opendaylight/md-sal/sal-akka-raft/src/test/java/org/opendaylight/controller/cluster/raft/ReplicationAndSnapshotsWithLaggingFollowerIntegrationTest.java

   1 /*
   2  * Copyright (c) 2015 Brocade Communications 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.cluster.raft;
   9
  10 import static org.junit.Assert.assertEquals;
  11 import static org.junit.Assert.assertNotNull;
  12
  13 import akka.actor.ActorRef;
  14 import akka.persistence.SaveSnapshotSuccess;
  15 import com.google.common.collect.ImmutableMap;
  16 import com.google.common.util.concurrent.Uninterruptibles;
  17 import java.util.Arrays;
  18 import java.util.HashSet;
  19 import java.util.List;
  20 import java.util.Map;
  21 import java.util.Set;
  22 import java.util.concurrent.TimeUnit;
  23 import javax.annotation.Nullable;
  24 import org.junit.Assert;
  25 import org.junit.Test;
  26 import org.opendaylight.controller.cluster.raft.MockRaftActorContext.MockPayload;
  27 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  28 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
  29 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  30 import org.opendaylight.controller.cluster.raft.behaviors.AbstractLeader;
  31 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  32 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  33 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  34 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  35 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  36 import org.opendaylight.controller.cluster.raft.persisted.ApplyJournalEntries;
  37 import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload;
  38 import org.opendaylight.controller.cluster.raft.persisted.ServerInfo;
  39 import org.opendaylight.controller.cluster.raft.persisted.UpdateElectionTerm;
  40 import org.opendaylight.controller.cluster.raft.utils.InMemoryJournal;
  41 import org.opendaylight.controller.cluster.raft.utils.InMemorySnapshotStore;
  42 import org.opendaylight.controller.cluster.raft.utils.MessageCollectorActor;
  43
  44 /**
  45  * Tests replication and snapshots end-to-end using real RaftActors and behavior communication with a
  46  * lagging follower.
  47  *
  48  * @author Thomas Pantelis
  49  */
  50 public class ReplicationAndSnapshotsWithLaggingFollowerIntegrationTest extends AbstractRaftActorIntegrationTest {
  51
  52     private void setup() {
  53         leaderId = factory.generateActorId("leader");
  54         follower1Id = factory.generateActorId("follower");
  55         follower2Id = factory.generateActorId("follower");
  56
  57         // Setup the persistent journal for the leader - just an election term and no journal/snapshots.
  58         InMemoryJournal.addEntry(leaderId, 1, new UpdateElectionTerm(initialTerm, leaderId));
  59
  60         // Create the leader and 2 follower actors.
  61         follower1Actor = newTestRaftActor(follower1Id, ImmutableMap.of(leaderId, testActorPath(leaderId),
  62                 follower2Id, testActorPath(follower2Id)), newFollowerConfigParams());
  63
  64         follower2Actor = newTestRaftActor(follower2Id, ImmutableMap.of(leaderId, testActorPath(leaderId),
  65                 follower1Id, testActorPath(follower1Id)), newFollowerConfigParams());
  66
  67         Map<String, String> leaderPeerAddresses = ImmutableMap.<String, String>builder()
  68                 .put(follower1Id, follower1Actor.path().toString())
  69                 .put(follower2Id, follower2Actor.path().toString()).build();
  70
  71         leaderConfigParams = newLeaderConfigParams();
  72         leaderActor = newTestRaftActor(leaderId, leaderPeerAddresses, leaderConfigParams);
  73
  74         waitUntilLeader(leaderActor);
  75
  76         leaderContext = leaderActor.underlyingActor().getRaftActorContext();
  77         leader = leaderActor.underlyingActor().getCurrentBehavior();
  78
  79         follower1Context = follower1Actor.underlyingActor().getRaftActorContext();
  80         follower1 = follower1Actor.underlyingActor().getCurrentBehavior();
  81
  82         follower2Context = follower2Actor.underlyingActor().getRaftActorContext();
  83         follower2 = follower2Actor.underlyingActor().getCurrentBehavior();
  84
  85         currentTerm = leaderContext.getTermInformation().getCurrentTerm();
  86         assertEquals("Current term > " + initialTerm, true, currentTerm > initialTerm);
  87
  88         leaderCollectorActor = leaderActor.underlyingActor().collectorActor();
  89         follower1CollectorActor = follower1Actor.underlyingActor().collectorActor();
  90         follower2CollectorActor = follower2Actor.underlyingActor().collectorActor();
  91
  92         testLog.info("Leader created and elected");
  93     }
  94
  95     /**
  96      * Send 2 payload instances with follower 2 lagging then resume the follower and verifies it gets
  97      * caught up via AppendEntries.
  98      */
  99     @Test
 100     public void testReplicationsWithLaggingFollowerCaughtUpViaAppendEntries() throws Exception {
 101         testLog.info("testReplicationsWithLaggingFollowerCaughtUpViaAppendEntries starting: sending 2 new payloads");
 102
 103         setup();
 104
 105         // Simulate lagging by dropping AppendEntries messages in follower 2.
 106         follower2Actor.underlyingActor().startDropMessages(AppendEntries.class);
 107
 108         // Send the payloads.
 109         MockPayload payload0 = sendPayloadData(leaderActor, "zero");
 110         MockPayload payload1 = sendPayloadData(leaderActor, "one");
 111
 112         // Verify the leader got consensus and applies each log entry even though follower 2 didn't respond.
 113         List<ApplyState> applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor, ApplyState.class, 2);
 114         verifyApplyState(applyStates.get(0), leaderCollectorActor, payload0.toString(), currentTerm, 0, payload0);
 115         verifyApplyState(applyStates.get(1), leaderCollectorActor, payload1.toString(), currentTerm, 1, payload1);
 116
 117         // Verify follower 1 applies each log entry.
