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 java.util.concurrent.ScheduledExecutorService;
  20 import java.util.concurrent.ScheduledFuture;
  21 import java.util.concurrent.TimeUnit;
  22 import javax.annotation.Nonnull;
  23 import javax.annotation.Nullable;
  24 import org.opendaylight.controller.md.sal.dom.api.DOMNotification;
  25 import org.opendaylight.controller.md.sal.dom.api.DOMRpcAvailabilityListener;
  26 import org.opendaylight.controller.md.sal.dom.api.DOMRpcException;
  27 import org.opendaylight.controller.md.sal.dom.api.DOMRpcResult;
  28 import org.opendaylight.controller.md.sal.dom.api.DOMRpcService;
  29 import org.opendaylight.netconf.sal.connect.api.RemoteDeviceHandler;
  30 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfDeviceCommunicator;
  31 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfSessionPreferences;
  32 import org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil;
  33 import org.opendaylight.netconf.sal.connect.util.RemoteDeviceId;
  34 import org.opendaylight.yangtools.concepts.ListenerRegistration;
  35 import org.opendaylight.yangtools.yang.data.api.schema.ContainerNode;
  36 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
  37 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  38 import org.opendaylight.yangtools.yang.model.api.SchemaPath;
  39 import org.slf4j.Logger;
  40 import org.slf4j.LoggerFactory;
  41
  42 /**
  43  * SalFacade proxy that invokes keepalive RPCs to prevent session shutdown from remote device
  44  * and to detect incorrect session drops (netconf session is inactive, but TCP/SSH connection is still present).
  45  * The keepalive RPC is a get-config with empty filter.
  46  */
  47 public final class KeepaliveSalFacade implements RemoteDeviceHandler<NetconfSessionPreferences> {
  48
  49     private static final Logger LOG = LoggerFactory.getLogger(KeepaliveSalFacade.class);
  50
  51     // 2 minutes keepalive delay by default
  52     private static final long DEFAULT_DELAY = TimeUnit.MINUTES.toSeconds(2);
  53
  54     // 1 minute transaction timeout by default
  55     private static final long DEFAULT_TRANSACTION_TIMEOUT_MILLI = TimeUnit.MILLISECONDS.toMillis(60000);
  56
  57     private final RemoteDeviceId id;
  58     private final RemoteDeviceHandler<NetconfSessionPreferences> salFacade;
  59     private final ScheduledExecutorService executor;
  60     private final long keepaliveDelaySeconds;
  61     private final ResetKeepalive resetKeepaliveTask;
  62     private final long defaultRequestTimeoutMillis;
  63
  64     private volatile NetconfDeviceCommunicator listener;
  65     private volatile ScheduledFuture<?> currentKeepalive;
  66     private volatile DOMRpcService currentDeviceRpc;
  67
  68     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
  69                               final ScheduledExecutorService executor, final long keepaliveDelaySeconds, final long defaultRequestTimeoutMillis) {
  70         this.id = id;
  71         this.salFacade = salFacade;
  72         this.executor = executor;
  73         this.keepaliveDelaySeconds = keepaliveDelaySeconds;
  74         this.defaultRequestTimeoutMillis = defaultRequestTimeoutMillis;
  75         this.resetKeepaliveTask = new ResetKeepalive();
  76     }
  77
  78     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
  79                               final ScheduledExecutorService executor) {
  80         this(id, salFacade, executor, DEFAULT_DELAY, DEFAULT_TRANSACTION_TIMEOUT_MILLI);
  81     }
  82
  83     /**
  84      * Set the netconf session listener whenever ready
  85      *
  86      * @param listener netconf session listener
  87      */
  88     public void setListener(final NetconfDeviceCommunicator listener) {
  89         this.listener = listener;
  90     }
  91
  92     /**
  93      * Just cancel current keepalive task.
  94      * If its already started, let it finish ... not such a big deal.
  95      *
  96      * Then schedule next keepalive.
