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 org.opendaylight.netconf.sal.connect.netconf.util.NetconfBaseOps.getSourceNode;
  11 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_GET_CONFIG_QNAME;
  12 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_RUNNING_QNAME;
  13 import static org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil.toPath;
  14
  15 import com.google.common.base.Preconditions;
  16 import com.google.common.util.concurrent.CheckedFuture;
  17 import com.google.common.util.concurrent.FutureCallback;
  18 import com.google.common.util.concurrent.Futures;
  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 javax.annotation.Nonnull;
  25 import javax.annotation.Nullable;
  26 import org.opendaylight.controller.md.sal.dom.api.DOMNotification;
  27 import org.opendaylight.controller.md.sal.dom.api.DOMRpcAvailabilityListener;
  28 import org.opendaylight.controller.md.sal.dom.api.DOMRpcException;
  29 import org.opendaylight.controller.md.sal.dom.api.DOMRpcResult;
  30 import org.opendaylight.controller.md.sal.dom.api.DOMRpcService;
  31 import org.opendaylight.netconf.sal.connect.api.RemoteDeviceHandler;
  32 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfDeviceCommunicator;
  33 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfSessionPreferences;
  34 import org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil;
  35 import org.opendaylight.netconf.sal.connect.util.RemoteDeviceId;
  36 import org.opendaylight.yangtools.concepts.ListenerRegistration;
  37 import org.opendaylight.yangtools.yang.data.api.schema.ContainerNode;
  38 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
  39 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  40 import org.opendaylight.yangtools.yang.model.api.SchemaPath;
  41 import org.slf4j.Logger;
  42 import org.slf4j.LoggerFactory;
  43
  44 /**
  45  * SalFacade proxy that invokes keepalive RPCs to prevent session shutdown from remote device
  46  * and to detect incorrect session drops (netconf session is inactive, but TCP/SSH connection is still present).
  47  * The keepalive RPC is a get-config with empty filter.
  48  */
  49 public final class KeepaliveSalFacade implements RemoteDeviceHandler<NetconfSessionPreferences> {
  50
  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 RemoteDeviceId id;
  60     private final RemoteDeviceHandler<NetconfSessionPreferences> salFacade;
  61     private final ScheduledExecutorService executor;
  62     private final long keepaliveDelaySeconds;
  63     private final ResetKeepalive resetKeepaliveTask;
  64     private final long defaultRequestTimeoutMillis;
  65
  66     private volatile NetconfDeviceCommunicator listener;
  67     private volatile ScheduledFuture<?> currentKeepalive;
  68     private volatile DOMRpcService currentDeviceRpc;
  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.
  97      * If its already started, let it finish ... not such a big deal.
  98      *
  99      * <p>
 100      * Then schedule next keepalive.
 101      */
 102     void resetKeepalive() {
 103         LOG.trace("{}: Resetting netconf keepalive timer", id);
 104         if (currentKeepalive != null) {
 105             currentKeepalive.cancel(false);
 106         }
 107         scheduleKeepalive();
 108     }
 109
 110     /**
 111      * Cancel current keepalive and also reset current deviceRpc.
