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 java.util.concurrent.atomic.AtomicBoolean;
  25 import javax.annotation.Nonnull;
  26 import javax.annotation.Nullable;
  27 import org.opendaylight.controller.md.sal.dom.api.DOMNotification;
  28 import org.opendaylight.controller.md.sal.dom.api.DOMRpcAvailabilityListener;
  29 import org.opendaylight.controller.md.sal.dom.api.DOMRpcException;
  30 import org.opendaylight.controller.md.sal.dom.api.DOMRpcResult;
  31 import org.opendaylight.controller.md.sal.dom.api.DOMRpcService;
  32 import org.opendaylight.netconf.sal.connect.api.RemoteDeviceHandler;
  33 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfDeviceCommunicator;
  34 import org.opendaylight.netconf.sal.connect.netconf.listener.NetconfSessionPreferences;
  35 import org.opendaylight.netconf.sal.connect.netconf.util.NetconfMessageTransformUtil;
  36 import org.opendaylight.netconf.sal.connect.util.RemoteDeviceId;
  37 import org.opendaylight.yangtools.concepts.ListenerRegistration;
  38 import org.opendaylight.yangtools.yang.data.api.schema.ContainerNode;
  39 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
  40 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  41 import org.opendaylight.yangtools.yang.model.api.SchemaPath;
  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<NetconfSessionPreferences> {
  51
  52     private static final Logger LOG = LoggerFactory.getLogger(KeepaliveSalFacade.class);
  53
  54     // 2 minutes keepalive delay by default
  55     private static final long DEFAULT_DELAY = TimeUnit.MINUTES.toSeconds(2);
  56
  57     // 1 minute transaction timeout by default
  58     private static final long DEFAULT_TRANSACTION_TIMEOUT_MILLI = TimeUnit.MILLISECONDS.toMillis(60000);
  59
  60     private final RemoteDeviceId id;
  61     private final RemoteDeviceHandler<NetconfSessionPreferences> salFacade;
  62     private final ScheduledExecutorService executor;
  63     private final long keepaliveDelaySeconds;
  64     private final ResetKeepalive resetKeepaliveTask;
  65     private final long defaultRequestTimeoutMillis;
  66
  67     private volatile NetconfDeviceCommunicator listener;
  68     private volatile ScheduledFuture<?> currentKeepalive;
  69     private volatile DOMRpcService currentDeviceRpc;
  70     private final AtomicBoolean lastKeepAliveSucceeded = new AtomicBoolean(false);
  71
  72     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
  73                               final ScheduledExecutorService executor, final long keepaliveDelaySeconds,
  74                               final long defaultRequestTimeoutMillis) {
  75         this.id = id;
  76         this.salFacade = salFacade;
  77         this.executor = executor;
  78         this.keepaliveDelaySeconds = keepaliveDelaySeconds;
  79         this.defaultRequestTimeoutMillis = defaultRequestTimeoutMillis;
  80         this.resetKeepaliveTask = new ResetKeepalive();
  81     }
  82
  83     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
  84                               final ScheduledExecutorService executor) {
  85         this(id, salFacade, executor, DEFAULT_DELAY, DEFAULT_TRANSACTION_TIMEOUT_MILLI);
  86     }
  87
  88     /**
  89      * Set the netconf session listener whenever ready.
  90      *
  91      * @param listener netconf session listener
  92      */
  93     public void setListener(final NetconfDeviceCommunicator listener) {
  94         this.listener = listener;
  95     }
  96
  97     /**
  98      * Just cancel current keepalive task.
  99      * If its already started, let it finish ... not such a big deal.
 100      *
 101      * <p>
 102      * Then schedule next keepalive.
 103      */
 104     void resetKeepalive() {
 105         LOG.trace("{}: Resetting netconf keepalive timer", id);
 106         if (currentKeepalive != null) {
 107             currentKeepalive.cancel(false);
 108         }
 109         scheduleKeepalives();
 110     }
 111
 112     /**
 113      * Cancel current keepalive and also reset current deviceRpc.
