Thomas Vachuska | 58de416 | 2015-09-10 16:15:33 -0700 | [diff] [blame] | 1 | /* |
Brian O'Connor | a09fe5b | 2017-08-03 21:12:30 -0700 | [diff] [blame] | 2 | * Copyright 2015-present Open Networking Foundation |
Thomas Vachuska | 58de416 | 2015-09-10 16:15:33 -0700 | [diff] [blame] | 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 16 | package org.onosproject.store.cluster.messaging.impl; |
| 17 | |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 18 | import javax.net.ssl.KeyManagerFactory; |
| 19 | import javax.net.ssl.SSLContext; |
| 20 | import javax.net.ssl.SSLEngine; |
| 21 | import javax.net.ssl.TrustManagerFactory; |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 22 | import java.io.File; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 23 | import java.io.FileInputStream; |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 24 | import java.io.FileNotFoundException; |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 25 | import java.net.ConnectException; |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 26 | import java.security.Key; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 27 | import java.security.KeyStore; |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 28 | import java.security.MessageDigest; |
| 29 | import java.security.cert.Certificate; |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 30 | import java.time.Duration; |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 31 | import java.util.ArrayList; |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 32 | import java.util.Enumeration; |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 33 | import java.util.Iterator; |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 34 | import java.util.List; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 35 | import java.util.Map; |
Madan Jampani | a9e70a6 | 2016-03-02 16:28:18 -0800 | [diff] [blame] | 36 | import java.util.Optional; |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 37 | import java.util.StringJoiner; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 38 | import java.util.concurrent.CompletableFuture; |
| 39 | import java.util.concurrent.ConcurrentHashMap; |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 40 | import java.util.concurrent.ExecutionException; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 41 | import java.util.concurrent.Executor; |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 42 | import java.util.concurrent.Executors; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 43 | import java.util.concurrent.RejectedExecutionException; |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 44 | import java.util.concurrent.ScheduledExecutorService; |
| 45 | import java.util.concurrent.ScheduledFuture; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 46 | import java.util.concurrent.TimeUnit; |
| 47 | import java.util.concurrent.TimeoutException; |
| 48 | import java.util.concurrent.atomic.AtomicBoolean; |
| 49 | import java.util.concurrent.atomic.AtomicLong; |
| 50 | import java.util.function.BiConsumer; |
| 51 | import java.util.function.BiFunction; |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 52 | import java.util.function.Function; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 53 | |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 54 | import com.google.common.cache.Cache; |
| 55 | import com.google.common.cache.CacheBuilder; |
| 56 | import com.google.common.collect.Lists; |
| 57 | import com.google.common.collect.Maps; |
| 58 | import com.google.common.util.concurrent.MoreExecutors; |
| 59 | import io.netty.bootstrap.Bootstrap; |
| 60 | import io.netty.bootstrap.ServerBootstrap; |
| 61 | import io.netty.buffer.PooledByteBufAllocator; |
| 62 | import io.netty.channel.Channel; |
| 63 | import io.netty.channel.ChannelFuture; |
| 64 | import io.netty.channel.ChannelHandler; |
| 65 | import io.netty.channel.ChannelHandlerContext; |
| 66 | import io.netty.channel.ChannelInitializer; |
| 67 | import io.netty.channel.ChannelOption; |
| 68 | import io.netty.channel.EventLoopGroup; |
| 69 | import io.netty.channel.ServerChannel; |
| 70 | import io.netty.channel.SimpleChannelInboundHandler; |
| 71 | import io.netty.channel.WriteBufferWaterMark; |
| 72 | import io.netty.channel.epoll.EpollEventLoopGroup; |
| 73 | import io.netty.channel.epoll.EpollServerSocketChannel; |
| 74 | import io.netty.channel.epoll.EpollSocketChannel; |
| 75 | import io.netty.channel.nio.NioEventLoopGroup; |
| 76 | import io.netty.channel.socket.SocketChannel; |
| 77 | import io.netty.channel.socket.nio.NioServerSocketChannel; |
| 78 | import io.netty.channel.socket.nio.NioSocketChannel; |
| 79 | import org.apache.commons.math3.stat.descriptive.DescriptiveStatistics; |
| 80 | import org.apache.commons.math3.stat.descriptive.SynchronizedDescriptiveStatistics; |
| 81 | import org.apache.felix.scr.annotations.Activate; |
| 82 | import org.apache.felix.scr.annotations.Component; |
| 83 | import org.apache.felix.scr.annotations.Deactivate; |
| 84 | import org.apache.felix.scr.annotations.Reference; |
| 85 | import org.apache.felix.scr.annotations.ReferenceCardinality; |
| 86 | import org.apache.felix.scr.annotations.Service; |
| 87 | import org.onosproject.cluster.ClusterMetadataService; |
| 88 | import org.onosproject.cluster.ControllerNode; |
| 89 | import org.onosproject.core.HybridLogicalClockService; |
| 90 | import org.onosproject.store.cluster.messaging.Endpoint; |
| 91 | import org.onosproject.store.cluster.messaging.MessagingException; |
| 92 | import org.onosproject.store.cluster.messaging.MessagingService; |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 93 | import org.slf4j.Logger; |
| 94 | import org.slf4j.LoggerFactory; |
| 95 | |
Yuta HIGUCHI | 90a1689 | 2016-07-20 20:36:08 -0700 | [diff] [blame] | 96 | import static org.onlab.util.Tools.groupedThreads; |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 97 | import static org.onosproject.security.AppGuard.checkPermission; |
| 98 | import static org.onosproject.security.AppPermission.Type.CLUSTER_WRITE; |
| 99 | |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 100 | /** |
| 101 | * Netty based MessagingService. |
| 102 | */ |
Sho SHIMIZU | 5c396e3 | 2016-08-12 15:19:12 -0700 | [diff] [blame] | 103 | @Component(immediate = true) |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 104 | @Service |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 105 | public class NettyMessagingManager implements MessagingService { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 106 | private static final long DEFAULT_TIMEOUT_MILLIS = 500; |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 107 | private static final long HISTORY_EXPIRE_MILLIS = Duration.ofMinutes(1).toMillis(); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 108 | private static final long MIN_TIMEOUT_MILLIS = 100; |
| 109 | private static final long MAX_TIMEOUT_MILLIS = 5000; |
| 110 | private static final long TIMEOUT_INTERVAL = 50; |
| 111 | private static final int WINDOW_SIZE = 100; |
| 112 | private static final double TIMEOUT_MULTIPLIER = 2.5; |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 113 | private static final int CHANNEL_POOL_SIZE = 8; |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 114 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 115 | private static final byte[] EMPTY_PAYLOAD = new byte[0]; |
