BUG-3335 Add keepalive mechanism to netconf-connector
[controller.git] / opendaylight / md-sal / sal-netconf-connector / src / main / java / org / opendaylight / controller / 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.controller.sal.connect.netconf.sal;
9
10 import static org.opendaylight.controller.sal.connect.netconf.util.NetconfBaseOps.getSourceNode;
11 import static org.opendaylight.controller.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_GET_CONFIG_QNAME;
12 import static org.opendaylight.controller.sal.connect.netconf.util.NetconfMessageTransformUtil.NETCONF_RUNNING_QNAME;
13 import static org.opendaylight.controller.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.controller.sal.connect.api.RemoteDeviceHandler;
30 import org.opendaylight.controller.sal.connect.netconf.listener.NetconfDeviceCommunicator;
31 import org.opendaylight.controller.sal.connect.netconf.listener.NetconfSessionPreferences;
32 import org.opendaylight.controller.sal.connect.netconf.util.NetconfMessageTransformUtil;
33 import org.opendaylight.controller.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     private final RemoteDeviceId id;
55     private final RemoteDeviceHandler<NetconfSessionPreferences> salFacade;
56     private final ScheduledExecutorService executor;
57     private final long keepaliveDelaySeconds;
58     private final ResetKeepalive resetKeepaliveTask;
59
60     private volatile NetconfDeviceCommunicator listener;
61     private volatile ScheduledFuture<?> currentKeepalive;
62     private volatile DOMRpcService currentDeviceRpc;
63
64     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
65                               final ScheduledExecutorService executor, final long keepaliveDelaySeconds) {
66         this.id = id;
67         this.salFacade = salFacade;
68         this.executor = executor;
69         this.keepaliveDelaySeconds = keepaliveDelaySeconds;
70         this.resetKeepaliveTask = new ResetKeepalive();
71     }
72
73     public KeepaliveSalFacade(final RemoteDeviceId id, final RemoteDeviceHandler<NetconfSessionPreferences> salFacade,
74                               final ScheduledExecutorService executor) {
75         this(id, salFacade, executor, DEFAULT_DELAY);
76     }
77
78     /**
79      * Set the netconf session listener whenever ready
80      *
81      * @param listener netconf session listener
82      */
83     public void setListener(final NetconfDeviceCommunicator listener) {
84         this.listener = listener;
85     }
86
87     /**
88      * Just cancel current keepalive task.
89      * If its already started, let it finish ... not such a big deal.
90      *
91      * Then schedule next keepalive.
92      */
93     private void resetKeepalive() {
94         LOG.trace("{}: Resetting netconf keepalive timer", id);
95         currentKeepalive.cancel(false);
96         scheduleKeepalive();
97     }
98
99     /**
100      * Cancel current keepalive and also reset current deviceRpc
101      */
102     private void stopKeepalives() {
103         currentKeepalive.cancel(false);
104         currentDeviceRpc = null;
105     }
106
107     private void reconnect() {
108         Preconditions.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 SchemaContext remoteSchemaContext, final NetconfSessionPreferences netconfSessionPreferences, final DOMRpcService deviceRpc) {
116         this.currentDeviceRpc = deviceRpc;
117         final DOMRpcService deviceRpc1 = new KeepaliveDOMRpcService(deviceRpc, resetKeepaliveTask);
118         salFacade.onDeviceConnected(remoteSchemaContext, netconfSessionPreferences, deviceRpc1);
119
120         LOG.debug("{}: Netconf session initiated, starting keepalives", id);
121         scheduleKeepalive();
122     }
123
124     private void scheduleKeepalive() {
125         Preconditions.checkState(currentDeviceRpc != null);
126         LOG.trace("{}: Scheduling next keepalive in {} {}", id, keepaliveDelaySeconds, TimeUnit.SECONDS);
127         currentKeepalive = executor.schedule(new Keepalive(), keepaliveDelaySeconds, TimeUnit.SECONDS);
128     }
129
130     @Override
131     public void onDeviceDisconnected() {
132         stopKeepalives();
133         salFacade.onDeviceDisconnected();
134     }
135
136     @Override
137     public void onDeviceFailed(final Throwable throwable) {
138         stopKeepalives();
139         salFacade.onDeviceFailed(throwable);
140     }
141
142     @Override
143     public void onNotification(final DOMNotification domNotification) {
144         resetKeepalive();
145         salFacade.onNotification(domNotification);
146     }
147
148     @Override
149     public void close() {
150         stopKeepalives();
151         salFacade.close();
152     }
153
154     // Keepalive RPC static resources
155     private static final SchemaPath PATH = toPath(NETCONF_GET_CONFIG_QNAME);
156     private static final ContainerNode KEEPALIVE_PAYLOAD =
157             NetconfMessageTransformUtil.wrap(NETCONF_GET_CONFIG_QNAME, getSourceNode(NETCONF_RUNNING_QNAME), NetconfMessageTransformUtil.EMPTY_FILTER);
158
159     /**
160      * Invoke keepalive RPC and check the response. In case of any received response the keepalive
161      * is considered successful and schedules next keepalive with a fixed delay. If the response is unsuccessful (no
162      * response received, or the rcp could not even be sent) immediate reconnect is triggered as netconf session
163      * is considered inactive/failed.