  97      */
  98     private void resetKeepalive() {
  99         LOG.trace("{}: Resetting netconf keepalive timer", id);
 100         if(currentKeepalive != null) {
 101             currentKeepalive.cancel(false);
 102         }
 103         scheduleKeepalive();
 104     }
 105
 106     /**
 107      * Cancel current keepalive and also reset current deviceRpc
 108      */
 109     private void stopKeepalives() {
 110         if(currentKeepalive != null) {
 111             currentKeepalive.cancel(false);
 112         }
 113         currentDeviceRpc = null;
 114     }
 115
 116     private void reconnect() {
 117         Preconditions.checkState(listener != null, "%s: Unable to reconnect, session listener is missing", id);
 118         stopKeepalives();
 119         LOG.info("{}: Reconnecting inactive netconf session", id);
 120         listener.disconnect();
 121     }
 122
 123     @Override
 124     public void onDeviceConnected(final SchemaContext remoteSchemaContext, final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc) {
 125         this.currentDeviceRpc = deviceRpc;
 126         final DOMRpcService deviceRpc1 = new KeepaliveDOMRpcService(deviceRpc, resetKeepaliveTask, defaultRequestTimeoutMillis, executor);
 127         salFacade.onDeviceConnected(remoteSchemaContext, netconfSessionPreferences, deviceRpc1);
 128
 129         LOG.debug("{}: Netconf session initiated, starting keepalives", id);
 130         scheduleKeepalive();
 131     }
 132
 133     private void scheduleKeepalive() {
 134         Preconditions.checkState(currentDeviceRpc != null);
 135         LOG.trace("{}: Scheduling next keepalive in {} {}", id, keepaliveDelaySeconds, TimeUnit.SECONDS);
 136         currentKeepalive = executor.schedule(new Keepalive(currentKeepalive), keepaliveDelaySeconds, TimeUnit.SECONDS);
 137     }
 138
 139     @Override
 140     public void onDeviceDisconnected() {
 141         stopKeepalives();
 142         salFacade.onDeviceDisconnected();
 143     }
 144
 145     @Override
 146     public void onDeviceFailed(final Throwable throwable) {
 147         stopKeepalives();
 148         salFacade.onDeviceFailed(throwable);
 149     }
 150
 151     @Override
 152     public void onNotification(final DOMNotification domNotification) {
 153         resetKeepalive();
 154         salFacade.onNotification(domNotification);
 155     }
 156
 157     @Override
 158     public void close() {
 159         stopKeepalives();
 160         salFacade.close();
 161     }
 162
 163     // Keepalive RPC static resources
 164     private static final SchemaPath PATH = toPath(NETCONF_GET_CONFIG_QNAME);
 165     private static final ContainerNode KEEPALIVE_PAYLOAD =
 166             NetconfMessageTransformUtil.wrap(NETCONF_GET_CONFIG_QNAME, getSourceNode(NETCONF_RUNNING_QNAME), NetconfMessageTransformUtil.EMPTY_FILTER);
 167
 168     /**
 169      * Invoke keepalive RPC and check the response. In case of any received response the keepalive
 170      * is considered successful and schedules next keepalive with a fixed delay. If the response is unsuccessful (no
 171      * response received, or the rcp could not even be sent) immediate reconnect is triggered as netconf session
 172      * is considered inactive/failed.
 173      */
 174     private class Keepalive implements Runnable, FutureCallback<DOMRpcResult> {
 175
 176         private final ScheduledFuture<?> previousKeepalive;
 177
 178         public Keepalive(final ScheduledFuture<?> previousKeepalive) {
 179             this.previousKeepalive = previousKeepalive;
 180         }
 181
 182         @Override
 183         public void run() {
 184             LOG.trace("{}: Invoking keepalive RPC", id);
 185
 186             try {
 187                 if(previousKeepalive != null && !previousKeepalive.isDone()) {
 188                     onFailure(new IllegalStateException("Previous keepalive timed out"));
 189                 } else {
 190                     Futures.addCallback(currentDeviceRpc.invokeRpc(PATH, KEEPALIVE_PAYLOAD), this);
 191                 }
 192             } catch (NullPointerException e) {
 193                 LOG.debug("{}: Skipping keepalive while reconnecting", id);
 194                 // Empty catch block intentional
 195                 // Do nothing. The currentDeviceRpc was null and it means we hit the reconnect window and
 196                 // attempted to send keepalive while we were reconnecting. Next keepalive will be scheduled
 197                 // after reconnect so no action necessary here.