| 116 | |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 117 | private final Logger log = LoggerFactory.getLogger(getClass()); |
| 118 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 119 | private final ClientConnection localClientConnection = new LocalClientConnection(); |
| 120 | private final ServerConnection localServerConnection = new LocalServerConnection(null); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 121 | |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 122 | //TODO CONFIG_DIR is duplicated from ConfigFileBasedClusterMetadataProvider |
| 123 | private static final String CONFIG_DIR = "../config"; |
| 124 | private static final String KS_FILE_NAME = "onos.jks"; |
| 125 | private static final File DEFAULT_KS_FILE = new File(CONFIG_DIR, KS_FILE_NAME); |
| 126 | private static final String DEFAULT_KS_PASSWORD = "changeit"; |
| 127 | |
Madan Jampani | 0583387 | 2016-07-12 23:01:39 -0700 | [diff] [blame] | 128 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 129 | protected HybridLogicalClockService clockService; |
| 130 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 131 | private Endpoint localEndpoint; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 132 | private int preamble; |
| 133 | private final AtomicBoolean started = new AtomicBoolean(false); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 134 | private final Map<String, BiConsumer<InternalRequest, ServerConnection>> handlers = new ConcurrentHashMap<>(); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 135 | private final Map<Channel, RemoteClientConnection> clientConnections = Maps.newConcurrentMap(); |
| 136 | private final Map<Channel, RemoteServerConnection> serverConnections = Maps.newConcurrentMap(); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 137 | private final AtomicLong messageIdGenerator = new AtomicLong(0); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 138 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 139 | private ScheduledFuture<?> timeoutFuture; |
| 140 | |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 141 | private final Map<Endpoint, List<CompletableFuture<Channel>>> channels = Maps.newConcurrentMap(); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 142 | |
| 143 | private EventLoopGroup serverGroup; |
| 144 | private EventLoopGroup clientGroup; |
| 145 | private Class<? extends ServerChannel> serverChannelClass; |
| 146 | private Class<? extends Channel> clientChannelClass; |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 147 | private ScheduledExecutorService timeoutExecutor; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 148 | |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 149 | protected static final boolean TLS_ENABLED = true; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 150 | protected static final boolean TLS_DISABLED = false; |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 151 | protected boolean enableNettyTls = TLS_ENABLED; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 152 | |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 153 | protected TrustManagerFactory trustManager; |
| 154 | protected KeyManagerFactory keyManager; |
JunHuy Lam | 39eb429 | 2015-06-26 17:24:23 +0900 | [diff] [blame] | 155 | |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 156 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
Madan Jampani | ec1df02 | 2015-10-13 21:23:03 -0700 | [diff] [blame] | 157 | protected ClusterMetadataService clusterMetadataService; |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 158 | |
| 159 | @Activate |
| 160 | public void activate() throws Exception { |
Madan Jampani | ec1df02 | 2015-10-13 21:23:03 -0700 | [diff] [blame] | 161 | ControllerNode localNode = clusterMetadataService.getLocalNode(); |
Jonathan Hart | d9df7bd | 2015-11-10 17:10:25 -0800 | [diff] [blame] | 162 | getTlsParameters(); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 163 | |
| 164 | if (started.get()) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 165 | log.warn("Already running at local endpoint: {}", localEndpoint); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 166 | return; |
| 167 | } |
| 168 | this.preamble = clusterMetadataService.getClusterMetadata().getName().hashCode(); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 169 | this.localEndpoint = new Endpoint(localNode.ip(), localNode.tcpPort()); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 170 | initEventLoopGroup(); |
| 171 | startAcceptingConnections(); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 172 | timeoutExecutor = Executors.newSingleThreadScheduledExecutor( |
| 173 | groupedThreads("NettyMessagingEvt", "timeout", log)); |
| 174 | timeoutFuture = timeoutExecutor.scheduleAtFixedRate( |
| 175 | this::timeoutAllCallbacks, TIMEOUT_INTERVAL, TIMEOUT_INTERVAL, TimeUnit.MILLISECONDS); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 176 | started.set(true); |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 177 | log.info("Started"); |
| 178 | } |
| 179 | |
| 180 | @Deactivate |
| 181 | public void deactivate() throws Exception { |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 182 | if (started.get()) { |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 183 | serverGroup.shutdownGracefully(); |
| 184 | clientGroup.shutdownGracefully(); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 185 | timeoutFuture.cancel(false); |
| 186 | timeoutExecutor.shutdown(); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 187 | started.set(false); |
| 188 | } |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 189 | log.info("Stopped"); |
| 190 | } |
JunHuy Lam | 39eb429 | 2015-06-26 17:24:23 +0900 | [diff] [blame] | 191 | |
Jonathan Hart | d9df7bd | 2015-11-10 17:10:25 -0800 | [diff] [blame] | 192 | private void getTlsParameters() { |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 193 | // default is TLS enabled unless key stores cannot be loaded |
| 194 | enableNettyTls = Boolean.parseBoolean(System.getProperty("enableNettyTLS", Boolean.toString(TLS_ENABLED))); |
| 195 | |
Jonathan Hart | d9df7bd | 2015-11-10 17:10:25 -0800 | [diff] [blame] | 196 | if (enableNettyTls) { |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 197 | enableNettyTls = loadKeyStores(); |
| 198 | } |
| 199 | } |
| 200 | |
| 201 | private boolean loadKeyStores() { |
| 202 | // Maintain a local copy of the trust and key managers in case anything goes wrong |
| 203 | TrustManagerFactory tmf; |
| 204 | KeyManagerFactory kmf; |
| 205 | try { |
| 206 | String ksLocation = System.getProperty("javax.net.ssl.keyStore", DEFAULT_KS_FILE.toString()); |
| 207 | String tsLocation = System.getProperty("javax.net.ssl.trustStore", DEFAULT_KS_FILE.toString()); |
| 208 | char[] ksPwd = System.getProperty("javax.net.ssl.keyStorePassword", DEFAULT_KS_PASSWORD).toCharArray(); |
| 209 | char[] tsPwd = System.getProperty("javax.net.ssl.trustStorePassword", DEFAULT_KS_PASSWORD).toCharArray(); |
| 210 | |
| 211 | tmf = TrustManagerFactory.getInstance(TrustManagerFactory.getDefaultAlgorithm()); |
| 212 | KeyStore ts = KeyStore.getInstance("JKS"); |
| 213 | ts.load(new FileInputStream(tsLocation), tsPwd); |
| 214 | tmf.init(ts); |
| 215 | |
| 216 | kmf = KeyManagerFactory.getInstance(KeyManagerFactory.getDefaultAlgorithm()); |
| 217 | KeyStore ks = KeyStore.getInstance("JKS"); |