 118         applyStates = MessageCollectorActor.expectMatching(follower1CollectorActor, ApplyState.class, 2);
 119         verifyApplyState(applyStates.get(0), null, null, currentTerm, 0, payload0);
 120         verifyApplyState(applyStates.get(1), null, null, currentTerm, 1, payload1);
 121
 122         // Ensure there's at least 1 more heartbeat.
 123         MessageCollectorActor.clearMessages(leaderCollectorActor);
 124         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, AppendEntriesReply.class);
 125
 126         // The leader should not have performed fake snapshots to trim the log because the entries have not
 127         // been replicated to follower 2.
 128         assertEquals("Leader snapshot term", -1, leaderContext.getReplicatedLog().getSnapshotTerm());
 129         assertEquals("Leader snapshot index", -1, leaderContext.getReplicatedLog().getSnapshotIndex());
 130         assertEquals("Leader journal log size", 2, leaderContext.getReplicatedLog().size());
 131         assertEquals("Leader journal last index", 1, leaderContext.getReplicatedLog().lastIndex());
 132         assertEquals("Leader commit index", 1, leaderContext.getCommitIndex());
 133         assertEquals("Leader last applied", 1, leaderContext.getLastApplied());
 134         assertEquals("Leader replicatedToAllIndex", -1, leader.getReplicatedToAllIndex());
 135
 136         testLog.info(
 137             "testReplicationsWithLaggingFollowerCaughtUpViaAppendEntries: new entries applied - resuming follower {}",
 138             follower2Id);
 139
 140         // Now stop dropping AppendEntries in follower 2.
 141         follower2Actor.underlyingActor().stopDropMessages(AppendEntries.class);
 142
 143         // Verify follower 2 applies each log entry.
 144         applyStates = MessageCollectorActor.expectMatching(follower2CollectorActor, ApplyState.class, 2);
 145         verifyApplyState(applyStates.get(0), null, null, currentTerm, 0, payload0);
 146         verifyApplyState(applyStates.get(1), null, null, currentTerm, 1, payload1);
 147
 148         // Ensure there's at least 1 more heartbeat.
 149         MessageCollectorActor.clearMessages(leaderCollectorActor);
 150         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, AppendEntriesReply.class);
 151
 152         // The leader should now have performed fake snapshots to trim the log.
 153         verifyLeadersTrimmedLog(1);
 154
 155         // Even though follower 2 lagged behind, the leader should not have tried to install a snapshot
 156         // to catch it up because no snapshotting was done so the follower's next index was present in the log.
 157         InstallSnapshot installSnapshot = MessageCollectorActor.getFirstMatching(follower2CollectorActor,
 158                 InstallSnapshot.class);
 159         Assert.assertNull("Follower 2 received unexpected InstallSnapshot", installSnapshot);
 160
 161         testLog.info("testReplicationsWithLaggingFollowerCaughtUpViaAppendEntries complete");
 162     }
 163
 164     /**
 165      * Send payloads to trigger a leader snapshot due to snapshotBatchCount reached with follower 2
 166      * lagging but not enough for the leader to trim its log from the last applied index. Follower 2's log
 167      * will be behind by several entries and, when it is resumed, it should be caught up via AppendEntries
 168      * sent by the leader.
 169      */
 170     @Test
 171     public void testLeaderSnapshotWithLaggingFollowerCaughtUpViaAppendEntries() throws Exception {
 172         testLog.info("testLeaderSnapshotWithLaggingFollowerCaughtUpViaAppendEntries starting");
 173
 174         setup();
 175
 176         sendInitialPayloadsReplicatedToAllFollowers("zero", "one");
 177
 178         // Configure follower 2 to drop messages and lag.
 179         follower2Actor.underlyingActor().startDropMessages(AppendEntries.class);
 180
 181         // Send the first payload and verify it gets applied by the leader and follower 1.
 182         MockPayload payload2 = sendPayloadData(leaderActor, "two");
 183
 184         ApplyState applyState = MessageCollectorActor.expectFirstMatching(leaderCollectorActor, ApplyState.class);
 185         verifyApplyState(applyState, leaderCollectorActor, payload2.toString(), currentTerm, 2, payload2);
 186
 187         applyState = MessageCollectorActor.expectFirstMatching(follower1CollectorActor, ApplyState.class);
 188         verifyApplyState(applyState, null, null, currentTerm, 2, payload2);
 189
 190         expSnapshotState.add(payload2);
 191
 192         MessageCollectorActor.clearMessages(leaderCollectorActor);
 193         MessageCollectorActor.clearMessages(follower1CollectorActor);
 194
 195         // Send another payload - this should cause a snapshot due to snapshotBatchCount reached.
 196         MockPayload payload3 = sendPayloadData(leaderActor, "three");
 197
 198         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 199
 200         testLog.info("testLeaderSnapshotWithLaggingFollowerCaughtUpViaAppendEntries: sending 2 more payloads");
 201
 202         // Send 2 more payloads - not enough to trigger another snapshot.
 203         MockPayload payload4 = sendPayloadData(leaderActor, "four");
 204         MockPayload payload5 = sendPayloadData(leaderActor, "five");
 205
 206         // Verify the leader got consensus and applies each log entry even though follower 2 didn't respond.
 207         List<ApplyState> applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor, ApplyState.class, 3);
 208         verifyApplyState(applyStates.get(0), leaderCollectorActor, payload3.toString(), currentTerm, 3, payload3);
 209         verifyApplyState(applyStates.get(1), leaderCollectorActor, payload4.toString(), currentTerm, 4, payload4);
 210         verifyApplyState(applyStates.get(2), leaderCollectorActor, payload5.toString(), currentTerm, 5, payload5);
 211
 212         // Verify follower 1 applies each log entry.