 114      */
 115     private void stopKeepalives() {
 116         if (currentKeepalive != null) {
 117             currentKeepalive.cancel(false);
 118         }
 119         currentDeviceRpc = null;
 120     }
 121
 122     void reconnect() {
 123         Preconditions.checkState(listener != null, "%s: Unable to reconnect, session listener is missing", id);
 124         stopKeepalives();
 125         LOG.info("{}: Reconnecting inactive netconf session", id);
 126         listener.disconnect();
 127     }
 128
 129     @Override
 130     public void onDeviceConnected(final SchemaContext remoteSchemaContext,
 131                           final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc) {
 132         this.currentDeviceRpc = deviceRpc;
 133         final DOMRpcService deviceRpc1 =
 134                 new KeepaliveDOMRpcService(deviceRpc, resetKeepaliveTask, defaultRequestTimeoutMillis, executor);
 135         salFacade.onDeviceConnected(remoteSchemaContext, netconfSessionPreferences, deviceRpc1);
 136
 137         LOG.debug("{}: Netconf session initiated, starting keepalives", id);
 138         scheduleKeepalives();
 139     }
 140
 141     private void scheduleKeepalives() {
 142         lastKeepAliveSucceeded.set(true);
 143         Preconditions.checkState(currentDeviceRpc != null);
 144         LOG.trace("{}: Scheduling keepalives every  {} {}", id, keepaliveDelaySeconds, TimeUnit.SECONDS);
 145         currentKeepalive = executor.scheduleWithFixedDelay(new Keepalive(),
 146           keepaliveDelaySeconds, keepaliveDelaySeconds, TimeUnit.SECONDS);
 147     }
 148
 149     @Override
 150     public void onDeviceDisconnected() {
 151         stopKeepalives();
 152         salFacade.onDeviceDisconnected();
 153     }
 154
 155     @Override
 156     public void onDeviceFailed(final Throwable throwable) {
 157         stopKeepalives();
 158         salFacade.onDeviceFailed(throwable);
 159     }
 160
 161     @Override
 162     public void onNotification(final DOMNotification domNotification) {
 163         resetKeepalive();
 164         salFacade.onNotification(domNotification);
 165     }
 166
 167     @Override
 168     public void close() {
 169         stopKeepalives();
 170         salFacade.close();
 171     }
 172
 173     // Keepalive RPC static resources
 174     private static final SchemaPath PATH = toPath(NETCONF_GET_CONFIG_QNAME);
 175     private static final ContainerNode KEEPALIVE_PAYLOAD = NetconfMessageTransformUtil.wrap(NETCONF_GET_CONFIG_QNAME,
 176             getSourceNode(NETCONF_RUNNING_QNAME), NetconfMessageTransformUtil.EMPTY_FILTER);
 177
 178     /**
 179      * Invoke keepalive RPC and check the response. In case of any received response the keepalive
 180      * is considered successful and schedules next keepalive with a fixed delay. If the response is unsuccessful (no
 181      * response received, or the rcp could not even be sent) immediate reconnect is triggered as netconf session
 182      * is considered inactive/failed.
 183      */
 184     private class Keepalive implements Runnable, FutureCallback<DOMRpcResult> {
 185
 186         @Override
 187         public void run() {
 188             LOG.trace("{}: Invoking keepalive RPC", id);
 189
 190             try {
 191                 boolean lastJobSucceeded = lastKeepAliveSucceeded.getAndSet(false);
 192                 if (!lastJobSucceeded) {
 193                     onFailure(new IllegalStateException("Previous keepalive timed out"));
 194                 } else {
 195                     Futures.addCallback(currentDeviceRpc.invokeRpc(PATH, KEEPALIVE_PAYLOAD), this,
 196                                         MoreExecutors.directExecutor());
 197                 }
 198             } catch (NullPointerException e) {
 199                 LOG.debug("{}: Skipping keepalive while reconnecting", id);
 200                 // Empty catch block intentional
 201                 // Do nothing. The currentDeviceRpc was null and it means we hit the reconnect window and
 202                 // attempted to send keepalive while we were reconnecting. Next keepalive will be scheduled
 203                 // after reconnect so no action necessary here.
 204             }
 205         }
 206
 207         @SuppressFBWarnings(value = "RCN_REDUNDANT_NULLCHECK_OF_NONNULL_VALUE",
 208                 justification = "Unrecognised NullableDecl")
 209         @Override
 210         public void onSuccess(final DOMRpcResult result) {
 211             // No matter what response we got, rpc-reply or rpc-error,
 212             // we got it from device so the netconf session is OK
 213             if (result != null && result.getResult() != null) {
 214                 lastKeepAliveSucceeded.set(true);
 215             }  else if (result != null && result.getErrors() != null) {
 216                 LOG.warn("{}: Keepalive RPC failed with error: {}", id, result.getErrors());
 217                 lastKeepAliveSucceeded.set(true);
 218             } else {
 219                 LOG.warn("{} Keepalive RPC returned null with response. Reconnecting netconf session", id);
 220                 reconnect();
 221             }
 222         }
 223
 224         @Override
 225         public void onFailure(@Nonnull final Throwable throwable) {
 226             LOG.warn("{}: Keepalive RPC failed. Reconnecting netconf session.", id, throwable);
 227             reconnect();
 228         }
 229     }
 230
 231     /**
 232      * Reset keepalive after each RPC response received.