| 218 | ks.load(new FileInputStream(ksLocation), ksPwd); |
| 219 | kmf.init(ks, ksPwd); |
| 220 | if (log.isInfoEnabled()) { |
| 221 | logKeyStore(ks, ksLocation, ksPwd); |
JunHuy Lam | 39eb429 | 2015-06-26 17:24:23 +0900 | [diff] [blame] | 222 | } |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 223 | } catch (FileNotFoundException e) { |
| 224 | log.warn("Disabling TLS for intra-cluster messaging; Could not load cluster key store: {}", e.getMessage()); |
| 225 | return TLS_DISABLED; |
| 226 | } catch (Exception e) { |
| 227 | //TODO we might want to catch exceptions more specifically |
| 228 | log.error("Error loading key store; disabling TLS for intra-cluster messaging", e); |
| 229 | return TLS_DISABLED; |
| 230 | } |
| 231 | this.trustManager = tmf; |
| 232 | this.keyManager = kmf; |
| 233 | return TLS_ENABLED; |
| 234 | } |
| 235 | |
| 236 | private void logKeyStore(KeyStore ks, String ksLocation, char[] ksPwd) { |
| 237 | if (log.isInfoEnabled()) { |
| 238 | log.info("Loaded cluster key store from: {}", ksLocation); |
| 239 | try { |
| 240 | for (Enumeration<String> e = ks.aliases(); e.hasMoreElements();) { |
| 241 | String alias = e.nextElement(); |
| 242 | Key key = ks.getKey(alias, ksPwd); |
| 243 | Certificate[] certs = ks.getCertificateChain(alias); |
| 244 | log.debug("{} -> {}", alias, certs); |
| 245 | final byte[] encodedKey; |
| 246 | if (certs != null && certs.length > 0) { |
| 247 | encodedKey = certs[0].getEncoded(); |
| 248 | } else { |
| 249 | log.info("Could not find cert chain for {}, using fingerprint of key instead...", alias); |
| 250 | encodedKey = key.getEncoded(); |
| 251 | } |
| 252 | // Compute the certificate's fingerprint (use the key if certificate cannot be found) |
| 253 | MessageDigest digest = MessageDigest.getInstance("SHA1"); |
| 254 | digest.update(encodedKey); |
| 255 | StringJoiner fingerprint = new StringJoiner(":"); |
| 256 | for (byte b : digest.digest()) { |
| 257 | fingerprint.add(String.format("%02X", b)); |
| 258 | } |
| 259 | log.info("{} -> {}", alias, fingerprint); |
| 260 | } |
| 261 | } catch (Exception e) { |
| 262 | log.warn("Unable to print contents of key store: {}", ksLocation, e); |
JunHuy Lam | 39eb429 | 2015-06-26 17:24:23 +0900 | [diff] [blame] | 263 | } |
| 264 | } |
| 265 | } |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 266 | |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 267 | private void initEventLoopGroup() { |
| 268 | // try Epoll first and if that does work, use nio. |
| 269 | try { |
Yuta HIGUCHI | 90a1689 | 2016-07-20 20:36:08 -0700 | [diff] [blame] | 270 | clientGroup = new EpollEventLoopGroup(0, groupedThreads("NettyMessagingEvt", "epollC-%d", log)); |
| 271 | serverGroup = new EpollEventLoopGroup(0, groupedThreads("NettyMessagingEvt", "epollS-%d", log)); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 272 | serverChannelClass = EpollServerSocketChannel.class; |
| 273 | clientChannelClass = EpollSocketChannel.class; |
| 274 | return; |
| 275 | } catch (Throwable e) { |
| 276 | log.debug("Failed to initialize native (epoll) transport. " |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 277 | + "Reason: {}. Proceeding with nio.", e.getMessage()); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 278 | } |
Yuta HIGUCHI | 90a1689 | 2016-07-20 20:36:08 -0700 | [diff] [blame] | 279 | clientGroup = new NioEventLoopGroup(0, groupedThreads("NettyMessagingEvt", "nioC-%d", log)); |
| 280 | serverGroup = new NioEventLoopGroup(0, groupedThreads("NettyMessagingEvt", "nioS-%d", log)); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 281 | serverChannelClass = NioServerSocketChannel.class; |
| 282 | clientChannelClass = NioSocketChannel.class; |
| 283 | } |
| 284 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 285 | /** |
| 286 | * Times out response callbacks. |
| 287 | */ |
| 288 | private void timeoutAllCallbacks() { |
| 289 | // Iterate through all connections and time out callbacks. |
| 290 | for (RemoteClientConnection connection : clientConnections.values()) { |
| 291 | connection.timeoutCallbacks(); |
| 292 | } |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 293 | } |
| 294 | |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 295 | @Override |
| 296 | public CompletableFuture<Void> sendAsync(Endpoint ep, String type, byte[] payload) { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 297 | checkPermission(CLUSTER_WRITE); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 298 | InternalRequest message = new InternalRequest(preamble, |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 299 | clockService.timeNow(), |
| 300 | messageIdGenerator.incrementAndGet(), |
| 301 | localEndpoint, |
| 302 | type, |
| 303 | payload); |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 304 | return executeOnPooledConnection(ep, type, c -> c.sendAsync(message), MoreExecutors.directExecutor()); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 305 | } |
| 306 | |
| 307 | @Override |
| 308 | public CompletableFuture<byte[]> sendAndReceive(Endpoint ep, String type, byte[] payload) { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 309 | checkPermission(CLUSTER_WRITE); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 310 | return sendAndReceive(ep, type, payload, MoreExecutors.directExecutor()); |
| 311 | } |
| 312 | |
| 313 | @Override |
| 314 | public CompletableFuture<byte[]> sendAndReceive(Endpoint ep, String type, byte[] payload, Executor executor) { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 315 | checkPermission(CLUSTER_WRITE); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 316 | Long messageId = messageIdGenerator.incrementAndGet(); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 317 | InternalRequest message = new InternalRequest(preamble, |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 318 | clockService.timeNow(), |
| 319 | messageId, |
| 320 | localEndpoint, |
| 321 | type, |
| 322 | payload); |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 323 | return executeOnPooledConnection(ep, type, c -> c.sendAndReceive(message), executor); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 324 | } |
Jordan Halterman | 041534b | 2017-03-21 10:37:33 -0700 | [diff] [blame] | 325 | |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 326 | private List<CompletableFuture<Channel>> getChannelPool(Endpoint endpoint) { |
| 327 | return channels.computeIfAbsent(endpoint, e -> { |
| 328 | List<CompletableFuture<Channel>> defaultList = new ArrayList<>(CHANNEL_POOL_SIZE); |
| 329 | for (int i = 0; i < CHANNEL_POOL_SIZE; i++) { |
| 330 | defaultList.add(null); |
| 331 | } |
| 332 | return Lists.newCopyOnWriteArrayList(defaultList); |
| 333 | }); |
| 334 | } |
| 335 | |
| 336 | private int getChannelOffset(String messageType) { |
| 337 | return Math.abs(messageType.hashCode() % CHANNEL_POOL_SIZE); |
| 338 | } |
| 339 | |
| 340 | private CompletableFuture<Channel> getChannel(Endpoint endpoint, String messageType) { |
| 341 | List<CompletableFuture<Channel>> channelPool = getChannelPool(endpoint); |
| 342 | int offset = getChannelOffset(messageType); |
| 343 | |
| 344 | CompletableFuture<Channel> channelFuture = channelPool.get(offset); |
| 345 | if (channelFuture == null || channelFuture.isCompletedExceptionally()) { |
| 346 | synchronized (channelPool) { |
| 347 | channelFuture = channelPool.get(offset); |
