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 akka.persistence.SaveSnapshotSuccess;
  12 import com.google.common.collect.ImmutableMap;
  13 import java.util.Arrays;
  14 import java.util.List;
  15 import java.util.Map;
  16 import org.junit.Assert;
  17 import org.junit.Test;
  18 import org.opendaylight.controller.cluster.raft.MockRaftActorContext.MockPayload;
  19 import org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm;
  20 import org.opendaylight.controller.cluster.raft.base.messages.ApplyJournalEntries;
  21 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  22 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
  23 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  24 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  25 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  26 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  27 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  28 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  29 import org.opendaylight.controller.cluster.raft.utils.InMemoryJournal;
  30 import org.opendaylight.controller.cluster.raft.utils.InMemorySnapshotStore;
  31 import org.opendaylight.controller.cluster.raft.utils.MessageCollectorActor;
  32
  33 /**
  34  * Tests replication and snapshots end-to-end using real RaftActors and behavior communication with a
  35  * lagging follower.
  36  *
  37  * @author Thomas Pantelis
  38  */
  39 public class ReplicationAndSnapshotsWithLaggingFollowerIntegrationTest extends AbstractRaftActorIntegrationTest {
  40
  41     private MockPayload payload9;
  42     private MockPayload payload11;
  43     private MockPayload payload12;
  44     private MockPayload payload13;
  45
  46     @Test
  47     public void runTest() throws Exception {
  48         testLog.info("testReplicationAndSnapshotsWithLaggingFollower starting");
  49
  50         leaderId = factory.generateActorId("leader");
  51         follower1Id = factory.generateActorId("follower");
  52         follower2Id = factory.generateActorId("follower");
  53
  54         // Setup the persistent journal for the leader - just an election term and no journal/snapshots.
  55         InMemoryJournal.addEntry(leaderId, 1, new UpdateElectionTerm(initialTerm, leaderId));
  56
  57         // Create the leader and 2 follower actors.
  58         follower1Actor = newTestRaftActor(follower1Id, ImmutableMap.of(leaderId, testActorPath(leaderId),
  59                 follower2Id, testActorPath(follower2Id)), newFollowerConfigParams());
  60
  61         follower2Actor = newTestRaftActor(follower2Id, ImmutableMap.of(leaderId, testActorPath(leaderId),
  62                 follower1Id, testActorPath(follower1Id)), newFollowerConfigParams());
  63
  64         Map<String, String> peerAddresses = ImmutableMap.<String, String>builder().
  65                 put(follower1Id, follower1Actor.path().toString()).
  66                 put(follower2Id, follower2Actor.path().toString()).build();
  67
  68         leaderConfigParams = newLeaderConfigParams();
  69         leaderActor = newTestRaftActor(leaderId, peerAddresses, leaderConfigParams);
  70
  71         waitUntilLeader(leaderActor);
  72
  73         leaderContext = leaderActor.underlyingActor().getRaftActorContext();
  74         leader = leaderActor.underlyingActor().getCurrentBehavior();
  75
  76         follower1Context = follower1Actor.underlyingActor().getRaftActorContext();
  77         follower1 = follower1Actor.underlyingActor().getCurrentBehavior();
  78
  79         follower2Context = follower2Actor.underlyingActor().getRaftActorContext();
  80         follower2 = follower2Actor.underlyingActor().getCurrentBehavior();
  81
  82         currentTerm = leaderContext.getTermInformation().getCurrentTerm();
  83         assertEquals("Current term > " + initialTerm, true, currentTerm > initialTerm);
  84
  85         leaderCollectorActor = leaderActor.underlyingActor().collectorActor();
  86         follower1CollectorActor = follower1Actor.underlyingActor().collectorActor();
  87         follower2CollectorActor = follower2Actor.underlyingActor().collectorActor();
  88
  89         testLog.info("Leader created and elected");
  90
  91         testInitialReplications();
  92
  93         testSubsequentReplicationsAndSnapshots();
  94
  95         testLeaderSnapshotTriggeredByMemoryThresholdExceeded();
  96
  97         testInstallSnapshotToLaggingFollower();
  98
  99         verifyNoSubsequentSnapshotAfterMemoryThresholdExceededSnapshot();
 100
 101         testFinalReplicationsAndSnapshot();
 102
 103         testLeaderReinstatement();
 104
 105         testLog.info("testReplicationAndSnapshotsWithLaggingFollower ending");
 106     }
 107
 108     /**
 109      * Send 3 payload instances with follower 2 temporarily lagging.