 213         applyStates = MessageCollectorActor.expectMatching(follower1CollectorActor, ApplyState.class, 3);
 214         verifyApplyState(applyStates.get(0), null, null, currentTerm, 3, payload3);
 215         verifyApplyState(applyStates.get(1), null, null, currentTerm, 4, payload4);
 216         verifyApplyState(applyStates.get(2), null, null, currentTerm, 5, payload5);
 217
 218         // The snapshot should have caused the leader to advanced the snapshot index to the
 219         // last previously applied index (1) that was replicated to all followers at the time of capture.
 220         // Note: since the log size (3) did not exceed the snapshot batch count (4), the leader should not
 221         // have trimmed the log to the last index actually applied (5).
 222         assertEquals("Leader snapshot term", currentTerm, leaderContext.getReplicatedLog().getSnapshotTerm());
 223         assertEquals("Leader snapshot index", 1, leaderContext.getReplicatedLog().getSnapshotIndex());
 224         assertEquals("Leader journal log size", 4, leaderContext.getReplicatedLog().size());
 225         assertEquals("Leader journal last index", 5, leaderContext.getReplicatedLog().lastIndex());
 226         assertEquals("Leader commit index", 5, leaderContext.getCommitIndex());
 227         assertEquals("Leader last applied", 5, leaderContext.getLastApplied());
 228         assertEquals("Leader replicatedToAllIndex", 1, leader.getReplicatedToAllIndex());
 229
 230         // Now stop dropping AppendEntries in follower 2.
 231         follower2Actor.underlyingActor().stopDropMessages(AppendEntries.class);
 232
 233         // Verify follower 2 applies each log entry. The leader should not install a snapshot b/c
 234         // follower 2's next index (3) is still present in the log.
 235         applyStates = MessageCollectorActor.expectMatching(follower2CollectorActor, ApplyState.class, 4);
 236         verifyApplyState(applyStates.get(0), null, null, currentTerm, 2, payload2);
 237         verifyApplyState(applyStates.get(1), null, null, currentTerm, 3, payload3);
 238         verifyApplyState(applyStates.get(2), null, null, currentTerm, 4, payload4);
 239         verifyApplyState(applyStates.get(3), null, null, currentTerm, 5, payload5);
 240
 241         // Verify the leader did not try to install a snapshot to catch up follower 2.
 242         InstallSnapshot installSnapshot = MessageCollectorActor.getFirstMatching(follower2CollectorActor,
 243                 InstallSnapshot.class);
 244         Assert.assertNull("Follower 2 received unexpected InstallSnapshot", installSnapshot);
 245
 246         // Ensure there's at least 1 more heartbeat.
 247         MessageCollectorActor.clearMessages(leaderCollectorActor);
 248         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, AppendEntriesReply.class);
 249
 250         // The leader should now have performed fake snapshots to advance the snapshot index and to trim
 251         // the log. In addition replicatedToAllIndex should've advanced.
 252         verifyLeadersTrimmedLog(5);
 253
 254         // Verify the leader's persisted snapshot.
 255         List<Snapshot> persistedSnapshots = InMemorySnapshotStore.getSnapshots(leaderId, Snapshot.class);
 256         assertEquals("Persisted snapshots size", 1, persistedSnapshots.size());
 257         verifySnapshot("Persisted", persistedSnapshots.get(0), currentTerm, 2, currentTerm, 3);
 258         List<ReplicatedLogEntry> unAppliedEntry = persistedSnapshots.get(0).getUnAppliedEntries();
 259         assertEquals("Persisted Snapshot getUnAppliedEntries size", 1, unAppliedEntry.size());
 260         verifyReplicatedLogEntry(unAppliedEntry.get(0), currentTerm, 3, payload3);
 261
 262         // Verify follower 1's log and snapshot indexes.
 263         MessageCollectorActor.clearMessages(follower1CollectorActor);
 264         MessageCollectorActor.expectFirstMatching(follower1CollectorActor, AppendEntries.class);
 265         verifyFollowersTrimmedLog(1, follower1Actor, 5);
 266
 267         // Verify follower 2's log and snapshot indexes.
 268         MessageCollectorActor.clearMessages(follower2CollectorActor);
 269         MessageCollectorActor.expectFirstMatching(follower2CollectorActor, AppendEntries.class);
 270         verifyFollowersTrimmedLog(2, follower2Actor, 5);
 271
 272         MessageCollectorActor.clearMessages(leaderCollectorActor);
 273         MessageCollectorActor.clearMessages(follower1CollectorActor);
 274         MessageCollectorActor.clearMessages(follower2CollectorActor);
 275
 276         expSnapshotState.add(payload3);
 277         expSnapshotState.add(payload4);
 278         expSnapshotState.add(payload5);
 279
 280         testLog.info("testLeaderSnapshotWithLaggingFollowerCaughtUpViaAppendEntries complete");
 281     }
 282
 283     /**
 284      * Send payloads to trigger a leader snapshot due to snapshotBatchCount reached with follower 2
 285      * lagging where the leader trims its log from the last applied index. Follower 2's log
 286      * will be behind by several entries and, when it is resumed, it should be caught up via a snapshot
 287      * installed by the leader.
 288      */
 289     @Test
 290     public void testLeaderSnapshotWithLaggingFollowerCaughtUpViaInstallSnapshot() throws Exception {
 291         testLog.info("testLeaderSnapshotWithLaggingFollowerCaughtUpViaInstallSnapshot starting");
 292
 293         setup();
 294
 295         sendInitialPayloadsReplicatedToAllFollowers("zero", "one");
 296
 297         // Configure follower 2 to drop messages and lag.
 298         follower2Actor.underlyingActor().startDropMessages(AppendEntries.class);
 299
 300         // Send 5 payloads - the second should cause a leader snapshot.