164      */
165     private class Keepalive implements Runnable, FutureCallback<DOMRpcResult> {
166
167         @Override
168         public void run() {
169             LOG.trace("{}: Invoking keepalive RPC", id);
170
171             try {
172                 Futures.addCallback(currentDeviceRpc.invokeRpc(PATH, KEEPALIVE_PAYLOAD), this);
173             } catch (NullPointerException e) {
174                 LOG.debug("{}: Skipping keepalive while reconnecting", id);
175                 // Empty catch block intentional
176                 // Do nothing. The currentDeviceRpc was null and it means we hit the reconnect window and
177                 // attempted to send keepalive while we were reconnecting. Next keepalive will be scheduled
178                 // after reconnect so no action necessary here.
179             }
180         }
181
182         @Override
183         public void onSuccess(final DOMRpcResult result) {
184             LOG.debug("{}: Keepalive RPC successful with response: {}", id, result.getResult());
185             scheduleKeepalive();
186         }
187
188         @Override
189         public void onFailure(@Nonnull final Throwable t) {
190             LOG.warn("{}: Keepalive RPC failed. Reconnecting netconf session.", id, t);
191             reconnect();
192         }
193     }
194
195     /**
196      * Reset keepalive after each RPC response received
197      */
198     private class ResetKeepalive implements com.google.common.util.concurrent.FutureCallback<DOMRpcResult> {
199         @Override
200         public void onSuccess(@Nullable final DOMRpcResult result) {
201             // No matter what response we got, rpc-reply or rpc-error, we got it from device so the netconf session is OK
202             resetKeepalive();
203         }
204
205         @Override
206         public void onFailure(@Nonnull final Throwable t) {
207             // User/Application RPC failed (The RPC did not reach the remote device or .. TODO what other reasons could cause this ?)
208             // There is no point in keeping this session. Reconnect.
209             LOG.warn("{}: Rpc failure detected. Reconnecting netconf session", id, t);
210             reconnect();
211         }
212     }
213
214     /**
215      * DOMRpcService proxy that attaches reset-keepalive-task to each RPC invocation.
216      */
217     private static final class KeepaliveDOMRpcService implements DOMRpcService {
218
219         private final DOMRpcService deviceRpc;
220         private ResetKeepalive resetKeepaliveTask;
221
222         public KeepaliveDOMRpcService(final DOMRpcService deviceRpc, final ResetKeepalive resetKeepaliveTask) {
223             this.deviceRpc = deviceRpc;
224             this.resetKeepaliveTask = resetKeepaliveTask;
225         }
226
227         @Nonnull
228         @Override
229         public CheckedFuture<DOMRpcResult, DOMRpcException> invokeRpc(@Nonnull final SchemaPath type, final NormalizedNode<?, ?> input) {
230             final CheckedFuture<DOMRpcResult, DOMRpcException> domRpcResultDOMRpcExceptionCheckedFuture = deviceRpc.invokeRpc(type, input);
231             Futures.addCallback(domRpcResultDOMRpcExceptionCheckedFuture, resetKeepaliveTask);
232             return domRpcResultDOMRpcExceptionCheckedFuture;
233         }
234
235         @Override
236         public <T extends DOMRpcAvailabilityListener> ListenerRegistration<T> registerRpcListener(@Nonnull final T listener) {
237             // There is no real communication with the device (yet), no reset here
238             return deviceRpc.registerRpcListener(listener);
239         }
240     }
241 }