 110      *
 111      * @throws Exception
 112      */
 113     private void testInitialReplications() throws Exception {
 114
 115         testLog.info("testInitialReplications starting: sending 2 new payloads");
 116
 117         // Simulate lagging by dropping AppendEntries messages in follower 2.
 118         follower2Actor.underlyingActor().startDropMessages(AppendEntries.class);
 119
 120         // Send the payloads.
 121         MockPayload payload0 = sendPayloadData(leaderActor, "zero");
 122         MockPayload payload1 = sendPayloadData(leaderActor, "one");
 123         MockPayload payload2 = sendPayloadData(leaderActor, "two");
 124
 125         // Verify the leader got consensus and applies each log entry even though follower 2 didn't respond.
 126         List<ApplyState> applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor, ApplyState.class, 3);
 127         verifyApplyState(applyStates.get(0), leaderCollectorActor, payload0.toString(), currentTerm, 0, payload0);
 128         verifyApplyState(applyStates.get(1), leaderCollectorActor, payload1.toString(), currentTerm, 1, payload1);
 129         verifyApplyState(applyStates.get(2), leaderCollectorActor, payload2.toString(), currentTerm, 2, payload2);
 130
 131         // Verify follower 1 applies each log entry.
 132         applyStates = MessageCollectorActor.expectMatching(follower1CollectorActor, ApplyState.class, 3);
 133         verifyApplyState(applyStates.get(0), null, null, currentTerm, 0, payload0);
 134         verifyApplyState(applyStates.get(1), null, null, currentTerm, 1, payload1);
 135         verifyApplyState(applyStates.get(2), null, null, currentTerm, 2, payload2);
 136
 137         // Ensure there's at least 1 more heartbeat.
 138         MessageCollectorActor.clearMessages(leaderCollectorActor);
 139         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, AppendEntriesReply.class);
 140
 141         // The leader should not have performed fake snapshots to trim the log because the entries have not
 142         // been replicated to follower 2.
 143         assertEquals("Leader snapshot term", -1, leaderContext.getReplicatedLog().getSnapshotTerm());
 144         assertEquals("Leader snapshot index", -1, leaderContext.getReplicatedLog().getSnapshotIndex());
 145         assertEquals("Leader journal log size", 3, leaderContext.getReplicatedLog().size());
 146         assertEquals("Leader journal last index", 2, leaderContext.getReplicatedLog().lastIndex());
 147         assertEquals("Leader commit index", 2, leaderContext.getCommitIndex());
 148         assertEquals("Leader last applied", 2, leaderContext.getLastApplied());
 149         assertEquals("Leader replicatedToAllIndex", -1, leader.getReplicatedToAllIndex());
 150
 151         testLog.info("Step 3: new entries applied - re-enabling follower {}", follower2Id);
 152
 153         // Now stop dropping AppendEntries in follower 2.
 154         follower2Actor.underlyingActor().stopDropMessages(AppendEntries.class);
 155
 156         // Verify follower 2 applies each log entry.
 157         applyStates = MessageCollectorActor.expectMatching(follower2CollectorActor, ApplyState.class, 3);
 158         verifyApplyState(applyStates.get(0), null, null, currentTerm, 0, payload0);
 159         verifyApplyState(applyStates.get(1), null, null, currentTerm, 1, payload1);
 160         verifyApplyState(applyStates.get(2), null, null, currentTerm, 2, payload2);
 161
 162         // Ensure there's at least 1 more heartbeat.
 163         MessageCollectorActor.clearMessages(leaderCollectorActor);
 164         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, AppendEntriesReply.class);
 165
 166         // The leader should now have performed fake snapshots to trim the log.
 167         assertEquals("Leader snapshot term", currentTerm, leaderContext.getReplicatedLog().getSnapshotTerm());
 168         assertEquals("Leader snapshot index", 1, leaderContext.getReplicatedLog().getSnapshotIndex());
 169         assertEquals("Leader journal log size", 1, leaderContext.getReplicatedLog().size());
 170         assertEquals("Leader journal last index", 2, leaderContext.getReplicatedLog().lastIndex());
 171         assertEquals("Leader commit index", 2, leaderContext.getCommitIndex());
 172         assertEquals("Leader last applied", 2, leaderContext.getLastApplied());
 173         // Note - replicatedToAllIndex always lags 1 behind last applied since it trims the log up to the
 174         // last applied index. The next entry successfully replicated to followers woild advance it.