 112      */
 113     private void stopKeepalives() {
 114         if (currentKeepalive != null) {
 115             currentKeepalive.cancel(false);
 116         }
 117         currentDeviceRpc = null;
 118     }
 119
 120     void reconnect() {
 121         Preconditions.checkState(listener != null, "%s: Unable to reconnect, session listener is missing", id);
 122         stopKeepalives();
 123         LOG.info("{}: Reconnecting inactive netconf session", id);
 124         listener.disconnect();
 125     }
 126
 127     @Override
 128     public void onDeviceConnected(final SchemaContext remoteSchemaContext,
 129                           final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc) {
 130         this.currentDeviceRpc = deviceRpc;
 131         final DOMRpcService deviceRpc1 =
 132                 new KeepaliveDOMRpcService(deviceRpc, resetKeepaliveTask, defaultRequestTimeoutMillis, executor);
 133         salFacade.onDeviceConnected(remoteSchemaContext, netconfSessionPreferences, deviceRpc1);
 134
 135         LOG.debug("{}: Netconf session initiated, starting keepalives", id);
 136         scheduleKeepalive();
 137     }
 138
 139     private void scheduleKeepalive() {
 140         Preconditions.checkState(currentDeviceRpc != null);
 141         LOG.trace("{}: Scheduling next keepalive in {} {}", id, keepaliveDelaySeconds, TimeUnit.SECONDS);
 142         currentKeepalive = executor.schedule(new Keepalive(currentKeepalive), keepaliveDelaySeconds, TimeUnit.SECONDS);
 143     }
 144
 145     @Override
 146     public void onDeviceDisconnected() {
 147         stopKeepalives();
 148         salFacade.onDeviceDisconnected();
 149     }
 150
 151     @Override
 152     public void onDeviceFailed(final Throwable throwable) {
 153         stopKeepalives();
 154         salFacade.onDeviceFailed(throwable);
 155     }
 156
 157     @Override
 158     public void onNotification(final DOMNotification domNotification) {
 159         resetKeepalive();
 160         salFacade.onNotification(domNotification);
 161     }
 162
 163     @Override
 164     public void close() {
 165         stopKeepalives();
 166         salFacade.close();
 167     }
 168
 169     // Keepalive RPC static resources
 170     private static final SchemaPath PATH = toPath(NETCONF_GET_CONFIG_QNAME);
 171     private static final ContainerNode KEEPALIVE_PAYLOAD = NetconfMessageTransformUtil.wrap(NETCONF_GET_CONFIG_QNAME,
 172             getSourceNode(NETCONF_RUNNING_QNAME), NetconfMessageTransformUtil.EMPTY_FILTER);
 173
 174     /**
 175      * Invoke keepalive RPC and check the response. In case of any received response the keepalive
 176      * is considered successful and schedules next keepalive with a fixed delay. If the response is unsuccessful (no
 177      * response received, or the rcp could not even be sent) immediate reconnect is triggered as netconf session
 178      * is considered inactive/failed.
 179      */
 180     private class Keepalive implements Runnable, FutureCallback<DOMRpcResult> {
 181
 182         private final ScheduledFuture<?> previousKeepalive;
 183
 184         Keepalive(final ScheduledFuture<?> previousKeepalive) {
 185             this.previousKeepalive = previousKeepalive;
 186         }
 187
 188         @Override
 189         public void run() {
 190             LOG.trace("{}: Invoking keepalive RPC", id);
 191
 192             try {
 193                 if (previousKeepalive != null && !previousKeepalive.isDone()) {
 194                     onFailure(new IllegalStateException("Previous keepalive timed out"));
 195                 } else {
 196                     Futures.addCallback(currentDeviceRpc.invokeRpc(PATH, KEEPALIVE_PAYLOAD), this,
 197                                         MoreExecutors.directExecutor());
 198                 }
 199             } catch (NullPointerException e) {
 200                 LOG.debug("{}: Skipping keepalive while reconnecting", id);
 201                 // Empty catch block intentional
 202                 // Do nothing. The currentDeviceRpc was null and it means we hit the reconnect window and
 203                 // attempted to send keepalive while we were reconnecting. Next keepalive will be scheduled
 204                 // after reconnect so no action necessary here.
 205             }
 206         }
 207
 208         @SuppressFBWarnings(value = "RCN_REDUNDANT_NULLCHECK_OF_NONNULL_VALUE",
 209                 justification = "Unrecognised NullableDecl")
 210         @Override
 211         public void onSuccess(final DOMRpcResult result) {
 212             // No matter what response we got, rpc-reply or rpc-error,
 213             // we got it from device so the netconf session is OK
 214             if (result != null && result.getResult() != null) {
 215                 LOG.debug("{}: Keepalive RPC successful with response: {}", id, result.getResult());
 216                 scheduleKeepalive();
 217             } else if (result != null && !result.getErrors().isEmpty()) {
 218                 LOG.warn("{}: Keepalive RPC failed with error: {}", id, result.getErrors());
 219                 scheduleKeepalive();
 220             } else {
 221                 LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 222                 reconnect();
 223             }
 224         }
 225
 226         @Override
 227         public void onFailure(@Nonnull final Throwable throwable) {
 228             LOG.warn("{}: Keepalive RPC failed. Reconnecting netconf session.", id, throwable);
 229             reconnect();
 230         }
 231     }
 232
 233     /**
 234      * Reset keepalive after each RPC response received.