| 348 | if (channelFuture == null || channelFuture.isCompletedExceptionally()) { |
| 349 | channelFuture = openChannel(endpoint); |
| 350 | channelPool.set(offset, channelFuture); |
| 351 | } |
| 352 | } |
| 353 | } |
| 354 | |
| 355 | CompletableFuture<Channel> future = new CompletableFuture<>(); |
| 356 | final CompletableFuture<Channel> finalFuture = channelFuture; |
| 357 | finalFuture.whenComplete((channel, error) -> { |
| 358 | if (error == null) { |
| 359 | if (!channel.isActive()) { |
| 360 | synchronized (channelPool) { |
| 361 | CompletableFuture<Channel> currentFuture = channelPool.get(offset); |
| 362 | if (currentFuture == finalFuture) { |
| 363 | channelPool.set(offset, null); |
| 364 | getChannel(endpoint, messageType).whenComplete((recursiveResult, recursiveError) -> { |
| 365 | if (recursiveError == null) { |
| 366 | future.complete(recursiveResult); |
| 367 | } else { |
| 368 | future.completeExceptionally(recursiveError); |
| 369 | } |
| 370 | }); |
| 371 | } else { |
| 372 | currentFuture.whenComplete((recursiveResult, recursiveError) -> { |
| 373 | if (recursiveError == null) { |
| 374 | future.complete(recursiveResult); |
| 375 | } else { |
| 376 | future.completeExceptionally(recursiveError); |
| 377 | } |
| 378 | }); |
| 379 | } |
| 380 | } |
| 381 | } else { |
| 382 | future.complete(channel); |
| 383 | } |
| 384 | } else { |
| 385 | future.completeExceptionally(error); |
| 386 | } |
| 387 | }); |
| 388 | return future; |
| 389 | } |
| 390 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 391 | private <T> CompletableFuture<T> executeOnPooledConnection( |
| 392 | Endpoint endpoint, |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 393 | String type, |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 394 | Function<ClientConnection, CompletableFuture<T>> callback, |
| 395 | Executor executor) { |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 396 | CompletableFuture<T> future = new CompletableFuture<T>(); |
| 397 | executeOnPooledConnection(endpoint, type, callback, executor, future); |
| 398 | return future; |
| 399 | } |
| 400 | |
| 401 | private <T> void executeOnPooledConnection( |
| 402 | Endpoint endpoint, |
| 403 | String type, |
| 404 | Function<ClientConnection, CompletableFuture<T>> callback, |
| 405 | Executor executor, |
| 406 | CompletableFuture<T> future) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 407 | if (endpoint.equals(localEndpoint)) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 408 | callback.apply(localClientConnection).whenComplete((result, error) -> { |
| 409 | if (error == null) { |
| 410 | executor.execute(() -> future.complete(result)); |
| 411 | } else { |
| 412 | executor.execute(() -> future.completeExceptionally(error)); |
| 413 | } |
| 414 | }); |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 415 | return; |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 416 | } |
| 417 | |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 418 | getChannel(endpoint, type).whenComplete((channel, channelError) -> { |
| 419 | if (channelError == null) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 420 | ClientConnection connection = clientConnections.computeIfAbsent(channel, RemoteClientConnection::new); |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 421 | callback.apply(connection).whenComplete((result, sendError) -> { |
| 422 | if (sendError == null) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 423 | executor.execute(() -> future.complete(result)); |
| 424 | } else { |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 425 | executor.execute(() -> future.completeExceptionally(sendError)); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 426 | } |
| 427 | }); |
| 428 | } else { |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 429 | executor.execute(() -> future.completeExceptionally(channelError)); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 430 | } |
Jordan Halterman | 041534b | 2017-03-21 10:37:33 -0700 | [diff] [blame] | 431 | }); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 432 | } |
| 433 | |
| 434 | @Override |
| 435 | public void registerHandler(String type, BiConsumer<Endpoint, byte[]> handler, Executor executor) { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 436 | checkPermission(CLUSTER_WRITE); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 437 | handlers.put(type, (message, connection) -> executor.execute(() -> |
| 438 | handler.accept(message.sender(), message.payload()))); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 439 | } |
| 440 | |
| 441 | @Override |
| 442 | public void registerHandler(String type, BiFunction<Endpoint, byte[], byte[]> handler, Executor executor) { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 443 | checkPermission(CLUSTER_WRITE); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 444 | handlers.put(type, (message, connection) -> executor.execute(() -> { |
Madan Jampani | a9e70a6 | 2016-03-02 16:28:18 -0800 | [diff] [blame] | 445 | byte[] responsePayload = null; |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 446 | InternalReply.Status status = InternalReply.Status.OK; |
Madan Jampani | a9e70a6 | 2016-03-02 16:28:18 -0800 | [diff] [blame] | 447 | try { |
| 448 | responsePayload = handler.apply(message.sender(), message.payload()); |
| 449 | } catch (Exception e) { |
Jordan Halterman | b6ee9e9 | 2017-06-21 15:26:28 -0700 | [diff] [blame] | 450 | log.debug("An error occurred in a message handler: {}", e); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 451 | status = InternalReply.Status.ERROR_HANDLER_EXCEPTION; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 452 | } |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 453 | connection.reply(message, status, Optional.ofNullable(responsePayload)); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 454 | })); |
| 455 | } |
| 456 | |
| 457 | @Override |
| 458 | public void registerHandler(String type, BiFunction<Endpoint, byte[], CompletableFuture<byte[]>> handler) { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 459 | checkPermission(CLUSTER_WRITE); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 460 | handlers.put(type, (message, connection) -> { |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 461 | handler.apply(message.sender(), message.payload()).whenComplete((result, error) -> { |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 462 | InternalReply.Status status; |
Jordan Halterman | b6ee9e9 | 2017-06-21 15:26:28 -0700 | [diff] [blame] | 463 | if (error == null) { |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 464 | status = InternalReply.Status.OK; |
Jordan Halterman | b6ee9e9 | 2017-06-21 15:26:28 -0700 | [diff] [blame] | 465 | } else { |
| 466 | log.debug("An error occurred in a message handler: {}", error); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 467 | status = InternalReply.Status.ERROR_HANDLER_EXCEPTION; |
Jordan Halterman | b6ee9e9 | 2017-06-21 15:26:28 -0700 | [diff] [blame] | 468 | } |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 469 | connection.reply(message, status, Optional.ofNullable(result)); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 470 | }); |
| 471 | }); |
| 472 | } |
| 473 | |
| 474 | @Override |
| 475 | public void unregisterHandler(String type) { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 476 | checkPermission(CLUSTER_WRITE); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 477 | handlers.remove(type); |