 233      */
 234     private class ResetKeepalive implements FutureCallback<DOMRpcResult> {
 235         @Override
 236         public void onSuccess(@Nullable final DOMRpcResult result) {
 237             // No matter what response we got,
 238             // rpc-reply or rpc-error, we got it from device so the netconf session is OK.
 239             resetKeepalive();
 240         }
 241
 242         @Override
 243         public void onFailure(@Nonnull final Throwable throwable) {
 244             // User/Application RPC failed (The RPC did not reach the remote device or ..
 245             // TODO what other reasons could cause this ?)
 246             // There is no point in keeping this session. Reconnect.
 247             LOG.warn("{}: Rpc failure detected. Reconnecting netconf session", id, throwable);
 248             reconnect();
 249         }
 250     }
 251
 252     /*
 253      * Request timeout task is called once the defaultRequestTimeoutMillis is
 254      * reached. At this moment, if the request is not yet finished, we cancel
 255      * it.
 256      */
 257     private static final class RequestTimeoutTask implements Runnable {
 258
 259         private final CheckedFuture<DOMRpcResult, DOMRpcException> rpcResultFuture;
 260
 261         RequestTimeoutTask(final CheckedFuture<DOMRpcResult, DOMRpcException> rpcResultFuture) {
 262             this.rpcResultFuture = rpcResultFuture;
 263         }
 264
 265         @Override
 266         public void run() {
 267             if (!rpcResultFuture.isDone()) {
 268                 rpcResultFuture.cancel(true);
 269             }
 270         }
 271     }
 272
 273     /**
 274      * DOMRpcService proxy that attaches reset-keepalive-task and schedule
 275      * request-timeout-task to each RPC invocation.
 276      */
 277     public static final class KeepaliveDOMRpcService implements DOMRpcService {
 278
 279         private final DOMRpcService deviceRpc;
 280         private final ResetKeepalive resetKeepaliveTask;
 281         private final long defaultRequestTimeoutMillis;
 282         private final ScheduledExecutorService executor;
 283
 284         KeepaliveDOMRpcService(final DOMRpcService deviceRpc, final ResetKeepalive resetKeepaliveTask,
 285                 final long defaultRequestTimeoutMillis, final ScheduledExecutorService executor) {
 286             this.deviceRpc = deviceRpc;
 287             this.resetKeepaliveTask = resetKeepaliveTask;
 288             this.defaultRequestTimeoutMillis = defaultRequestTimeoutMillis;
 289             this.executor = executor;
 290         }
 291
 292         public DOMRpcService getDeviceRpc() {
 293             return deviceRpc;
 294         }
 295
 296         @Nonnull
 297         @Override
 298         public CheckedFuture<DOMRpcResult, DOMRpcException> invokeRpc(@Nonnull final SchemaPath type,
 299                                                                       final NormalizedNode<?, ?> input) {
 300             final CheckedFuture<DOMRpcResult, DOMRpcException> domRpcResultDOMRpcExceptionCheckedFuture =
 301                     deviceRpc.invokeRpc(type, input);
 302             Futures.addCallback(domRpcResultDOMRpcExceptionCheckedFuture, resetKeepaliveTask,
 303                                 MoreExecutors.directExecutor());
 304
 305             final RequestTimeoutTask timeoutTask = new RequestTimeoutTask(domRpcResultDOMRpcExceptionCheckedFuture);
 306             executor.schedule(timeoutTask, defaultRequestTimeoutMillis, TimeUnit.MILLISECONDS);
 307
 308             return domRpcResultDOMRpcExceptionCheckedFuture;
 309         }
 310
 311         @Override
 312         public <T extends DOMRpcAvailabilityListener> ListenerRegistration<T> registerRpcListener(
 313                 @Nonnull final T listener) {
 314             // There is no real communication with the device (yet), no reset here
 315             return deviceRpc.registerRpcListener(listener);
 316         }
 317     }
 318 }