2 * Copyright (c) 2016 Cisco Systems, Inc. and others. All rights reserved.
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
8 package org.opendaylight.controller.cluster.databroker.actors.dds;
10 import akka.actor.ActorRef;
11 import akka.actor.Status;
12 import com.google.common.base.Throwables;
13 import com.google.common.base.Verify;
14 import java.util.ArrayList;
15 import java.util.Collection;
17 import java.util.concurrent.ConcurrentHashMap;
18 import java.util.concurrent.atomic.AtomicLong;
19 import java.util.concurrent.locks.StampedLock;
20 import org.opendaylight.controller.cluster.access.client.BackendInfoResolver;
21 import org.opendaylight.controller.cluster.access.client.ClientActorBehavior;
22 import org.opendaylight.controller.cluster.access.client.ClientActorContext;
23 import org.opendaylight.controller.cluster.access.client.ConnectedClientConnection;
24 import org.opendaylight.controller.cluster.access.client.ConnectionEntry;
25 import org.opendaylight.controller.cluster.access.client.ReconnectForwarder;
26 import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
27 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
28 import org.slf4j.Logger;
29 import org.slf4j.LoggerFactory;
32 * {@link ClientActorBehavior} acting as an intermediary between the backend actors and the DistributedDataStore
36 * This class is not visible outside of this package because it breaks the actor containment. Services provided to
37 * Java world outside of actor containment are captured in {@link DataStoreClient}.
40 * IMPORTANT: this class breaks actor containment via methods implementing {@link DataStoreClient} contract.
41 * When touching internal state, be mindful of the execution context from which execution context, Actor
42 * or POJO, is the state being accessed or modified.
45 * THREAD SAFETY: this class must always be kept thread-safe, so that both the Actor System thread and the application
46 * threads can run concurrently. All state transitions must be made in a thread-safe manner. When in
47 * doubt, feel free to synchronize on this object.
50 * PERFORMANCE: this class lies in a performance-critical fast path. All code needs to be concise and efficient, but
51 * performance must not come at the price of correctness. Any optimizations need to be carefully analyzed
52 * for correctness and performance impact.
55 * TRADE-OFFS: part of the functionality runs in application threads without switching contexts, which makes it ideal
56 * for performing work and charging applications for it. That has two positive effects:
57 * - CPU usage is distributed across applications, minimizing work done in the actor thread
58 * - CPU usage provides back-pressure towards the application.
60 * @author Robert Varga
62 abstract class AbstractDataStoreClientBehavior extends ClientActorBehavior<ShardBackendInfo>
63 implements DataStoreClient {
64 private static final Logger LOG = LoggerFactory.getLogger(AbstractDataStoreClientBehavior.class);
66 private final Map<LocalHistoryIdentifier, ClientLocalHistory> histories = new ConcurrentHashMap<>();
67 private final AtomicLong nextHistoryId = new AtomicLong(1);
68 private final StampedLock lock = new StampedLock();
69 private final SingleClientHistory singleHistory;
71 private volatile Throwable aborted;
73 AbstractDataStoreClientBehavior(final ClientActorContext context,
74 final BackendInfoResolver<ShardBackendInfo> resolver) {
75 super(context, resolver);
76 singleHistory = new SingleClientHistory(this, new LocalHistoryIdentifier(getIdentifier(), 0));
81 // Methods below are invoked from the client actor thread
86 protected final void haltClient(final Throwable cause) {
87 // If we have encountered a previous problem there is no cleanup necessary, as we have already cleaned up
88 // Thread safely is not an issue, as both this method and any failures are executed from the same (client actor)
90 if (aborted != null) {
91 abortOperations(cause);
95 private void abortOperations(final Throwable cause) {
96 final long stamp = lock.writeLock();
98 // This acts as a barrier, application threads check this after they have added an entry in the maps,
99 // and if they observe aborted being non-null, they will perform their cleanup and not return the handle.
102 for (ClientLocalHistory h : histories.values()) {
107 lock.unlockWrite(stamp);
111 private AbstractDataStoreClientBehavior shutdown(final ClientActorBehavior<ShardBackendInfo> currentBehavior) {
112 abortOperations(new IllegalStateException("Client " + getIdentifier() + " has been shut down"));
117 protected final AbstractDataStoreClientBehavior onCommand(final Object command) {
118 if (command instanceof GetClientRequest) {
119 ((GetClientRequest) command).getReplyTo().tell(new Status.Success(this), ActorRef.noSender());
121 LOG.warn("{}: ignoring unhandled command {}", persistenceId(), command);
128 * The connection has resolved, which means we have to potentially perform message adaptation. This is a bit more
129 * involved, as the messages need to be replayed to the individual proxies.
132 protected final ConnectionConnectCohort connectionUp(final ConnectedClientConnection<ShardBackendInfo> newConn) {
133 final long stamp = lock.writeLock();
135 // Step 1: Freeze all AbstractProxyHistory instances pointing to that shard. This indirectly means that no
136 // further TransactionProxies can be created and we can safely traverse maps without risking
138 final Collection<HistoryReconnectCohort> cohorts = new ArrayList<>();
139 startReconnect(singleHistory, newConn, cohorts);
140 for (ClientLocalHistory h : histories.values()) {
141 startReconnect(h, newConn, cohorts);
144 return previousEntries -> finishReconnect(newConn, stamp, cohorts, previousEntries);
147 private ReconnectForwarder finishReconnect(final ConnectedClientConnection<ShardBackendInfo> newConn,
148 final long stamp, final Collection<HistoryReconnectCohort> cohorts,
149 final Collection<ConnectionEntry> previousEntries) {
151 // Step 2: Collect previous successful requests from the cohorts. We do not want to expose
152 // the non-throttling interface to the connection, hence we use a wrapper consumer
153 for (HistoryReconnectCohort c : cohorts) {
154 c.replayRequests(previousEntries);
157 // Step 3: Install a forwarder, which will forward requests back to affected cohorts. Any outstanding
158 // requests will be immediately sent to it and requests being sent concurrently will get
159 // forwarded once they hit the new connection.
160 return BouncingReconnectForwarder.forCohorts(newConn, cohorts);
163 // Step 4: Complete switchover of the connection. The cohorts can resume normal operations.
164 for (HistoryReconnectCohort c : cohorts) {
168 lock.unlockWrite(stamp);
173 private static void startReconnect(final AbstractClientHistory history,
174 final ConnectedClientConnection<ShardBackendInfo> newConn,
175 final Collection<HistoryReconnectCohort> cohorts) {
176 final HistoryReconnectCohort cohort = history.startReconnect(newConn);
177 if (cohort != null) {
184 // Methods below are invoked from application threads
189 public final ClientLocalHistory createLocalHistory() {
190 final LocalHistoryIdentifier historyId = new LocalHistoryIdentifier(getIdentifier(),
191 nextHistoryId.getAndIncrement());
193 final long stamp = lock.readLock();
195 if (aborted != null) {
196 Throwables.throwIfUnchecked(aborted);
197 throw new RuntimeException(aborted);
200 final ClientLocalHistory history = new ClientLocalHistory(this, historyId);
201 LOG.debug("{}: creating a new local history {}", persistenceId(), history);
203 Verify.verify(histories.put(historyId, history) == null);
206 lock.unlockRead(stamp);
211 public final ClientTransaction createTransaction() {
212 return singleHistory.createTransaction();
216 public final ClientSnapshot createSnapshot() {
217 return singleHistory.takeSnapshot();
221 public void close() {
223 context().executeInActor(this::shutdown);
226 abstract Long resolveShardForPath(YangInstanceIdentifier path);