 175         assertEquals("Leader replicatedToAllIndex", 1, leader.getReplicatedToAllIndex());
 176
 177         // Even though follower 2 lagged behind, the leader should not have tried to install a snapshot
 178         // to catch it up because no snapshotting was done so the follower's next index was present in the log.
 179         InstallSnapshot installSnapshot = MessageCollectorActor.getFirstMatching(follower2CollectorActor,
 180                 InstallSnapshot.class);
 181         Assert.assertNull("Follower 2 received unexpected InstallSnapshot", installSnapshot);
 182
 183         MessageCollectorActor.clearMessages(leaderCollectorActor);
 184         MessageCollectorActor.clearMessages(follower1CollectorActor);
 185         MessageCollectorActor.clearMessages(follower2CollectorActor);
 186
 187         expSnapshotState.add(payload0);
 188         expSnapshotState.add(payload1);
 189         expSnapshotState.add(payload2);
 190
 191         testLog.info("testInitialReplications complete");
 192     }
 193
 194     /**
 195      * Send 5 more payloads with follower 2 lagging. Since the snapshotBatch count is 4, this should cause
 196      * 2 leader snapshots and follower 2's log will be behind by 5 entries.
 197      *
 198      * @throws Exception
 199      */
 200     private void testSubsequentReplicationsAndSnapshots() throws Exception {
 201         testLog.info("testSubsequentReplicationsAndSnapshots starting: sending first payload, replicatedToAllIndex: {}",
 202                 leader.getReplicatedToAllIndex());
 203
 204         follower2Actor.underlyingActor().startDropMessages(AppendEntries.class);
 205
 206         // Send the first payload - this should cause the first snapshot.
 207         MockPayload payload3 = sendPayloadData(leaderActor, "three");
 208
 209         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 210
 211         expSnapshotState.add(payload3);
 212
 213         testLog.info("testSubsequentReplicationsAndSnapshots: sending 4 more payloads");
 214
 215         // Send the next 4. The last one should cause the second snapshot.
 216         MockPayload payload4 = sendPayloadData(leaderActor, "four");
 217         MockPayload payload5 = sendPayloadData(leaderActor, "five");
 218         MockPayload payload6 = sendPayloadData(leaderActor, "six");
 219         MockPayload payload7 = sendPayloadData(leaderActor, "seven");
 220
 221         // Verify the leader got consensus and applies each log entry even though follower 2 didn't respond.
 222         List<ApplyState> applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor, ApplyState.class, 5);
 223         verifyApplyState(applyStates.get(0), leaderCollectorActor, payload3.toString(), currentTerm, 3, payload3);
 224         verifyApplyState(applyStates.get(1), leaderCollectorActor, payload4.toString(), currentTerm, 4, payload4);
 225         verifyApplyState(applyStates.get(2), leaderCollectorActor, payload5.toString(), currentTerm, 5, payload5);
 226         verifyApplyState(applyStates.get(3), leaderCollectorActor, payload6.toString(), currentTerm, 6, payload6);
 227         verifyApplyState(applyStates.get(4), leaderCollectorActor, payload7.toString(), currentTerm, 7, payload7);
 228
 229         // Verify follower 1 applies each log entry.
 230         applyStates = MessageCollectorActor.expectMatching(follower1CollectorActor, ApplyState.class, 5);
 231         verifyApplyState(applyStates.get(0), null, null, currentTerm, 3, payload3);
 232         verifyApplyState(applyStates.get(1), null, null, currentTerm, 4, payload4);
 233         verifyApplyState(applyStates.get(2), null, null, currentTerm, 5, payload5);
 234         verifyApplyState(applyStates.get(3), null, null, currentTerm, 6, payload6);
 235         verifyApplyState(applyStates.get(4), null, null, currentTerm, 7, payload7);
 236
 237         // Wait for snapshot completion.
 238         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 239
 240         // The first new entry applied should have caused the leader to advanced the snapshot index to the
 241         // last previously applied index (2) that was replicated to all followers.