 301         final MockPayload payload2 = sendPayloadData(leaderActor, "two");
 302         final MockPayload payload3 = sendPayloadData(leaderActor, "three");
 303         final MockPayload payload4 = sendPayloadData(leaderActor, "four");
 304         final MockPayload payload5 = sendPayloadData(leaderActor, "five");
 305         final MockPayload payload6 = sendPayloadData(leaderActor, "six");
 306
 307         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 308
 309         // Verify the leader got consensus and applies each log entry even though follower 2 didn't respond.
 310         List<ApplyState> applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor, ApplyState.class, 5);
 311         verifyApplyState(applyStates.get(0), leaderCollectorActor, payload2.toString(), currentTerm, 2, payload2);
 312         verifyApplyState(applyStates.get(2), leaderCollectorActor, payload4.toString(), currentTerm, 4, payload4);
 313         verifyApplyState(applyStates.get(4), leaderCollectorActor, payload6.toString(), currentTerm, 6, payload6);
 314
 315         MessageCollectorActor.clearMessages(leaderCollectorActor);
 316
 317         testLog.info("testLeaderSnapshotWithLaggingFollowerCaughtUpViaInstallSnapshot: "
 318                 + "sending 1 more payload to trigger second snapshot");
 319
 320         // Sleep for at least the election timeout interval so follower 2 is deemed inactive by the leader.
 321         Uninterruptibles.sleepUninterruptibly(leaderConfigParams.getElectionTimeOutInterval().toMillis() + 5,
 322                 TimeUnit.MILLISECONDS);
 323
 324         // Send another payload to trigger a second leader snapshot.
 325         MockPayload payload7 = sendPayloadData(leaderActor, "seven");
 326
 327         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 328
 329         ApplyState applyState = MessageCollectorActor.expectFirstMatching(leaderCollectorActor, ApplyState.class);
 330         verifyApplyState(applyState, leaderCollectorActor, payload7.toString(), currentTerm, 7, payload7);
 331
 332         // Verify follower 1 applies each log entry.
 333         applyStates = MessageCollectorActor.expectMatching(follower1CollectorActor, ApplyState.class, 6);
 334         verifyApplyState(applyStates.get(0), null, null, currentTerm, 2, payload2);
 335         verifyApplyState(applyStates.get(2), null, null, currentTerm, 4, payload4);
 336         verifyApplyState(applyStates.get(5), null, null, currentTerm, 7, payload7);
 337
 338         // The snapshot should have caused the leader to advanced the snapshot index to the leader's last
 339         // applied index (6) since the log size should have exceed the snapshot batch count (4).
 340         // replicatedToAllIndex should remain at 1 since follower 2 is lagging.
 341         verifyLeadersTrimmedLog(7, 1);
 342
 343         expSnapshotState.add(payload2);
 344         expSnapshotState.add(payload3);
 345         expSnapshotState.add(payload4);
 346         expSnapshotState.add(payload5);
 347         expSnapshotState.add(payload6);
 348
 349         MessageCollectorActor.clearMessages(leaderCollectorActor);
 350         MessageCollectorActor.clearMessages(follower1CollectorActor);
 351
 352         // Send a server config change to test that the install snapshot includes the server config.
 353
 354         ServerConfigurationPayload serverConfig = new ServerConfigurationPayload(Arrays.asList(
 355                 new ServerInfo(leaderId, true),
 356                 new ServerInfo(follower1Id, false),
 357                 new ServerInfo(follower2Id, false)));
 358         leaderContext.updatePeerIds(serverConfig);
 359         ((AbstractLeader)leader).updateMinReplicaCount();
 360         leaderActor.tell(serverConfig, ActorRef.noSender());
 361
 362         applyState = MessageCollectorActor.expectFirstMatching(leaderCollectorActor, ApplyState.class);
 363         verifyApplyState(applyState, leaderCollectorActor, "serverConfig", currentTerm, 8, serverConfig);
 364
 365         applyState = MessageCollectorActor.expectFirstMatching(follower1CollectorActor, ApplyState.class);
 366         verifyApplyState(applyState, null, null, currentTerm, 8, serverConfig);
 367
 368         // Verify the leader's persisted snapshot.
 369         List<Snapshot> persistedSnapshots = InMemorySnapshotStore.getSnapshots(leaderId, Snapshot.class);
 370         assertEquals("Persisted snapshots size", 1, persistedSnapshots.size());
 371         verifySnapshot("Persisted", persistedSnapshots.get(0), currentTerm, 6, currentTerm, 7);
 372         List<ReplicatedLogEntry> unAppliedEntry = persistedSnapshots.get(0).getUnAppliedEntries();
 373         assertEquals("Persisted Snapshot getUnAppliedEntries size", 1, unAppliedEntry.size());
 374         verifyReplicatedLogEntry(unAppliedEntry.get(0), currentTerm, 7, payload7);
 375
 376         expSnapshotState.add(payload7);
 377
 378         verifyInstallSnapshotToLaggingFollower(8, serverConfig);
 379
 380         testLog.info("testLeaderSnapshotWithLaggingFollowerCaughtUpViaInstallSnapshot complete");
 381     }
 382
 383     /**
 384      * Send payloads with follower 2 lagging with the last payload having a large enough size to trigger a
 385      * leader snapshot such that the leader trims its log from the last applied index.. Follower 2's log will
 386      * be behind by several entries and, when it is resumed, it should be caught up via a snapshot installed
 387      * by the leader.
 388      */
 389     @Test
 390     public void testLeaderSnapshotTriggeredByMemoryThresholdExceededWithLaggingFollower() throws Exception {
 391         testLog.info("testLeaderSnapshotTriggeredByMemoryThresholdExceededWithLaggingFollower starting");
 392
 393         snapshotBatchCount = 5;
 394         setup();
 395
 396         sendInitialPayloadsReplicatedToAllFollowers("zero");
 397
 398         leaderActor.underlyingActor().setMockTotalMemory(1000);
 399
 400         // We'll expect a ReplicatedLogImplEntry message and an ApplyJournalEntries message added to the journal.
