Add retry after exceptions during a cluster topology change event callback

This commit is contained in:
Chris Eager
2023-03-23 09:50:15 -05:00
committed by Chris Eager
parent 0cc84131de
commit 3ccfeb490b
13 changed files with 323 additions and 132 deletions

View File

@@ -515,8 +515,8 @@ public class WhisperServerService extends Application<WhisperServerConfiguration
StoredVerificationCodeManager pendingAccountsManager = new StoredVerificationCodeManager(pendingAccounts);
StoredVerificationCodeManager pendingDevicesManager = new StoredVerificationCodeManager(pendingDevices);
ProfilesManager profilesManager = new ProfilesManager(profiles, cacheCluster);
MessagesCache messagesCache = new MessagesCache(messagesCluster, messagesCluster, Clock.systemUTC(),
keyspaceNotificationDispatchExecutor, messageDeliveryScheduler, messageDeletionAsyncExecutor);
MessagesCache messagesCache = new MessagesCache(messagesCluster, messagesCluster,
keyspaceNotificationDispatchExecutor, messageDeliveryScheduler, messageDeletionAsyncExecutor, clock);
PushLatencyManager pushLatencyManager = new PushLatencyManager(metricsCluster, dynamicConfigurationManager);
ReportMessageManager reportMessageManager = new ReportMessageManager(reportMessageDynamoDb, rateLimitersCluster,
config.getReportMessageConfiguration().getCounterTtl());

View File

@@ -1,5 +1,5 @@
/*
* Copyright 2013-2020 Signal Messenger, LLC
* Copyright 2013 Signal Messenger, LLC
* SPDX-License-Identifier: AGPL-3.0-only
*/
@@ -18,7 +18,6 @@ import io.lettuce.core.RedisFuture;
import io.lettuce.core.ScriptOutputType;
import io.lettuce.core.cluster.SlotHash;
import io.lettuce.core.cluster.api.sync.RedisAdvancedClusterCommands;
import io.lettuce.core.cluster.event.ClusterTopologyChangedEvent;
import io.lettuce.core.cluster.models.partitions.RedisClusterNode;
import io.lettuce.core.cluster.pubsub.RedisClusterPubSubAdapter;
import java.io.IOException;
@@ -86,8 +85,10 @@ public class ClientPresenceManager extends RedisClusterPubSubAdapter<String, Str
final ExecutorService keyspaceNotificationExecutorService) throws IOException {
this.presenceCluster = presenceCluster;
this.pubSubConnection = this.presenceCluster.createPubSubConnection();
this.clearPresenceScript = ClusterLuaScript.fromResource(presenceCluster, "lua/clear_presence.lua", ScriptOutputType.INTEGER);
this.renewPresenceScript = ClusterLuaScript.fromResource(presenceCluster, "lua/renew_presence.lua", ScriptOutputType.VALUE);
this.clearPresenceScript = ClusterLuaScript.fromResource(presenceCluster, "lua/clear_presence.lua",
ScriptOutputType.INTEGER);
this.renewPresenceScript = ClusterLuaScript.fromResource(presenceCluster, "lua/renew_presence.lua",
ScriptOutputType.VALUE);
this.scheduledExecutorService = scheduledExecutorService;
this.keyspaceNotificationExecutorService = keyspaceNotificationExecutorService;
@@ -112,9 +113,6 @@ public class ClientPresenceManager extends RedisClusterPubSubAdapter<String, Str
public void start() {
pubSubConnection.usePubSubConnection(connection -> {
connection.addListener(this);
connection.getResources().eventBus().get()
.filter(event -> event instanceof ClusterTopologyChangedEvent)
.subscribe(event -> resubscribeAll());
final String presenceChannel = getManagerPresenceChannel(managerId);
final int slot = SlotHash.getSlot(presenceChannel);
@@ -124,6 +122,8 @@ public class ClientPresenceManager extends RedisClusterPubSubAdapter<String, Str
.subscribe(presenceChannel);
});
pubSubConnection.subscribeToClusterTopologyChangedEvents(this::resubscribeAll);
presenceCluster.useCluster(connection -> connection.sync().sadd(MANAGER_SET_KEY, managerId));
pruneMissingPeersFuture = scheduledExecutorService.scheduleWithFixedDelay(() -> {