 198             }
 199         }
 200
 201         @Override
 202         public void onSuccess(final DOMRpcResult result) {
 203             if (result != null && result.getResult() != null) {
 204                 LOG.debug("{}: Keepalive RPC successful with response: {}", id, result.getResult());
 205                 scheduleKeepalive();
 206             } else {
 207                 LOG.warn("{} Keepalive RPC returned null with response: {}. Reconnecting netconf session", id, result);
 208                 reconnect();
 209             }
 210         }
 211
 212         @Override
 213         public void onFailure(@Nonnull final Throwable t) {
 214             LOG.warn("{}: Keepalive RPC failed. Reconnecting netconf session.", id, t);
 215             reconnect();
 216         }
 217     }
 218
 219     /**
 220      * Reset keepalive after each RPC response received
 221      */
 222     private class ResetKeepalive implements FutureCallback<DOMRpcResult> {
 223         @Override
 224         public void onSuccess(@Nullable final DOMRpcResult result) {
 225             // No matter what response we got, rpc-reply or rpc-error, we got it from device so the netconf session is OK
 226             resetKeepalive();
 227         }
 228
 229         @Override
 230         public void onFailure(@Nonnull final Throwable t) {
 231             // User/Application RPC failed (The RPC did not reach the remote device or .. TODO what other reasons could cause this ?)
 232             // There is no point in keeping this session. Reconnect.
 233             LOG.warn("{}: Rpc failure detected. Reconnecting netconf session", id, t);
 234             reconnect();
 235         }
 236     }
 237
 238     /*
 239      * Request timeout task is called once the defaultRequestTimeoutMillis is
 240      * reached. At this moment, if the request is not yet finished, we cancel
 241      * it.
 242      */
 243     private static final class RequestTimeoutTask implements Runnable {
 244
 245         private final CheckedFuture<DOMRpcResult, DOMRpcException> rpcResultFuture;
 246
 247         public RequestTimeoutTask(final CheckedFuture<DOMRpcResult, DOMRpcException> rpcResultFuture) {
 248             this.rpcResultFuture = rpcResultFuture;
 249         }
 250
 251         @Override
 252         public void run() {
 253             if (!rpcResultFuture.isDone()) {
 254                 rpcResultFuture.cancel(true);
 255             }
 256         }
 257     }
 258
 259     /**
 260      * DOMRpcService proxy that attaches reset-keepalive-task and schedule
 261      * request-timeout-task to each RPC invocation.
 262      */
 263     private static final class KeepaliveDOMRpcService implements DOMRpcService {
 264
 265         private final DOMRpcService deviceRpc;
 266         private ResetKeepalive resetKeepaliveTask;
 267         private final long defaultRequestTimeoutMillis;
 268         private final ScheduledExecutorService executor;
 269
 270         public KeepaliveDOMRpcService(final DOMRpcService deviceRpc, final ResetKeepalive resetKeepaliveTask,
 271                 final long defaultRequestTimeoutMillis, final ScheduledExecutorService executor) {
 272             this.deviceRpc = deviceRpc;
 273             this.resetKeepaliveTask = resetKeepaliveTask;
 274             this.defaultRequestTimeoutMillis = defaultRequestTimeoutMillis;
 275             this.executor = executor;
 276         }
 277
 278         @Nonnull
 279         @Override
 280         public CheckedFuture<DOMRpcResult, DOMRpcException> invokeRpc(@Nonnull final SchemaPath type, final NormalizedNode<?, ?> input) {
 281             final CheckedFuture<DOMRpcResult, DOMRpcException> domRpcResultDOMRpcExceptionCheckedFuture = deviceRpc.invokeRpc(type, input);
 282             Futures.addCallback(domRpcResultDOMRpcExceptionCheckedFuture, resetKeepaliveTask);
 283
 284             final RequestTimeoutTask timeoutTask = new RequestTimeoutTask(domRpcResultDOMRpcExceptionCheckedFuture);
 285             executor.schedule(timeoutTask, defaultRequestTimeoutMillis, TimeUnit.MILLISECONDS);
 286
 287             return domRpcResultDOMRpcExceptionCheckedFuture;
 288         }
 289
 290         @Override
 291         public <T extends DOMRpcAvailabilityListener> ListenerRegistration<T> registerRpcListener(@Nonnull final T listener) {
 292             // There is no real communication with the device (yet), no reset here
 293             return deviceRpc.registerRpcListener(listener);
 294         }
 295     }
 296 }