 242         assertEquals("Leader snapshot term", currentTerm, leaderContext.getReplicatedLog().getSnapshotTerm());
 243         assertEquals("Leader snapshot index", 2, leaderContext.getReplicatedLog().getSnapshotIndex());
 244         assertEquals("Leader journal log size", 5, leaderContext.getReplicatedLog().size());
 245         assertEquals("Leader journal last index", 7, leaderContext.getReplicatedLog().lastIndex());
 246         assertEquals("Leader commit index", 7, leaderContext.getCommitIndex());
 247         assertEquals("Leader last applied", 7, leaderContext.getLastApplied());
 248         assertEquals("Leader replicatedToAllIndex", 2, leader.getReplicatedToAllIndex());
 249
 250         // Now stop dropping AppendEntries in follower 2.
 251         follower2Actor.underlyingActor().stopDropMessages(AppendEntries.class);
 252
 253         // Verify follower 2 applies each log entry.
 254         applyStates = MessageCollectorActor.expectMatching(follower2CollectorActor, ApplyState.class, 5);
 255         verifyApplyState(applyStates.get(0), null, null, currentTerm, 3, payload3);
 256         verifyApplyState(applyStates.get(1), null, null, currentTerm, 4, payload4);
 257         verifyApplyState(applyStates.get(2), null, null, currentTerm, 5, payload5);
 258         verifyApplyState(applyStates.get(3), null, null, currentTerm, 6, payload6);
 259         verifyApplyState(applyStates.get(4), null, null, currentTerm, 7, payload7);
 260
 261         // Ensure there's at least 1 more heartbeat.
 262         MessageCollectorActor.clearMessages(leaderCollectorActor);
 263         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, AppendEntriesReply.class);
 264
 265         // The leader should now have performed fake snapshots to advance the snapshot index and to trim
 266         // the log. In addition replicatedToAllIndex should've advanced.
 267         assertEquals("Leader snapshot term", currentTerm, leaderContext.getReplicatedLog().getSnapshotTerm());
 268         assertEquals("Leader snapshot index", 6, leaderContext.getReplicatedLog().getSnapshotIndex());
 269         assertEquals("Leader journal log size", 1, leaderContext.getReplicatedLog().size());
 270         assertEquals("Leader journal last index", 7, leaderContext.getReplicatedLog().lastIndex());
 271         assertEquals("Leader replicatedToAllIndex", 6, leader.getReplicatedToAllIndex());
 272
 273         // Verify the leader's persisted snapshot.
 274         List<Snapshot> persistedSnapshots = InMemorySnapshotStore.getSnapshots(leaderId, Snapshot.class);
 275         assertEquals("Persisted snapshots size", 1, persistedSnapshots.size());
 276         verifySnapshot("Persisted", persistedSnapshots.get(0), currentTerm, 3, currentTerm, 7);
 277         List<ReplicatedLogEntry> unAppliedEntry = persistedSnapshots.get(0).getUnAppliedEntries();
 278         assertEquals("Persisted Snapshot getUnAppliedEntries size", 4, unAppliedEntry.size());
 279         verifyReplicatedLogEntry(unAppliedEntry.get(0), currentTerm, 4, payload4);
 280         verifyReplicatedLogEntry(unAppliedEntry.get(1), currentTerm, 5, payload5);
 281         verifyReplicatedLogEntry(unAppliedEntry.get(2), currentTerm, 6, payload6);
 282         verifyReplicatedLogEntry(unAppliedEntry.get(3), currentTerm, 7, payload7);
 283
 284         // Even though follower 2's log was behind by 5 entries and 2 snapshots were done, the leader
 285         // should not have tried to install a snapshot to catch it up because replicatedToAllIndex was also
 286         // behind. Instead of installing a snapshot the leader would've sent AppendEntries with the log entries.
 287         InstallSnapshot installSnapshot = MessageCollectorActor.getFirstMatching(follower2CollectorActor, InstallSnapshot.class);
 288         Assert.assertNull("Follower 2 received unexpected InstallSnapshot", installSnapshot);
 289
 290         // Verify follower 1's log and snapshot indexes.
