336         }
 337     }
 338
 339     /**
 340      * Proxy for {@link Rpcs} which attaches a reset-keepalive-task and schedule request-timeout-task to each RPC
 341      * invocation. Version for {@link Rpcs.Normalized}.
 342      */
 343     private final class NormalizedKeepaliveRpcs implements Rpcs.Normalized {
 344         private final Rpcs.Normalized delegate;
 345
 346         NormalizedKeepaliveRpcs(final Rpcs.Normalized delegate) {
 347             this.delegate = requireNonNull(delegate);
 348         }
 349
 350         @Override
 351         public ListenableFuture<? extends DOMRpcResult> invokeRpc(final QName type, final ContainerNode input) {
 352             // FIXME: what happens if we disable keepalive and then invokeRpc() throws?
 353             disableKeepalive();
 354             return scheduleTimeout(delegate.invokeRpc(type, input));
 355         }
 356
 357         @Override
 358         public <T extends DOMRpcAvailabilityListener> ListenerRegistration<T> registerRpcListener(
 359             final T rpcListener) {
 360             // There is no real communication with the device (yet), hence no recordActivity() or anything
 361             return delegate.registerRpcListener(rpcListener);
 362         }
 363     }
 364
 365     /**
 366      * Proxy for {@link Rpcs} which attaches a reset-keepalive-task and schedule request-timeout-task to each RPC
 367      * invocation. Version for {@link Rpcs.Schemaless}.
 368      */
 369     private final class SchemalessKeepaliveRpcs implements Rpcs.Schemaless {
 370         private final Rpcs.Schemaless delegate;
 371
 372         SchemalessKeepaliveRpcs(final Rpcs.Schemaless delegate) {
 373             this.delegate = requireNonNull(delegate);
 374         }
 375
 376         @Override
 377         public ListenableFuture<? extends DOMRpcResult> invokeNetconf(final QName type, final ContainerNode input) {
 378             // FIXME: what happens if we disable keepalive and then invokeRpc() throws?
 379             disableKeepalive();
 380             return scheduleTimeout(delegate.invokeNetconf(type, input));
 381         }
 382
 383         @Override
 384         public ListenableFuture<? extends DOMSource> invokeRpc(final QName type, final DOMSource input) {
 385             // FIXME: what happens if we disable keepalive and then invokeRpc() throws?
 386             disableKeepalive();
 387             return scheduleTimeout(delegate.invokeRpc(type, input));
 388         }
 389     }
 390 }