| 478 | } |
| 479 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 480 | private Bootstrap bootstrapClient(Endpoint endpoint) { |
| 481 | Bootstrap bootstrap = new Bootstrap(); |
| 482 | bootstrap.option(ChannelOption.ALLOCATOR, PooledByteBufAllocator.DEFAULT); |
| 483 | bootstrap.option(ChannelOption.WRITE_BUFFER_WATER_MARK, |
| 484 | new WriteBufferWaterMark(10 * 32 * 1024, 10 * 64 * 1024)); |
| 485 | bootstrap.option(ChannelOption.SO_SNDBUF, 1048576); |
| 486 | bootstrap.option(ChannelOption.CONNECT_TIMEOUT_MILLIS, 1000); |
| 487 | bootstrap.group(clientGroup); |
| 488 | // TODO: Make this faster: |
| 489 | // http://normanmaurer.me/presentations/2014-facebook-eng-netty/slides.html#37.0 |
| 490 | bootstrap.channel(clientChannelClass); |
| 491 | bootstrap.option(ChannelOption.SO_KEEPALIVE, true); |
| 492 | bootstrap.remoteAddress(endpoint.host().toInetAddress(), endpoint.port()); |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 493 | if (enableNettyTls) { |
| 494 | bootstrap.handler(new SslClientCommunicationChannelInitializer()); |
| 495 | } else { |
| 496 | bootstrap.handler(new BasicChannelInitializer()); |
| 497 | } |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 498 | return bootstrap; |
| 499 | } |
| 500 | |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 501 | private void startAcceptingConnections() throws InterruptedException { |
| 502 | ServerBootstrap b = new ServerBootstrap(); |
Jon Hall | 9a44d6a | 2017-03-02 18:14:37 -0800 | [diff] [blame] | 503 | b.childOption(ChannelOption.WRITE_BUFFER_WATER_MARK, |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 504 | new WriteBufferWaterMark(8 * 1024, 32 * 1024)); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 505 | b.option(ChannelOption.SO_RCVBUF, 1048576); |
| 506 | b.option(ChannelOption.TCP_NODELAY, true); |
Yuta HIGUCHI | b47c953 | 2016-08-22 09:41:23 -0700 | [diff] [blame] | 507 | b.childOption(ChannelOption.ALLOCATOR, PooledByteBufAllocator.DEFAULT); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 508 | b.group(serverGroup, clientGroup); |
| 509 | b.channel(serverChannelClass); |
| 510 | if (enableNettyTls) { |
| 511 | b.childHandler(new SslServerCommunicationChannelInitializer()); |
| 512 | } else { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 513 | b.childHandler(new BasicChannelInitializer()); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 514 | } |
| 515 | b.option(ChannelOption.SO_BACKLOG, 128); |
| 516 | b.childOption(ChannelOption.SO_KEEPALIVE, true); |
| 517 | |
| 518 | // Bind and start to accept incoming connections. |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 519 | b.bind(localEndpoint.port()).sync().addListener(future -> { |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 520 | if (future.isSuccess()) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 521 | log.info("{} accepting incoming connections on port {}", |
| 522 | localEndpoint.host(), localEndpoint.port()); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 523 | } else { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 524 | log.warn("{} failed to bind to port {} due to {}", |
| 525 | localEndpoint.host(), localEndpoint.port(), future.cause()); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 526 | } |
| 527 | }); |
| 528 | } |
| 529 | |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 530 | private CompletableFuture<Channel> openChannel(Endpoint ep) { |
| 531 | Bootstrap bootstrap = bootstrapClient(ep); |
| 532 | CompletableFuture<Channel> retFuture = new CompletableFuture<>(); |
| 533 | ChannelFuture f = bootstrap.connect(); |
| 534 | |
| 535 | f.addListener(future -> { |
| 536 | if (future.isSuccess()) { |
| 537 | retFuture.complete(f.channel()); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 538 | } else { |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 539 | retFuture.completeExceptionally(future.cause()); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 540 | } |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 541 | }); |
| 542 | log.debug("Established a new connection to {}", ep); |
| 543 | return retFuture; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 544 | } |
| 545 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 546 | /** |
| 547 | * Channel initializer for TLS servers. |
| 548 | */ |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 549 | private class SslServerCommunicationChannelInitializer extends ChannelInitializer<SocketChannel> { |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 550 | private final ChannelHandler dispatcher = new InboundMessageDispatcher(); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 551 | |
| 552 | @Override |
| 553 | protected void initChannel(SocketChannel channel) throws Exception { |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 554 | SSLContext serverContext = SSLContext.getInstance("TLS"); |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 555 | serverContext.init(keyManager.getKeyManagers(), trustManager.getTrustManagers(), null); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 556 | |
| 557 | SSLEngine serverSslEngine = serverContext.createSSLEngine(); |
| 558 | |
| 559 | serverSslEngine.setNeedClientAuth(true); |
| 560 | serverSslEngine.setUseClientMode(false); |
| 561 | serverSslEngine.setEnabledProtocols(serverSslEngine.getSupportedProtocols()); |
| 562 | serverSslEngine.setEnabledCipherSuites(serverSslEngine.getSupportedCipherSuites()); |
| 563 | serverSslEngine.setEnableSessionCreation(true); |
| 564 | |
| 565 | channel.pipeline().addLast("ssl", new io.netty.handler.ssl.SslHandler(serverSslEngine)) |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 566 | .addLast("encoder", new MessageEncoder(localEndpoint, preamble)) |
Madan Jampani | b825aeb | 2016-04-01 15:18:25 -0700 | [diff] [blame] | 567 | .addLast("decoder", new MessageDecoder()) |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 568 | .addLast("handler", dispatcher); |
| 569 | } |
| 570 | } |
| 571 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 572 | /** |
| 573 | * Channel initializer for TLS clients. |
| 574 | */ |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 575 | private class SslClientCommunicationChannelInitializer extends ChannelInitializer<SocketChannel> { |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 576 | private final ChannelHandler dispatcher = new InboundMessageDispatcher(); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 577 | |
| 578 | @Override |
| 579 | protected void initChannel(SocketChannel channel) throws Exception { |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 580 | SSLContext clientContext = SSLContext.getInstance("TLS"); |
Brian O'Connor | 740e98c | 2017-06-29 17:07:17 -0700 | [diff] [blame] | 581 | clientContext.init(keyManager.getKeyManagers(), trustManager.getTrustManagers(), null); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 582 | |
| 583 | SSLEngine clientSslEngine = clientContext.createSSLEngine(); |
| 584 | |
| 585 | clientSslEngine.setUseClientMode(true); |
| 586 | clientSslEngine.setEnabledProtocols(clientSslEngine.getSupportedProtocols()); |
| 587 | clientSslEngine.setEnabledCipherSuites(clientSslEngine.getSupportedCipherSuites()); |
| 588 | clientSslEngine.setEnableSessionCreation(true); |
| 589 | |