View File

@@ -10,28 +10,41 @@ import static org.whispersystems.textsecuregcm.metrics.MetricsUtil.name;
import io.github.resilience4j.circuitbreaker.CircuitBreaker;
import io.github.resilience4j.retry.Retry;
import io.lettuce.core.RedisException;
import io.lettuce.core.cluster.event.ClusterTopologyChangedEvent;
import io.lettuce.core.cluster.pubsub.StatefulRedisClusterPubSubConnection;
import io.micrometer.core.instrument.Metrics;
import io.micrometer.core.instrument.Timer;
import java.util.function.Consumer;
import java.util.function.Function;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.whispersystems.textsecuregcm.util.CircuitBreakerUtil;
import reactor.core.scheduler.Scheduler;
public class FaultTolerantPubSubConnection<K, V> {
private static final Logger logger = LoggerFactory.getLogger(FaultTolerantPubSubConnection.class);
private final String name;
private final StatefulRedisClusterPubSubConnection<K, V> pubSubConnection;
private final CircuitBreaker circuitBreaker;
private final Retry retry;
private final Retry resubscribeRetry;
private final Scheduler topologyChangedEventScheduler;
private final Timer executeTimer;
public FaultTolerantPubSubConnection(final String name,
final StatefulRedisClusterPubSubConnection<K, V> pubSubConnection, final CircuitBreaker circuitBreaker,
final Retry retry) {
final Retry retry, final Retry resubscribeRetry, final Scheduler topologyChangedEventScheduler) {
this.name = name;
this.pubSubConnection = pubSubConnection;
this.circuitBreaker = circuitBreaker;
this.retry = retry;
this.resubscribeRetry = resubscribeRetry;
this.topologyChangedEventScheduler = topologyChangedEventScheduler;
this.pubSubConnection.setNodeMessagePropagation(true);
@@ -58,11 +71,32 @@ public class FaultTolerantPubSubConnection<K, V> {
return circuitBreaker.executeCheckedSupplier(
() -> retry.executeCallable(() -> executeTimer.record(() -> function.apply(pubSubConnection))));
} catch (final Throwable t) {
if (t instanceof RedisException) {
throw (RedisException) t;
} else {
throw new RedisException(t);
}
if (t instanceof RedisException) {
throw (RedisException) t;
} else {
throw new RedisException(t);
}
}
}
public void subscribeToClusterTopologyChangedEvents(final Runnable eventHandler) {
usePubSubConnection(connection -> connection.getResources().eventBus().get()
.filter(event -> event instanceof ClusterTopologyChangedEvent)
.subscribeOn(topologyChangedEventScheduler)
.subscribe(event -> {
logger.info("Got topology change event for {}, resubscribing all keyspace notifications", name);
resubscribeRetry.executeRunnable(() -> {
try {
eventHandler.run();
} catch (final RuntimeException e) {
logger.warn("Resubscribe for {} failed", name, e);
throw e;
}
});
}));
}
}