 291         MessageCollectorActor.clearMessages(follower1CollectorActor);
 292         MessageCollectorActor.expectFirstMatching(follower1CollectorActor, AppendEntries.class);
 293         assertEquals("Follower 1 snapshot term", currentTerm, follower1Context.getReplicatedLog().getSnapshotTerm());
 294         assertEquals("Follower 1 snapshot index", 6, follower1Context.getReplicatedLog().getSnapshotIndex());
 295         assertEquals("Follower 1 journal log size", 1, follower1Context.getReplicatedLog().size());
 296         assertEquals("Follower 1 journal last index", 7, follower1Context.getReplicatedLog().lastIndex());
 297         assertEquals("Follower 1 commit index", 7, follower1Context.getCommitIndex());
 298         assertEquals("Follower 1 last applied", 7, follower1Context.getLastApplied());
 299         assertEquals("Follower 1 replicatedToAllIndex", 6, follower1.getReplicatedToAllIndex());
 300
 301         // Verify follower 2's log and snapshot indexes.
 302         MessageCollectorActor.clearMessages(follower2CollectorActor);
 303         MessageCollectorActor.expectFirstMatching(follower2CollectorActor, AppendEntries.class);
 304         assertEquals("Follower 2 snapshot term", currentTerm, follower2Context.getReplicatedLog().getSnapshotTerm());
 305         assertEquals("Follower 2 snapshot index", 6, follower2Context.getReplicatedLog().getSnapshotIndex());
 306         assertEquals("Follower 2 journal log size", 1, follower2Context.getReplicatedLog().size());
 307         assertEquals("Follower 2 journal last index", 7, follower2Context.getReplicatedLog().lastIndex());
 308         assertEquals("Follower 2 commit index", 7, follower2Context.getCommitIndex());
 309         assertEquals("Follower 2 last applied", 7, follower2Context.getLastApplied());
 310         assertEquals("Follower 2 replicatedToAllIndex", 6, follower2.getReplicatedToAllIndex());
 311
 312         MessageCollectorActor.clearMessages(leaderCollectorActor);
 313         MessageCollectorActor.clearMessages(follower1CollectorActor);
 314         MessageCollectorActor.clearMessages(follower2CollectorActor);
 315
 316         expSnapshotState.add(payload4);
 317         expSnapshotState.add(payload5);
 318         expSnapshotState.add(payload6);
 319         expSnapshotState.add(payload7);
 320
 321         testLog.info("testSubsequentReplicationsAndSnapshots complete");
 322     }
 323
 324     /**
 325      * Send a couple more payloads with follower 2 lagging. The last payload will have a large enough size
 326      * to trigger a leader snapshot.
 327      *
 328      * @throws Exception
 329      */
 330     private void testLeaderSnapshotTriggeredByMemoryThresholdExceeded() throws Exception {
 331         testLog.info("testLeaderSnapshotTriggeredByMemoryThresholdExceeded starting: sending 3 payloads, replicatedToAllIndex: {}",
 332                 leader.getReplicatedToAllIndex());
 333
 334         leaderActor.underlyingActor().setMockTotalMemory(1000);
 335
 336         // We'll expect a ReplicatedLogImplEntry message and an ApplyJournalEntries message added to the journal.
 337         InMemoryJournal.addWriteMessagesCompleteLatch(leaderId, 2);
 338
 339         follower2Actor.underlyingActor().startDropMessages(AppendEntries.class);
 340
 341         // Send a payload with a large relative size but not enough to trigger a snapshot.
 342         MockPayload payload8 = sendPayloadData(leaderActor, "eight", 500);
 343
 344         // Verify the leader got consensus and applies the first log entry even though follower 2 didn't respond.
 345         List<ApplyState> applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor, ApplyState.class, 1);
 346         verifyApplyState(applyStates.get(0), leaderCollectorActor, payload8.toString(), currentTerm, 8, payload8);
 347
 348         // Wait for all the ReplicatedLogImplEntry and ApplyJournalEntries messages to be added to the journal
 349         // before the snapshot so the snapshot sequence # will be higher to ensure the snapshot gets
 350         // purged from the snapshot store after subsequent snapshots.
 351         InMemoryJournal.waitForWriteMessagesComplete(leaderId);
 352
 353         // Verify a snapshot is not triggered.
 354         CaptureSnapshot captureSnapshot = MessageCollectorActor.getFirstMatching(leaderCollectorActor, CaptureSnapshot.class);
 355         Assert.assertNull("Leader received unexpected CaptureSnapshot", captureSnapshot);
 356
 357         expSnapshotState.add(payload8);
 358
 359         // Send another payload with a large enough relative size in combination with the last payload
 360         // that exceeds the memory threshold (70% * 1000 = 700) - this should do a snapshot.