 235      */
 236     private class ResetKeepalive implements FutureCallback<DOMRpcResult> {
 237         @Override
 238         public void onSuccess(@Nullable final DOMRpcResult result) {
 239             // No matter what response we got,
 240             // rpc-reply or rpc-error, we got it from device so the netconf session is OK.
 241             resetKeepalive();
 242         }
 243
 244         @Override
 245         public void onFailure(@Nonnull final Throwable throwable) {
 246             // User/Application RPC failed (The RPC did not reach the remote device or ..
 247             // TODO what other reasons could cause this ?)
 248             // There is no point in keeping this session. Reconnect.
 249             LOG.warn("{}: Rpc failure detected. Reconnecting netconf session", id, throwable);
 250             reconnect();
 251         }
 252     }
 253
 254     /*
 255      * Request timeout task is called once the defaultRequestTimeoutMillis is
 256      * reached. At this moment, if the request is not yet finished, we cancel
 257      * it.
 258      */
 259     private static final class RequestTimeoutTask implements Runnable {
 260
 261         private final CheckedFuture<DOMRpcResult, DOMRpcException> rpcResultFuture;
 262
 263         RequestTimeoutTask(final CheckedFuture<DOMRpcResult, DOMRpcException> rpcResultFuture) {
 264             this.rpcResultFuture = rpcResultFuture;
 265         }
 266
 267         @Override
 268         public void run() {
 269             if (!rpcResultFuture.isDone()) {
 270                 rpcResultFuture.cancel(true);
 271             }
 272         }
 273     }
 274
 275     /**
 276      * DOMRpcService proxy that attaches reset-keepalive-task and schedule
 277      * request-timeout-task to each RPC invocation.
 278      */
 279     public static final class KeepaliveDOMRpcService implements DOMRpcService {
 280
 281         private final DOMRpcService deviceRpc;
 282         private final ResetKeepalive resetKeepaliveTask;
 283         private final long defaultRequestTimeoutMillis;
 284         private final ScheduledExecutorService executor;
 285
 286         KeepaliveDOMRpcService(final DOMRpcService deviceRpc, final ResetKeepalive resetKeepaliveTask,
 287                 final long defaultRequestTimeoutMillis, final ScheduledExecutorService executor) {
 288             this.deviceRpc = deviceRpc;
 289             this.resetKeepaliveTask = resetKeepaliveTask;
 290             this.defaultRequestTimeoutMillis = defaultRequestTimeoutMillis;
 291             this.executor = executor;
 292         }
 293
 294         public DOMRpcService getDeviceRpc() {
 295             return deviceRpc;
 296         }
 297
 298         @Nonnull
 299         @Override
 300         public CheckedFuture<DOMRpcResult, DOMRpcException> invokeRpc(@Nonnull final SchemaPath type,
 301                                                                       final NormalizedNode<?, ?> input) {
 302             final CheckedFuture<DOMRpcResult, DOMRpcException> domRpcResultDOMRpcExceptionCheckedFuture =
 303                     deviceRpc.invokeRpc(type, input);
 304             Futures.addCallback(domRpcResultDOMRpcExceptionCheckedFuture, resetKeepaliveTask,
 305                                 MoreExecutors.directExecutor());
 306
 307             final RequestTimeoutTask timeoutTask = new RequestTimeoutTask(domRpcResultDOMRpcExceptionCheckedFuture);
 308             executor.schedule(timeoutTask, defaultRequestTimeoutMillis, TimeUnit.MILLISECONDS);
 309
 310             return domRpcResultDOMRpcExceptionCheckedFuture;
 311         }
 312
 313         @Override
 314         public <T extends DOMRpcAvailabilityListener> ListenerRegistration<T> registerRpcListener(
 315                 @Nonnull final T listener) {
 316             // There is no real communication with the device (yet), no reset here
 317             return deviceRpc.registerRpcListener(listener);
 318         }
 319     }
 320 }