View File

@@ -7,9 +7,11 @@ package org.whispersystems.textsecuregcm.redis;
import com.google.common.annotations.VisibleForTesting;
import io.github.resilience4j.circuitbreaker.CircuitBreaker;
import io.github.resilience4j.core.IntervalFunction;
import io.github.resilience4j.reactor.circuitbreaker.operator.CircuitBreakerOperator;
import io.github.resilience4j.reactor.retry.RetryOperator;
import io.github.resilience4j.retry.Retry;
import io.github.resilience4j.retry.RetryConfig;
import io.lettuce.core.ClientOptions.DisconnectedBehavior;
import io.lettuce.core.RedisCommandTimeoutException;
import io.lettuce.core.RedisException;
@@ -31,6 +33,7 @@ import org.whispersystems.textsecuregcm.configuration.RedisClusterConfiguration;
import org.whispersystems.textsecuregcm.configuration.RetryConfiguration;
import org.whispersystems.textsecuregcm.util.CircuitBreakerUtil;
import reactor.core.publisher.Flux;
import reactor.core.scheduler.Schedulers;
/**
* A fault-tolerant access manager for a Redis cluster. A fault-tolerant Redis cluster provides managed,
@@ -38,51 +41,61 @@ import reactor.core.publisher.Flux;
*/
public class FaultTolerantRedisCluster {
private final String name;
private final String name;
private final RedisClusterClient clusterClient;
private final RedisClusterClient clusterClient;
private final StatefulRedisClusterConnection<String, String> stringConnection;
private final StatefulRedisClusterConnection<byte[], byte[]> binaryConnection;
private final StatefulRedisClusterConnection<String, String> stringConnection;
private final StatefulRedisClusterConnection<byte[], byte[]> binaryConnection;
private final List<StatefulRedisClusterPubSubConnection<?, ?>> pubSubConnections = new ArrayList<>();
private final List<StatefulRedisClusterPubSubConnection<?, ?>> pubSubConnections = new ArrayList<>();
private final CircuitBreaker circuitBreaker;
private final Retry retry;
private final CircuitBreaker circuitBreaker;
private final Retry retry;
private final Retry topologyChangedEventRetry;
public FaultTolerantRedisCluster(final String name, final RedisClusterConfiguration clusterConfiguration, final ClientResources clientResources) {
this(name,
RedisClusterClient.create(clientResources, clusterConfiguration.getConfigurationUri()),
clusterConfiguration.getTimeout(),
clusterConfiguration.getCircuitBreakerConfiguration(),
clusterConfiguration.getRetryConfiguration());
}
public FaultTolerantRedisCluster(final String name, final RedisClusterConfiguration clusterConfiguration,
final ClientResources clientResources) {
this(name,
RedisClusterClient.create(clientResources, clusterConfiguration.getConfigurationUri()),
clusterConfiguration.getTimeout(),
clusterConfiguration.getCircuitBreakerConfiguration(),
clusterConfiguration.getRetryConfiguration());
}
@VisibleForTesting
FaultTolerantRedisCluster(final String name, final RedisClusterClient clusterClient, final Duration commandTimeout, final CircuitBreakerConfiguration circuitBreakerConfiguration, final RetryConfiguration retryConfiguration) {
this.name = name;
@VisibleForTesting
FaultTolerantRedisCluster(final String name, final RedisClusterClient clusterClient, final Duration commandTimeout,
final CircuitBreakerConfiguration circuitBreakerConfiguration, final RetryConfiguration retryConfiguration) {
this.name = name;
this.clusterClient = clusterClient;
this.clusterClient.setDefaultTimeout(commandTimeout);
this.clusterClient.setOptions(ClusterClientOptions.builder()
.disconnectedBehavior(DisconnectedBehavior.REJECT_COMMANDS)
.validateClusterNodeMembership(false)
.topologyRefreshOptions(ClusterTopologyRefreshOptions.builder()
.enableAllAdaptiveRefreshTriggers()
.build())
.publishOnScheduler(true)
.build());
this.clusterClient = clusterClient;
this.clusterClient.setDefaultTimeout(commandTimeout);
this.clusterClient.setOptions(ClusterClientOptions.builder()
.disconnectedBehavior(DisconnectedBehavior.REJECT_COMMANDS)
.validateClusterNodeMembership(false)
.topologyRefreshOptions(ClusterTopologyRefreshOptions.builder()
.enableAllAdaptiveRefreshTriggers()
.build())
.publishOnScheduler(true)
.build());
this.stringConnection = clusterClient.connect();
this.binaryConnection = clusterClient.connect(ByteArrayCodec.INSTANCE);
this.stringConnection = clusterClient.connect();
this.binaryConnection = clusterClient.connect(ByteArrayCodec.INSTANCE);
this.circuitBreaker = CircuitBreaker.of(name + "-breaker", circuitBreakerConfiguration.toCircuitBreakerConfig());
this.retry = Retry.of(name + "-retry", retryConfiguration.toRetryConfigBuilder()
.retryOnException(exception -> exception instanceof RedisCommandTimeoutException).build());
this.circuitBreaker = CircuitBreaker.of(name + "-breaker", circuitBreakerConfiguration.toCircuitBreakerConfig());
this.retry = Retry.of(name + "-retry", retryConfiguration.toRetryConfigBuilder()
.retryOnException(exception -> exception instanceof RedisCommandTimeoutException).build());
final RetryConfig topologyChangedEventRetryConfig = RetryConfig.custom()
.maxAttempts(Integer.MAX_VALUE)
.intervalFunction(
IntervalFunction.ofExponentialRandomBackoff(Duration.ofSeconds(1), 1.5, Duration.ofSeconds(30)))
.build();
this.topologyChangedEventRetry = Retry.of(name + "-topologyChangedRetry", topologyChangedEventRetryConfig);
CircuitBreakerUtil.registerMetrics(circuitBreaker, FaultTolerantRedisCluster.class);
CircuitBreakerUtil.registerMetrics(retry, FaultTolerantRedisCluster.class);
}
}
void shutdown() {
stringConnection.close();
@@ -158,6 +171,7 @@ public class FaultTolerantRedisCluster {
final StatefulRedisClusterPubSubConnection<String, String> pubSubConnection = clusterClient.connectPubSub();
pubSubConnections.add(pubSubConnection);
return new FaultTolerantPubSubConnection<>(name, pubSubConnection, circuitBreaker, retry);
return new FaultTolerantPubSubConnection<>(name, pubSubConnection, circuitBreaker, retry, topologyChangedEventRetry,
Schedulers.newSingle(name + "-redisPubSubEvents"));
}
}