 361         payload9 = sendPayloadData(leaderActor, "nine", 201);
 362
 363         // Verify the leader applies the last log entry.
 364         applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor, ApplyState.class, 2);
 365         verifyApplyState(applyStates.get(1), leaderCollectorActor, payload9.toString(), currentTerm, 9, payload9);
 366
 367         // Verify follower 1 applies each log entry.
 368         applyStates = MessageCollectorActor.expectMatching(follower1CollectorActor, ApplyState.class, 2);
 369         verifyApplyState(applyStates.get(0), null, null, currentTerm, 8, payload8);
 370         verifyApplyState(applyStates.get(1), null, null, currentTerm, 9, payload9);
 371
 372         // A snapshot should've occurred - wait for it to complete.
 373         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 374
 375         // Because the snapshot was triggered by exceeding the memory threshold the leader should've advanced
 376         // the snapshot index to the last applied index and trimmed the log even though the entries weren't
 377         // replicated to all followers.
 378         assertEquals("Leader snapshot term", currentTerm, leaderContext.getReplicatedLog().getSnapshotTerm());
 379         assertEquals("Leader snapshot index", 8, leaderContext.getReplicatedLog().getSnapshotIndex());
 380         assertEquals("Leader journal log size", 1, leaderContext.getReplicatedLog().size());
 381         assertEquals("Leader journal last index", 9, leaderContext.getReplicatedLog().lastIndex());
 382         assertEquals("Leader commit index", 9, leaderContext.getCommitIndex());
 383         assertEquals("Leader last applied", 9, leaderContext.getLastApplied());
 384         // Note: replicatedToAllIndex should not be advanced since log entries 8 and 9 haven't yet been
 385         // replicated to follower 2.
 386         assertEquals("Leader replicatedToAllIndex", 7, leader.getReplicatedToAllIndex());
 387
 388         // Verify the leader's persisted snapshot.
 389         List<Snapshot> persistedSnapshots = InMemorySnapshotStore.getSnapshots(leaderId, Snapshot.class);
 390         assertEquals("Persisted snapshots size", 1, persistedSnapshots.size());
 391         verifySnapshot("Persisted", persistedSnapshots.get(0), currentTerm, 8, currentTerm, 9);
 392         List<ReplicatedLogEntry> unAppliedEntry = persistedSnapshots.get(0).getUnAppliedEntries();
 393         assertEquals("Persisted Snapshot getUnAppliedEntries size", 1, unAppliedEntry.size());
 394         verifyReplicatedLogEntry(unAppliedEntry.get(0), currentTerm, 9, payload9);
 395
 396         testLog.info("testLeaderSnapshotTriggeredByMemoryThresholdExceeded ending");
 397     }
 398
 399     /**
 400      * Send another payload to verify another snapshot is not done since the last snapshot trimmed the
 401      * first log entry so the memory threshold should not be exceeded.
 402      *
 403      * @throws Exception
 404      */
 405     private void verifyNoSubsequentSnapshotAfterMemoryThresholdExceededSnapshot() throws Exception {
 406         ApplyState applyState;
 407         CaptureSnapshot captureSnapshot;
 408
 409         MockPayload payload10 = sendPayloadData(leaderActor, "ten");
 410
 411         // Verify the leader applies the state.
 412         applyState = MessageCollectorActor.expectFirstMatching(leaderCollectorActor, ApplyState.class);
 413         verifyApplyState(applyState, leaderCollectorActor, payload10.toString(), currentTerm, 10, payload10);
 414
 415         captureSnapshot = MessageCollectorActor.getFirstMatching(leaderCollectorActor, CaptureSnapshot.class);
 416         Assert.assertNull("Leader received unexpected CaptureSnapshot", captureSnapshot);
 417
 418         // Verify the follower 1 applies the state.
 419         applyState = MessageCollectorActor.expectFirstMatching(follower1CollectorActor, ApplyState.class);
 420         verifyApplyState(applyState, null, null, currentTerm, 10, payload10);
 421
 422         // Verify the follower 2 applies the state.
 423         applyState = MessageCollectorActor.expectFirstMatching(follower2CollectorActor, ApplyState.class);
 424         verifyApplyState(applyState, null, null, currentTerm, 10, payload10);
 425
 426         // Verify the leader's state.