| 590 | channel.pipeline().addLast("ssl", new io.netty.handler.ssl.SslHandler(clientSslEngine)) |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 591 | .addLast("encoder", new MessageEncoder(localEndpoint, preamble)) |
Madan Jampani | b825aeb | 2016-04-01 15:18:25 -0700 | [diff] [blame] | 592 | .addLast("decoder", new MessageDecoder()) |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 593 | .addLast("handler", dispatcher); |
| 594 | } |
| 595 | } |
| 596 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 597 | /** |
| 598 | * Channel initializer for basic connections. |
| 599 | */ |
| 600 | private class BasicChannelInitializer extends ChannelInitializer<SocketChannel> { |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 601 | private final ChannelHandler dispatcher = new InboundMessageDispatcher(); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 602 | |
| 603 | @Override |
| 604 | protected void initChannel(SocketChannel channel) throws Exception { |
| 605 | channel.pipeline() |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 606 | .addLast("encoder", new MessageEncoder(localEndpoint, preamble)) |
Madan Jampani | b825aeb | 2016-04-01 15:18:25 -0700 | [diff] [blame] | 607 | .addLast("decoder", new MessageDecoder()) |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 608 | .addLast("handler", dispatcher); |
| 609 | } |
| 610 | } |
| 611 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 612 | /** |
| 613 | * Channel inbound handler that dispatches messages to the appropriate handler. |
| 614 | */ |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 615 | @ChannelHandler.Sharable |
Yuta HIGUCHI | c012dda | 2016-08-17 00:43:46 -0700 | [diff] [blame] | 616 | private class InboundMessageDispatcher extends SimpleChannelInboundHandler<Object> { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 617 | // Effectively SimpleChannelInboundHandler<InternalMessage>, |
| 618 | // had to specify <Object> to avoid Class Loader not being able to find some classes. |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 619 | |
| 620 | @Override |
Yuta HIGUCHI | c012dda | 2016-08-17 00:43:46 -0700 | [diff] [blame] | 621 | protected void channelRead0(ChannelHandlerContext ctx, Object rawMessage) throws Exception { |
| 622 | InternalMessage message = (InternalMessage) rawMessage; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 623 | try { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 624 | if (message.isRequest()) { |
| 625 | RemoteServerConnection connection = |
| 626 | serverConnections.computeIfAbsent(ctx.channel(), RemoteServerConnection::new); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 627 | connection.dispatch((InternalRequest) message); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 628 | } else { |
| 629 | RemoteClientConnection connection = |
| 630 | clientConnections.computeIfAbsent(ctx.channel(), RemoteClientConnection::new); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 631 | connection.dispatch((InternalReply) message); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 632 | } |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 633 | } catch (RejectedExecutionException e) { |
| 634 | log.warn("Unable to dispatch message due to {}", e.getMessage()); |
| 635 | } |
| 636 | } |
| 637 | |
| 638 | @Override |
| 639 | public void exceptionCaught(ChannelHandlerContext context, Throwable cause) { |
| 640 | log.error("Exception inside channel handling pipeline.", cause); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 641 | |
| 642 | RemoteClientConnection clientConnection = clientConnections.remove(context.channel()); |
| 643 | if (clientConnection != null) { |
| 644 | clientConnection.close(); |
| 645 | } |
| 646 | |
| 647 | RemoteServerConnection serverConnection = serverConnections.remove(context.channel()); |
| 648 | if (serverConnection != null) { |
| 649 | serverConnection.close(); |
| 650 | } |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 651 | context.close(); |
| 652 | } |
Yuta HIGUCHI | c012dda | 2016-08-17 00:43:46 -0700 | [diff] [blame] | 653 | |
| 654 | /** |
| 655 | * Returns true if the given message should be handled. |
| 656 | * |
| 657 | * @param msg inbound message |
| 658 | * @return true if {@code msg} is {@link InternalMessage} instance. |
| 659 | * |
| 660 | * @see SimpleChannelInboundHandler#acceptInboundMessage(Object) |
| 661 | */ |
| 662 | @Override |
| 663 | public final boolean acceptInboundMessage(Object msg) { |
| 664 | return msg instanceof InternalMessage; |
| 665 | } |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 666 | } |
Yuta HIGUCHI | c012dda | 2016-08-17 00:43:46 -0700 | [diff] [blame] | 667 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 668 | /** |
| 669 | * Wraps a {@link CompletableFuture} and tracks its type and creation time. |
| 670 | */ |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 671 | private final class Callback { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 672 | private final String type; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 673 | private final CompletableFuture<byte[]> future; |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 674 | private final long time = System.currentTimeMillis(); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 675 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 676 | Callback(String type, CompletableFuture<byte[]> future) { |
| 677 | this.type = type; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 678 | this.future = future; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 679 | } |
| 680 | |
| 681 | public void complete(byte[] value) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 682 | future.complete(value); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 683 | } |
| 684 | |
| 685 | public void completeExceptionally(Throwable error) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 686 | future.completeExceptionally(error); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 687 | } |
| 688 | } |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 689 | |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 690 | /** |
| 691 | * Represents the client side of a connection to a local or remote server. |
| 692 | */ |
| 693 | private interface ClientConnection { |
| 694 | |
| 695 | /** |
| 696 | * Sends a message to the other side of the connection. |
| 697 | * |
| 698 | * @param message the message to send |
| 699 | * @return a completable future to be completed once the message has been sent |
| 700 | */ |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 701 | CompletableFuture<Void> sendAsync(InternalRequest message); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 702 | |
| 703 | /** |
| 704 | * Sends a message to the other side of the connection, awaiting a reply. |
| 705 | * |
| 706 | * @param message the message to send |
| 707 | * @return a completable future to be completed once a reply is received or the request times out |
| 708 | */ |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 709 | CompletableFuture<byte[]> sendAndReceive(InternalRequest message); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 710 | |
| 711 | /** |
| 712 | * Closes the connection. |
| 713 | */ |
| 714 | default void close() { |
| 715 | } |
| 716 | } |
| 717 | |
| 718 | /** |
| 719 | * Represents the server side of a connection. |
| 720 | */ |