View File

@@ -14,7 +14,6 @@ import io.lettuce.core.ScoredValue;
import io.lettuce.core.ScriptOutputType;
import io.lettuce.core.ZAddArgs;
import io.lettuce.core.cluster.SlotHash;
import io.lettuce.core.cluster.event.ClusterTopologyChangedEvent;
import io.lettuce.core.cluster.models.partitions.RedisClusterNode;
import io.lettuce.core.cluster.pubsub.RedisClusterPubSubAdapter;
import io.micrometer.core.instrument.Counter;
@@ -104,8 +103,8 @@ public class MessagesCache extends RedisClusterPubSubAdapter<String, String> imp
private static final Logger logger = LoggerFactory.getLogger(MessagesCache.class);
public MessagesCache(final FaultTolerantRedisCluster insertCluster, final FaultTolerantRedisCluster readDeleteCluster,
final Clock clock, final ExecutorService notificationExecutorService, final Scheduler messageDeliveryScheduler,
final ExecutorService messageDeletionExecutorService) throws IOException {
final ExecutorService notificationExecutorService, final Scheduler messageDeliveryScheduler,
final ExecutorService messageDeletionExecutorService, final Clock clock) throws IOException {
this.readDeleteCluster = readDeleteCluster;
this.pubSubConnection = readDeleteCluster.createPubSubConnection();
@@ -128,12 +127,8 @@ public class MessagesCache extends RedisClusterPubSubAdapter<String, String> imp
@Override
public void start() {
pubSubConnection.usePubSubConnection(connection -> {
connection.addListener(this);
connection.getResources().eventBus().get()
.filter(event -> event instanceof ClusterTopologyChangedEvent)
.subscribe(event -> resubscribeAll());
});
pubSubConnection.usePubSubConnection(connection -> connection.addListener(this));
pubSubConnection.subscribeToClusterTopologyChangedEvents(this::resubscribeAll);
}
@Override
@@ -142,7 +137,6 @@ public class MessagesCache extends RedisClusterPubSubAdapter<String, String> imp
}
private void resubscribeAll() {
logger.info("Got topology change event, resubscribing all keyspace notifications");
final Set<String> queueNames;

View File

@@ -201,7 +201,7 @@ public class AssignUsernameCommand extends EnvironmentCommand<WhisperServerConfi
ClientPresenceManager clientPresenceManager = new ClientPresenceManager(clientPresenceCluster,
Executors.newSingleThreadScheduledExecutor(), keyspaceNotificationDispatchExecutor);
MessagesCache messagesCache = new MessagesCache(messageInsertCacheCluster, messageReadDeleteCluster,
Clock.systemUTC(), keyspaceNotificationDispatchExecutor, messageDeliveryScheduler, messageDeletionExecutor);
keyspaceNotificationDispatchExecutor, messageDeliveryScheduler, messageDeletionExecutor, Clock.systemUTC());
DirectoryQueue directoryQueue = new DirectoryQueue(
configuration.getDirectoryConfiguration().getSqsConfiguration());
ProfilesManager profilesManager = new ProfilesManager(profiles, cacheCluster);

View File

@@ -180,7 +180,7 @@ record CommandDependencies(
ClientPresenceManager clientPresenceManager = new ClientPresenceManager(clientPresenceCluster,
Executors.newSingleThreadScheduledExecutor(), keyspaceNotificationDispatchExecutor);
MessagesCache messagesCache = new MessagesCache(messageInsertCacheCluster, messageReadDeleteCluster,
Clock.systemUTC(), keyspaceNotificationDispatchExecutor, messageDeliveryScheduler, messageDeletionExecutor);
keyspaceNotificationDispatchExecutor, messageDeliveryScheduler, messageDeletionExecutor, Clock.systemUTC());
DirectoryQueue directoryQueue = new DirectoryQueue(
configuration.getDirectoryConfiguration().getSqsConfiguration());
ProfilesManager profilesManager = new ProfilesManager(profiles, cacheCluster);