 401         InMemoryJournal.addWriteMessagesCompleteLatch(leaderId, 2);
 402
 403         follower2Actor.underlyingActor().startDropMessages(AppendEntries.class);
 404
 405         // Send a payload with a large relative size but not enough to trigger a snapshot.
 406         MockPayload payload1 = sendPayloadData(leaderActor, "one", 500);
 407
 408         // Verify the leader got consensus and applies the first log entry even though follower 2 didn't respond.
 409         List<ApplyState> applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor, ApplyState.class, 1);
 410         verifyApplyState(applyStates.get(0), leaderCollectorActor, payload1.toString(), currentTerm, 1, payload1);
 411
 412         // Wait for all the ReplicatedLogImplEntry and ApplyJournalEntries messages to be added to the journal
 413         // before the snapshot so the snapshot sequence # will be higher to ensure the snapshot gets
 414         // purged from the snapshot store after subsequent snapshots.
 415         InMemoryJournal.waitForWriteMessagesComplete(leaderId);
 416
 417         // Verify a snapshot is not triggered.
 418         CaptureSnapshot captureSnapshot = MessageCollectorActor.getFirstMatching(leaderCollectorActor,
 419                 CaptureSnapshot.class);
 420         Assert.assertNull("Leader received unexpected CaptureSnapshot", captureSnapshot);
 421
 422         expSnapshotState.add(payload1);
 423
 424         // Sleep for at least the election timeout interval so follower 2 is deemed inactive by the leader.
 425         Uninterruptibles.sleepUninterruptibly(leaderConfigParams.getElectionTimeOutInterval().toMillis() + 5,
 426                 TimeUnit.MILLISECONDS);
 427
 428         // Send another payload with a large enough relative size in combination with the last payload
 429         // that exceeds the memory threshold (70% * 1000 = 700) - this should do a snapshot.
 430         MockPayload payload2 = sendPayloadData(leaderActor, "two", 201);
 431
 432         // Verify the leader applies the last log entry.
 433         applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor, ApplyState.class, 2);
 434         verifyApplyState(applyStates.get(1), leaderCollectorActor, payload2.toString(), currentTerm, 2, payload2);
 435
 436         // Verify follower 1 applies each log entry.
 437         applyStates = MessageCollectorActor.expectMatching(follower1CollectorActor, ApplyState.class, 2);
 438         verifyApplyState(applyStates.get(0), null, null, currentTerm, 1, payload1);
 439         verifyApplyState(applyStates.get(1), null, null, currentTerm, 2, payload2);
 440
 441         // A snapshot should've occurred - wait for it to complete.
 442         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 443
 444         // Because the snapshot was triggered by exceeding the memory threshold the leader should've advanced
 445         // the snapshot index to the last applied index and trimmed the log even though the entries weren't
 446         // replicated to all followers.
 447         verifyLeadersTrimmedLog(2, 0);
 448
 449         // Verify the leader's persisted snapshot.
 450         List<Snapshot> persistedSnapshots = InMemorySnapshotStore.getSnapshots(leaderId, Snapshot.class);
 451         assertEquals("Persisted snapshots size", 1, persistedSnapshots.size());
 452         verifySnapshot("Persisted", persistedSnapshots.get(0), currentTerm, 1, currentTerm, 2);
 453         List<ReplicatedLogEntry> unAppliedEntry = persistedSnapshots.get(0).getUnAppliedEntries();
 454         assertEquals("Persisted Snapshot getUnAppliedEntries size", 1, unAppliedEntry.size());
 455         verifyReplicatedLogEntry(unAppliedEntry.get(0), currentTerm, 2, payload2);
 456
 457         expSnapshotState.add(payload2);
 458
 459         verifyInstallSnapshotToLaggingFollower(2L, null);
 460
 461         // Sends a payload with index 3.
 462         verifyNoSubsequentSnapshotAfterMemoryThresholdExceededSnapshot();
 463
 464         // Sends 3 payloads with indexes 4, 5 and 6.
 465         verifyReplicationsAndSnapshotWithNoLaggingAfterInstallSnapshot();
 466
 467         // Recover the leader from persistence and verify.
 468         long leadersLastIndexOnRecovery = 6;
 469
 470         // The leader's last snapshot was triggered by index 4 so the last applied index in the snapshot was 3.
 471         long leadersSnapshotIndexOnRecovery = 3;
 472
 473         // The recovered journal should have 3 entries starting at index 4.
 474         long leadersFirstJournalEntryIndexOnRecovery = 4;
 475
 476         verifyLeaderRecoveryAfterReinstatement(leadersLastIndexOnRecovery, leadersSnapshotIndexOnRecovery,
 477                 leadersFirstJournalEntryIndexOnRecovery);
 478
 479         testLog.info("testLeaderSnapshotTriggeredByMemoryThresholdExceeded ending");
 480     }
 481
 482     /**
 483      * Send another payload to verify another snapshot is not done since the last snapshot trimmed the
 484      * first log entry so the memory threshold should not be exceeded.
 485      */
 486     private void verifyNoSubsequentSnapshotAfterMemoryThresholdExceededSnapshot() throws Exception {
 487         ApplyState applyState;
 488         CaptureSnapshot captureSnapshot;
 489
 490         MockPayload payload3 = sendPayloadData(leaderActor, "three");
 491
 492         // Verify the leader applies the state.