 427         assertEquals("Leader snapshot term", currentTerm, leaderContext.getReplicatedLog().getSnapshotTerm());
 428         assertEquals("Leader snapshot index", 9, leaderContext.getReplicatedLog().getSnapshotIndex());
 429         assertEquals("Leader journal log size", 1, leaderContext.getReplicatedLog().size());
 430         assertEquals("Leader journal last index", 10, leaderContext.getReplicatedLog().lastIndex());
 431         assertEquals("Leader commit index", 10, leaderContext.getCommitIndex());
 432         assertEquals("Leader last applied", 10, leaderContext.getLastApplied());
 433         assertEquals("Leader replicatedToAllIndex", 9, leader.getReplicatedToAllIndex());
 434
 435         // Verify follower 1's state.
 436         assertEquals("Follower 1 snapshot term", currentTerm, follower1Context.getReplicatedLog().getSnapshotTerm());
 437         assertEquals("Follower 1 snapshot index", 9, follower1Context.getReplicatedLog().getSnapshotIndex());
 438         assertEquals("Follower 1 journal log size", 1, follower1Context.getReplicatedLog().size());
 439         assertEquals("Follower 1 journal last index", 10, follower1Context.getReplicatedLog().lastIndex());
 440         assertEquals("Follower 1 commit index", 10, follower1Context.getCommitIndex());
 441         assertEquals("Follower 1 last applied", 10, follower1Context.getLastApplied());
 442         assertEquals("Follower 1 replicatedToAllIndex", 9, follower1.getReplicatedToAllIndex());
 443
 444         // Verify follower 2's state.
 445         assertEquals("Follower 2 snapshot term", currentTerm, follower2Context.getReplicatedLog().getSnapshotTerm());
 446         assertEquals("Follower 2 snapshot index", 9, follower2Context.getReplicatedLog().getSnapshotIndex());
 447         assertEquals("Follower 2 journal log size", 1, follower2Context.getReplicatedLog().size());
 448         assertEquals("Follower 2 journal last index", 10, follower2Context.getReplicatedLog().lastIndex());
 449         assertEquals("Follower 2 commit index", 10, follower2Context.getCommitIndex());
 450         assertEquals("Follower 2 last applied", 10, follower2Context.getLastApplied());
 451         assertEquals("Follower 2 replicatedToAllIndex", 9, follower2.getReplicatedToAllIndex());
 452
 453         // Revert back to JVM total memory.
 454         leaderActor.underlyingActor().setMockTotalMemory(0);
 455
 456         MessageCollectorActor.clearMessages(leaderCollectorActor);
 457         MessageCollectorActor.clearMessages(follower1CollectorActor);
 458         MessageCollectorActor.clearMessages(follower2CollectorActor);
 459
 460         expSnapshotState.add(payload10);
 461     }
 462
 463     /**
 464      * Following a snapshot due memory threshold exceeded, resume the lagging follower and verify it receives
 465      * an install snapshot from the leader.
 466      *
 467      * @throws Exception
 468      */
 469     private void testInstallSnapshotToLaggingFollower() throws Exception {
 470         List<Snapshot> persistedSnapshots;
 471         List<ReplicatedLogEntry> unAppliedEntry;
 472         ApplyState applyState;
 473         ApplySnapshot applySnapshot;
 474         InstallSnapshot installSnapshot;
 475         InstallSnapshotReply installSnapshotReply;
 476
 477         expSnapshotState.add(payload9);
 478
 479         // Now stop dropping AppendEntries in follower 2.