| 721 | private interface ServerConnection { |
| 722 | |
| 723 | /** |
| 724 | * Sends a reply to the other side of the connection. |
| 725 | * |
| 726 | * @param message the message to which to reply |
| 727 | * @param status the reply status |
| 728 | * @param payload the response payload |
| 729 | */ |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 730 | void reply(InternalRequest message, InternalReply.Status status, Optional<byte[]> payload); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 731 | |
| 732 | /** |
| 733 | * Closes the connection. |
| 734 | */ |
| 735 | default void close() { |
| 736 | } |
| 737 | } |
| 738 | |
| 739 | /** |
| 740 | * Local connection implementation. |
| 741 | */ |
| 742 | private final class LocalClientConnection implements ClientConnection { |
| 743 | @Override |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 744 | public CompletableFuture<Void> sendAsync(InternalRequest message) { |
| 745 | BiConsumer<InternalRequest, ServerConnection> handler = handlers.get(message.subject()); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 746 | if (handler != null) { |
| 747 | handler.accept(message, localServerConnection); |
| 748 | } else { |
| 749 | log.debug("No handler for message type {} from {}", message.type(), message.sender()); |
| 750 | } |
| 751 | return CompletableFuture.completedFuture(null); |
| 752 | } |
| 753 | |
| 754 | @Override |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 755 | public CompletableFuture<byte[]> sendAndReceive(InternalRequest message) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 756 | CompletableFuture<byte[]> future = new CompletableFuture<>(); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 757 | BiConsumer<InternalRequest, ServerConnection> handler = handlers.get(message.subject()); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 758 | if (handler != null) { |
| 759 | handler.accept(message, new LocalServerConnection(future)); |
| 760 | } else { |
| 761 | log.debug("No handler for message type {} from {}", message.type(), message.sender()); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 762 | new LocalServerConnection(future) |
| 763 | .reply(message, InternalReply.Status.ERROR_NO_HANDLER, Optional.empty()); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 764 | } |
| 765 | return future; |
| 766 | } |
| 767 | } |
| 768 | |
| 769 | /** |
| 770 | * Local server connection. |
| 771 | */ |
| 772 | private final class LocalServerConnection implements ServerConnection { |
| 773 | private final CompletableFuture<byte[]> future; |
| 774 | |
| 775 | LocalServerConnection(CompletableFuture<byte[]> future) { |
| 776 | this.future = future; |
| 777 | } |
| 778 | |
| 779 | @Override |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 780 | public void reply(InternalRequest message, InternalReply.Status status, Optional<byte[]> payload) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 781 | if (future != null) { |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 782 | if (status == InternalReply.Status.OK) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 783 | future.complete(payload.orElse(EMPTY_PAYLOAD)); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 784 | } else if (status == InternalReply.Status.ERROR_NO_HANDLER) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 785 | future.completeExceptionally(new MessagingException.NoRemoteHandler()); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 786 | } else if (status == InternalReply.Status.ERROR_HANDLER_EXCEPTION) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 787 | future.completeExceptionally(new MessagingException.RemoteHandlerFailure()); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 788 | } else if (status == InternalReply.Status.PROTOCOL_EXCEPTION) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 789 | future.completeExceptionally(new MessagingException.ProtocolException()); |
| 790 | } |
| 791 | } |
| 792 | } |
| 793 | } |
| 794 | |
| 795 | /** |
| 796 | * Remote connection implementation. |
| 797 | */ |
| 798 | private final class RemoteClientConnection implements ClientConnection { |
| 799 | private final Channel channel; |
| 800 | private final Map<Long, Callback> futures = Maps.newConcurrentMap(); |
| 801 | private final AtomicBoolean closed = new AtomicBoolean(false); |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 802 | private final Cache<String, TimeoutHistory> timeoutHistories = CacheBuilder.newBuilder() |
| 803 | .expireAfterAccess(HISTORY_EXPIRE_MILLIS, TimeUnit.MILLISECONDS) |
| 804 | .build(); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 805 | |
| 806 | RemoteClientConnection(Channel channel) { |
| 807 | this.channel = channel; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 808 | } |
| 809 | |
| 810 | /** |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 811 | * Times out callbacks for this connection. |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 812 | */ |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 813 | private void timeoutCallbacks() { |
| 814 | // Store the current time. |
| 815 | long currentTime = System.currentTimeMillis(); |
| 816 | |
| 817 | // Iterate through future callbacks and time out callbacks that have been alive |
| 818 | // longer than the current timeout according to the message type. |
| 819 | Iterator<Map.Entry<Long, Callback>> iterator = futures.entrySet().iterator(); |
| 820 | while (iterator.hasNext()) { |
| 821 | Callback callback = iterator.next().getValue(); |
| 822 | try { |
| 823 | TimeoutHistory timeoutHistory = timeoutHistories.get(callback.type, TimeoutHistory::new); |
| 824 | long currentTimeout = timeoutHistory.currentTimeout; |
| 825 | if (currentTime - callback.time > currentTimeout) { |
| 826 | iterator.remove(); |
| 827 | long elapsedTime = currentTime - callback.time; |
| 828 | timeoutHistory.addReplyTime(elapsedTime); |
| 829 | callback.completeExceptionally( |
| 830 | new TimeoutException("Request timed out in " + elapsedTime + " milliseconds")); |
| 831 | } |
| 832 | } catch (ExecutionException e) { |
| 833 | throw new AssertionError(); |
| 834 | } |
| 835 | } |
Jordan Halterman | 66e6e3b | 2017-07-10 11:26:34 -0700 | [diff] [blame] | 836 | |
| 837 | // Iterate through all timeout histories and recompute the timeout. |
| 838 | for (TimeoutHistory timeoutHistory : timeoutHistories.asMap().values()) { |
| 839 | timeoutHistory.recomputeTimeoutMillis(); |
| 840 | } |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 841 | } |
| 842 | |
| 843 | @Override |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 844 | public CompletableFuture<Void> sendAsync(InternalRequest message) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 845 | CompletableFuture<Void> future = new CompletableFuture<>(); |
| 846 | channel.writeAndFlush(message).addListener(channelFuture -> { |
| 847 | if (!channelFuture.isSuccess()) { |
| 848 | future.completeExceptionally(channelFuture.cause()); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 849 | } else { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 850 | future.complete(null); |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 851 | } |
| 852 | }); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 853 | return future; |
| 854 | } |
| 855 | |