 493         applyState = MessageCollectorActor.expectFirstMatching(leaderCollectorActor, ApplyState.class);
 494         verifyApplyState(applyState, leaderCollectorActor, payload3.toString(), currentTerm, 3, payload3);
 495
 496         captureSnapshot = MessageCollectorActor.getFirstMatching(leaderCollectorActor, CaptureSnapshot.class);
 497         Assert.assertNull("Leader received unexpected CaptureSnapshot", captureSnapshot);
 498
 499         // Verify the follower 1 applies the state.
 500         applyState = MessageCollectorActor.expectFirstMatching(follower1CollectorActor, ApplyState.class);
 501         verifyApplyState(applyState, null, null, currentTerm, 3, payload3);
 502
 503         // Verify the follower 2 applies the state.
 504         applyState = MessageCollectorActor.expectFirstMatching(follower2CollectorActor, ApplyState.class);
 505         verifyApplyState(applyState, null, null, currentTerm, 3, payload3);
 506
 507         // Verify the leader's state.
 508         verifyLeadersTrimmedLog(3);
 509
 510         // Verify follower 1's state.
 511         verifyFollowersTrimmedLog(1, follower1Actor, 3);
 512
 513         // Verify follower 2's state.
 514         verifyFollowersTrimmedLog(2, follower2Actor, 3);
 515
 516         // Revert back to JVM total memory.
 517         leaderActor.underlyingActor().setMockTotalMemory(0);
 518
 519         MessageCollectorActor.clearMessages(leaderCollectorActor);
 520         MessageCollectorActor.clearMessages(follower1CollectorActor);
 521         MessageCollectorActor.clearMessages(follower2CollectorActor);
 522
 523         expSnapshotState.add(payload3);
 524     }
 525
 526     /**
 527      * Resume the lagging follower 2 and verify it receives an install snapshot from the leader.
 528      */
 529     private void verifyInstallSnapshotToLaggingFollower(long lastAppliedIndex,
 530             @Nullable ServerConfigurationPayload expServerConfig) throws Exception {
 531         testLog.info("testInstallSnapshotToLaggingFollower starting");
 532
 533         MessageCollectorActor.clearMessages(leaderCollectorActor);
 534
 535         // Now stop dropping AppendEntries in follower 2.
 536         follower2Actor.underlyingActor().stopDropMessages(AppendEntries.class);
 537
 538
 539         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 540
 541         // Verify the leader's persisted snapshot. The previous snapshot (currently) won't be deleted from
 542         // the snapshot store because the second snapshot was initiated by the follower install snapshot and
 543         // not because the batch count was reached so the persisted journal sequence number wasn't advanced
 544         // far enough to cause the previous snapshot to be deleted. This is because
 545         // RaftActor#trimPersistentData subtracts the snapshotBatchCount from the snapshot's sequence number.
 546         // This is OK - the next snapshot should delete it. In production, even if the system restarted
 547         // before another snapshot, they would both get applied which wouldn't hurt anything.
 548         List<Snapshot> persistedSnapshots = InMemorySnapshotStore.getSnapshots(leaderId, Snapshot.class);
 549         Assert.assertTrue("Expected at least 1 persisted snapshots", persistedSnapshots.size() > 0);
 550         Snapshot persistedSnapshot = persistedSnapshots.get(persistedSnapshots.size() - 1);
 551         verifySnapshot("Persisted", persistedSnapshot, currentTerm, lastAppliedIndex, currentTerm, lastAppliedIndex);
 552         List<ReplicatedLogEntry> unAppliedEntry = persistedSnapshot.getUnAppliedEntries();
 553         assertEquals("Persisted Snapshot getUnAppliedEntries size", 0, unAppliedEntry.size());
 554
 555         int snapshotSize = persistedSnapshot.getState().length;
 556         final int expTotalChunks = snapshotSize / SNAPSHOT_CHUNK_SIZE
 557                 + (snapshotSize % SNAPSHOT_CHUNK_SIZE > 0 ? 1 : 0);
 558
 559         InstallSnapshot installSnapshot = MessageCollectorActor.expectFirstMatching(follower2CollectorActor,
 560                 InstallSnapshot.class);
 561         assertEquals("InstallSnapshot getTerm", currentTerm, installSnapshot.getTerm());
 562         assertEquals("InstallSnapshot getLeaderId", leaderId, installSnapshot.getLeaderId());
 563         assertEquals("InstallSnapshot getChunkIndex", 1, installSnapshot.getChunkIndex());
 564         assertEquals("InstallSnapshot getTotalChunks", expTotalChunks, installSnapshot.getTotalChunks());
 565         assertEquals("InstallSnapshot getLastIncludedTerm", currentTerm, installSnapshot.getLastIncludedTerm());
 566         assertEquals("InstallSnapshot getLastIncludedIndex", lastAppliedIndex, installSnapshot.getLastIncludedIndex());
 567         //assertArrayEquals("InstallSnapshot getData", snapshot, installSnapshot.getData().toByteArray());
 568
 569         List<InstallSnapshotReply> installSnapshotReplies = MessageCollectorActor.expectMatching(
 570                 leaderCollectorActor, InstallSnapshotReply.class, expTotalChunks);
 571         int index = 1;
 572         for (InstallSnapshotReply installSnapshotReply: installSnapshotReplies) {
 573             assertEquals("InstallSnapshotReply getTerm", currentTerm, installSnapshotReply.getTerm());
 574             assertEquals("InstallSnapshotReply getChunkIndex", index++, installSnapshotReply.getChunkIndex());
 575             assertEquals("InstallSnapshotReply getFollowerId", follower2Id, installSnapshotReply.getFollowerId());
 576             assertEquals("InstallSnapshotReply isSuccess", true, installSnapshotReply.isSuccess());
 577         }
 578
 579         // Verify follower 2 applies the snapshot.
 580         ApplySnapshot applySnapshot = MessageCollectorActor.expectFirstMatching(follower2CollectorActor,
 581                 ApplySnapshot.class);
 582         verifySnapshot("Follower 2", applySnapshot.getSnapshot(), currentTerm, lastAppliedIndex, currentTerm,
 583                 lastAppliedIndex);
 584         assertEquals("Persisted Snapshot getUnAppliedEntries size", 0,
 585                 applySnapshot.getSnapshot().getUnAppliedEntries().size());
 586
 587         // Wait for the snapshot to complete.
 588         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 589
 590         // Ensure there's at least 1 more heartbeat.
 591         MessageCollectorActor.clearMessages(leaderCollectorActor);
 592         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, AppendEntriesReply.class);
 593
 594         // The leader should now have performed fake snapshots to advance the snapshot index and to trim
 595         // the log. In addition replicatedToAllIndex should've advanced.
 596         verifyLeadersTrimmedLog(lastAppliedIndex);
 597
 598         if (expServerConfig != null) {
 599             Set<ServerInfo> expServerInfo = new HashSet<>(expServerConfig.getServerConfig());
 600             assertEquals("Leader snapshot server config", expServerInfo,
 601                     new HashSet<>(persistedSnapshot.getServerConfiguration().getServerConfig()));
 602
 603             assertEquals("Follower 2 snapshot server config", expServerInfo,
 604                     new HashSet<>(applySnapshot.getSnapshot().getServerConfiguration().getServerConfig()));
 605
 606             ServerConfigurationPayload follower2ServerConfig = follower2Context.getPeerServerInfo(true);
 607             assertNotNull("Follower 2 server config is null", follower2ServerConfig);
 608
 609             assertEquals("Follower 2 server config", expServerInfo,
 610                     new HashSet<>(follower2ServerConfig.getServerConfig()));
 611         }
 612
 613         MessageCollectorActor.clearMessages(leaderCollectorActor);
 614         MessageCollectorActor.clearMessages(follower1CollectorActor);
 615         MessageCollectorActor.clearMessages(follower2CollectorActor);
 616
 617         testLog.info("testInstallSnapshotToLaggingFollower complete");
 618     }
 619
 620     /**
 621      * Do another round of payloads and snapshot to verify replicatedToAllIndex gets back on track and
 622      * snapshots works as expected after doing a follower snapshot. In this step we don't lag a follower.
 623      */
 624     private void verifyReplicationsAndSnapshotWithNoLaggingAfterInstallSnapshot() throws Exception {
 625         testLog.info("testReplicationsAndSnapshotAfterInstallSnapshot starting: replicatedToAllIndex: {}",
 626                 leader.getReplicatedToAllIndex());
 627
 628         // Send another payload - a snapshot should occur.
 629         MockPayload payload4 = sendPayloadData(leaderActor, "four");
 630
 631         // Wait for the snapshot to complete.
 632         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 633
 634         ApplyState applyState = MessageCollectorActor.expectFirstMatching(leaderCollectorActor, ApplyState.class);
 635         verifyApplyState(applyState, leaderCollectorActor, payload4.toString(), currentTerm, 4, payload4);
 636
 637         // Verify the leader's last persisted snapshot (previous ones may not be purged yet).
 638         List<Snapshot> persistedSnapshots = InMemorySnapshotStore.getSnapshots(leaderId, Snapshot.class);
 639         Snapshot persistedSnapshot = persistedSnapshots.get(persistedSnapshots.size() - 1);
 640         verifySnapshot("Persisted", persistedSnapshot, currentTerm, 3, currentTerm, 4);
 641         List<ReplicatedLogEntry> unAppliedEntry = persistedSnapshot.getUnAppliedEntries();
 642         assertEquals("Persisted Snapshot getUnAppliedEntries size", 1, unAppliedEntry.size());
 643         verifyReplicatedLogEntry(unAppliedEntry.get(0), currentTerm, 4, payload4);
 644
 645         // Send a couple more payloads.
 646         MockPayload payload5 = sendPayloadData(leaderActor, "five");
 647         MockPayload payload6 = sendPayloadData(leaderActor, "six");
 648
 649         // Verify the leader applies the 2 log entries.
 650         List<ApplyState> applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor, ApplyState.class, 3);
 651         verifyApplyState(applyStates.get(1), leaderCollectorActor, payload5.toString(), currentTerm, 5, payload5);
 652         verifyApplyState(applyStates.get(2), leaderCollectorActor, payload6.toString(), currentTerm, 6, payload6);
 653
 654         // Verify the leader applies a log entry for at least the last entry index.
 655         verifyApplyJournalEntries(leaderCollectorActor, 6);
 656
 657         // Ensure there's at least 1 more heartbeat to trim the log.
 658         MessageCollectorActor.clearMessages(leaderCollectorActor);
 659         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, AppendEntriesReply.class);
 660
 661         // Verify the leader's final state.
 662         verifyLeadersTrimmedLog(6);
 663
 664         InMemoryJournal.dumpJournal(leaderId);
 665
 666         // Verify the leaders's persisted journal log - it should only contain the last 2 ReplicatedLogEntries
 667         // added after the snapshot as the persisted journal should've been purged to the snapshot
 668         // sequence number.
 669         verifyPersistedJournal(leaderId, Arrays.asList(new ReplicatedLogImplEntry(5, currentTerm, payload5),
 670                 new ReplicatedLogImplEntry(6, currentTerm, payload6)));
 671
 672         // Verify the leaders's persisted journal contains an ApplyJournalEntries for at least the last entry index.
 673         List<ApplyJournalEntries> persistedApplyJournalEntries =
 674                 InMemoryJournal.get(leaderId, ApplyJournalEntries.class);
 675         boolean found = false;
 676         for (ApplyJournalEntries entry: persistedApplyJournalEntries) {
 677             if (entry.getToIndex() == 6) {
 678                 found = true;
 679                 break;
 680             }
 681         }
 682
 683         Assert.assertTrue(String.format("ApplyJournalEntries with index %d not found in leader's persisted journal", 6),
 684                 found);
 685
 686         // Verify follower 1 applies the 3 log entries.