| 856 | @Override |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 857 | public CompletableFuture<byte[]> sendAndReceive(InternalRequest message) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 858 | CompletableFuture<byte[]> future = new CompletableFuture<>(); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 859 | Callback callback = new Callback(message.subject(), future); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 860 | futures.put(message.id(), callback); |
| 861 | channel.writeAndFlush(message).addListener(channelFuture -> { |
| 862 | if (!channelFuture.isSuccess()) { |
| 863 | futures.remove(message.id()); |
| 864 | callback.completeExceptionally(channelFuture.cause()); |
| 865 | } |
| 866 | }); |
| 867 | return future; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 868 | } |
| 869 | |
| 870 | /** |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 871 | * Dispatches a message to a local handler. |
| 872 | * |
| 873 | * @param message the message to dispatch |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 874 | */ |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 875 | private void dispatch(InternalReply message) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 876 | if (message.preamble() != preamble) { |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 877 | log.debug("Received {} with invalid preamble", message.type()); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 878 | return; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 879 | } |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 880 | |
| 881 | clockService.recordEventTime(message.time()); |
| 882 | |
| 883 | Callback callback = futures.remove(message.id()); |
| 884 | if (callback != null) { |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 885 | if (message.status() == InternalReply.Status.OK) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 886 | callback.complete(message.payload()); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 887 | } else if (message.status() == InternalReply.Status.ERROR_NO_HANDLER) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 888 | callback.completeExceptionally(new MessagingException.NoRemoteHandler()); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 889 | } else if (message.status() == InternalReply.Status.ERROR_HANDLER_EXCEPTION) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 890 | callback.completeExceptionally(new MessagingException.RemoteHandlerFailure()); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 891 | } else if (message.status() == InternalReply.Status.PROTOCOL_EXCEPTION) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 892 | callback.completeExceptionally(new MessagingException.ProtocolException()); |
| 893 | } |
| 894 | |
| 895 | try { |
| 896 | TimeoutHistory timeoutHistory = timeoutHistories.get(callback.type, TimeoutHistory::new); |
| 897 | timeoutHistory.addReplyTime(System.currentTimeMillis() - callback.time); |
| 898 | } catch (ExecutionException e) { |
| 899 | throw new AssertionError(); |
| 900 | } |
| 901 | } else { |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 902 | log.debug("Received a reply for message id:[{}] " |
| 903 | + "but was unable to locate the" |
| 904 | + " request handle", message.id()); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 905 | } |
| 906 | } |
| 907 | |
| 908 | @Override |
| 909 | public void close() { |
| 910 | if (closed.compareAndSet(false, true)) { |
| 911 | timeoutFuture.cancel(false); |
| 912 | for (Callback callback : futures.values()) { |
| 913 | callback.completeExceptionally(new ConnectException()); |
| 914 | } |
| 915 | } |
| 916 | } |
| 917 | } |
| 918 | |
| 919 | /** |
| 920 | * Remote server connection. |
| 921 | */ |
| 922 | private final class RemoteServerConnection implements ServerConnection { |
| 923 | private final Channel channel; |
| 924 | |
| 925 | RemoteServerConnection(Channel channel) { |
| 926 | this.channel = channel; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 927 | } |
| 928 | |
| 929 | /** |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 930 | * Dispatches a message to a local handler. |
| 931 | * |
| 932 | * @param message the message to dispatch |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 933 | */ |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 934 | private void dispatch(InternalRequest message) { |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 935 | if (message.preamble() != preamble) { |
| 936 | log.debug("Received {} with invalid preamble from {}", message.type(), message.sender()); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 937 | reply(message, InternalReply.Status.PROTOCOL_EXCEPTION, Optional.empty()); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 938 | return; |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 939 | } |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 940 | |
| 941 | clockService.recordEventTime(message.time()); |
| 942 | |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 943 | BiConsumer<InternalRequest, ServerConnection> handler = handlers.get(message.subject()); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 944 | if (handler != null) { |
| 945 | handler.accept(message, this); |
| 946 | } else { |
| 947 | log.debug("No handler for message type {} from {}", message.type(), message.sender()); |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 948 | reply(message, InternalReply.Status.ERROR_NO_HANDLER, Optional.empty()); |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 949 | } |
| 950 | } |
| 951 | |
| 952 | @Override |
Jordan Halterman | e3813a9 | 2017-07-29 14:10:31 -0700 | [diff] [blame] | 953 | public void reply(InternalRequest message, InternalReply.Status status, Optional<byte[]> payload) { |
| 954 | InternalReply response = new InternalReply(preamble, |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 955 | clockService.timeNow(), |
| 956 | message.id(), |
Jordan Halterman | f7c7f6f | 2017-05-05 03:02:34 -0700 | [diff] [blame] | 957 | payload.orElse(EMPTY_PAYLOAD), |
| 958 | status); |
| 959 | channel.writeAndFlush(response); |
| 960 | } |
| 961 | } |
| 962 | |
| 963 | /** |
| 964 | * Request-reply timeout history tracker. |
| 965 | */ |
| 966 | private static final class TimeoutHistory { |
| 967 | private final DescriptiveStatistics timeoutHistory = new SynchronizedDescriptiveStatistics(WINDOW_SIZE); |
| 968 | private final AtomicLong maxReplyTime = new AtomicLong(); |
| 969 | private volatile long currentTimeout = DEFAULT_TIMEOUT_MILLIS; |
| 970 | |
| 971 | /** |
| 972 | * Adds a reply time to the history. |
| 973 | * |
| 974 | * @param replyTime the reply time to add to the history |
| 975 | */ |
| 976 | void addReplyTime(long replyTime) { |
| 977 | maxReplyTime.getAndAccumulate(replyTime, Math::max); |
| 978 | } |
| 979 | |
| 980 | /** |
| 981 | * Computes the current timeout. |
| 982 | */ |
| 983 | private void recomputeTimeoutMillis() { |
| 984 | double nextTimeout = maxReplyTime.getAndSet(0) * TIMEOUT_MULTIPLIER; |
| 985 | timeoutHistory.addValue( |
| 986 | Math.min(Math.max(nextTimeout, MIN_TIMEOUT_MILLIS), MAX_TIMEOUT_MILLIS)); |
| 987 | if (timeoutHistory.getN() == WINDOW_SIZE) { |
| 988 | this.currentTimeout = (long) timeoutHistory.getMax(); |
| 989 | } |
Aaron Kruglikov | 1b72738 | 2016-02-09 16:17:47 -0800 | [diff] [blame] | 990 | } |
| 991 | } |
JunHuy Lam | 39eb429 | 2015-06-26 17:24:23 +0900 | [diff] [blame] | 992 | } |