 687         applyStates = MessageCollectorActor.expectMatching(follower1CollectorActor, ApplyState.class, 3);
 688         verifyApplyState(applyStates.get(0), null, null, currentTerm, 4, payload4);
 689         verifyApplyState(applyStates.get(1), null, null, currentTerm, 5, payload5);
 690         verifyApplyState(applyStates.get(2), null, null, currentTerm, 6, payload6);
 691
 692         // Verify follower 1's log state.
 693         verifyFollowersTrimmedLog(1, follower1Actor, 6);
 694
 695         // Verify follower 2 applies the 3 log entries.
 696         applyStates = MessageCollectorActor.expectMatching(follower2CollectorActor, ApplyState.class, 3);
 697         verifyApplyState(applyStates.get(0), null, null, currentTerm, 4, payload4);
 698         verifyApplyState(applyStates.get(1), null, null, currentTerm, 5, payload5);
 699         verifyApplyState(applyStates.get(2), null, null, currentTerm, 6, payload6);
 700
 701         // Verify follower 2's log state.
 702         verifyFollowersTrimmedLog(2, follower2Actor, 6);
 703
 704         expSnapshotState.add(payload4);
 705         expSnapshotState.add(payload5);
 706         expSnapshotState.add(payload6);
 707
 708         testLog.info("testReplicationsAndSnapshotAfterInstallSnapshot ending");
 709     }
 710
 711     /**
 712      * Kill the leader actor, reinstate it and verify the recovered journal.
 713      */
 714     private void verifyLeaderRecoveryAfterReinstatement(long lastIndex, long snapshotIndex,
 715             long firstJournalEntryIndex) {
 716         testLog.info("testLeaderReinstatement starting");
 717
 718         killActor(leaderActor);
 719
 720         leaderActor = newTestRaftActor(leaderId, peerAddresses, leaderConfigParams);
 721         TestRaftActor testRaftActor = leaderActor.underlyingActor();
 722
 723         testRaftActor.startDropMessages(RequestVoteReply.class);
 724
 725         leaderContext = testRaftActor.getRaftActorContext();
 726
 727         testRaftActor.waitForRecoveryComplete();
 728
 729         int logSize = (int) (expSnapshotState.size() - firstJournalEntryIndex);
 730         assertEquals("Leader snapshot term", currentTerm, leaderContext.getReplicatedLog().getSnapshotTerm());
 731         assertEquals("Leader snapshot index", snapshotIndex, leaderContext.getReplicatedLog().getSnapshotIndex());
 732         assertEquals("Leader journal log size", logSize, leaderContext.getReplicatedLog().size());
 733         assertEquals("Leader journal last index", lastIndex, leaderContext.getReplicatedLog().lastIndex());
 734         assertEquals("Leader commit index", lastIndex, leaderContext.getCommitIndex());
 735         assertEquals("Leader last applied", lastIndex, leaderContext.getLastApplied());
 736
 737         for (long i = firstJournalEntryIndex; i < expSnapshotState.size(); i++) {
 738             verifyReplicatedLogEntry(leaderContext.getReplicatedLog().get(i), currentTerm, i,
 739                     expSnapshotState.get((int) i));
 740         }
 741
 742         assertEquals("Leader applied state", expSnapshotState, testRaftActor.getState());
 743
 744         testLog.info("testLeaderReinstatement ending");
 745     }
 746
 747     private void sendInitialPayloadsReplicatedToAllFollowers(String... data) {
 748
 749         // Send the payloads.
 750         for (String d: data) {
 751             expSnapshotState.add(sendPayloadData(leaderActor, d));
 752         }
 753
 754         int numEntries = data.length;
 755
 756         // Verify the leader got consensus and applies each log entry even though follower 2 didn't respond.
 757         List<ApplyState> applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor,
 758                 ApplyState.class, numEntries);
 759         for (int i = 0; i < expSnapshotState.size(); i++) {
 760             MockPayload payload = expSnapshotState.get(i);
 761             verifyApplyState(applyStates.get(i), leaderCollectorActor, payload.toString(), currentTerm, i, payload);
 762         }
 763
 764         // Verify follower 1 applies each log entry.
 765         applyStates = MessageCollectorActor.expectMatching(follower1CollectorActor, ApplyState.class, numEntries);
 766         for (int i = 0; i < expSnapshotState.size(); i++) {
 767             MockPayload payload = expSnapshotState.get(i);
 768             verifyApplyState(applyStates.get(i), null, null, currentTerm, i, payload);
 769         }
 770
 771         // Verify follower 2 applies each log entry.
 772         applyStates = MessageCollectorActor.expectMatching(follower2CollectorActor, ApplyState.class, numEntries);
 773         for (int i = 0; i < expSnapshotState.size(); i++) {
 774             MockPayload payload = expSnapshotState.get(i);
 775             verifyApplyState(applyStates.get(i), null, null, currentTerm, i, payload);
 776         }
 777
 778         // Ensure there's at least 1 more heartbeat.
 779         MessageCollectorActor.clearMessages(leaderCollectorActor);
 780         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, AppendEntriesReply.class);
 781
 782         // The leader should have performed fake snapshots to trim the log to the last index replicated to
 783         // all followers.
 784         verifyLeadersTrimmedLog(numEntries - 1);
 785
 786         MessageCollectorActor.clearMessages(leaderCollectorActor);
 787         MessageCollectorActor.clearMessages(follower1CollectorActor);
 788         MessageCollectorActor.clearMessages(follower2CollectorActor);
 789     }
 790 }