 480         follower2Actor.underlyingActor().stopDropMessages(AppendEntries.class);
 481
 482         installSnapshot = MessageCollectorActor.expectFirstMatching(follower2CollectorActor, InstallSnapshot.class);
 483         assertEquals("InstallSnapshot getTerm", currentTerm, installSnapshot.getTerm());
 484         assertEquals("InstallSnapshot getLeaderId", leaderId, installSnapshot.getLeaderId());
 485         assertEquals("InstallSnapshot getChunkIndex", 1, installSnapshot.getChunkIndex());
 486         assertEquals("InstallSnapshot getTotalChunks", 1, installSnapshot.getTotalChunks());
 487         assertEquals("InstallSnapshot getLastIncludedTerm", currentTerm, installSnapshot.getLastIncludedTerm());
 488         assertEquals("InstallSnapshot getLastIncludedIndex", 8, installSnapshot.getLastIncludedIndex());
 489         //assertArrayEquals("InstallSnapshot getData", snapshot, installSnapshot.getData().toByteArray());
 490
 491         installSnapshotReply = MessageCollectorActor.expectFirstMatching(leaderCollectorActor, InstallSnapshotReply.class);
 492         assertEquals("InstallSnapshotReply getTerm", currentTerm, installSnapshotReply.getTerm());
 493         assertEquals("InstallSnapshotReply getChunkIndex", 1, installSnapshotReply.getChunkIndex());
 494         assertEquals("InstallSnapshotReply getFollowerId", follower2Id, installSnapshotReply.getFollowerId());
 495         assertEquals("InstallSnapshotReply isSuccess", true, installSnapshotReply.isSuccess());
 496
 497         // Verify follower 2 applies the snapshot.
 498         applySnapshot = MessageCollectorActor.expectFirstMatching(follower2CollectorActor, ApplySnapshot.class);
 499         verifySnapshot("Follower 2", applySnapshot.getSnapshot(), currentTerm, 8, currentTerm, 8);
 500         assertEquals("Persisted Snapshot getUnAppliedEntries size", 0, applySnapshot.getSnapshot().getUnAppliedEntries().size());
 501
 502         // Verify follower 2 only applies the second log entry (9) as the first one (8) was in the snapshot.
 503         applyState = MessageCollectorActor.expectFirstMatching(follower2CollectorActor, ApplyState.class);
 504         verifyApplyState(applyState, null, null, currentTerm, 9, payload9);
 505
 506         // Wait for the snapshot to complete.
 507         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 508
 509         // Ensure there's at least 1 more heartbeat.
 510         MessageCollectorActor.clearMessages(leaderCollectorActor);
 511         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, AppendEntriesReply.class);
 512
 513         // The leader should now have performed fake snapshots to advance the snapshot index and to trim
 514         // the log. In addition replicatedToAllIndex should've advanced.
 515         assertEquals("Leader snapshot term", currentTerm, leaderContext.getReplicatedLog().getSnapshotTerm());
 516         assertEquals("Leader snapshot index", 8, leaderContext.getReplicatedLog().getSnapshotIndex());
 517         assertEquals("Leader journal log size", 1, leaderContext.getReplicatedLog().size());
 518         assertEquals("Leader commit index", 9, leaderContext.getCommitIndex());
 519         assertEquals("Leader last applied", 9, leaderContext.getLastApplied());
 520         assertEquals("Leader replicatedToAllIndex", 8, leader.getReplicatedToAllIndex());
 521
 522         // Verify the leader's persisted snapshot. The previous snapshot (currently) won't be deleted from
 523         // the snapshot store because the second snapshot was initiated by the follower install snapshot and
 524         // not because the batch count was reached so the persisted journal sequence number wasn't advanced
 525         // far enough to cause the previous snapshot to be deleted. This is because
 526         // RaftActor#trimPersistentData subtracts the snapshotBatchCount from the snapshot's sequence number.
 527         // This is OK - the next snapshot should delete it. In production, even if the system restarted
 528         // before another snapshot, they would both get applied which wouldn't hurt anything.
 529         persistedSnapshots = InMemorySnapshotStore.getSnapshots(leaderId, Snapshot.class);
 530         Assert.assertTrue("Expected at least 1 persisted snapshots", persistedSnapshots.size() > 0);
 531         Snapshot persistedSnapshot = persistedSnapshots.get(persistedSnapshots.size() - 1);
 532         verifySnapshot("Persisted", persistedSnapshot, currentTerm, 9, currentTerm, 9);
 533         unAppliedEntry = persistedSnapshot.getUnAppliedEntries();
 534         assertEquals("Persisted Snapshot getUnAppliedEntries size", 0, unAppliedEntry.size());
 535
 536         MessageCollectorActor.clearMessages(leaderCollectorActor);
 537         MessageCollectorActor.clearMessages(follower1CollectorActor);
 538         MessageCollectorActor.clearMessages(follower2CollectorActor);
 539     }
 540
 541     /**
 542      * Do another round of payloads and snapshot to verify replicatedToAllIndex gets back on track and
 543      * snapshots works as expected after doing a follower snapshot. In this step we don't lag a follower.
 544      * @throws Exception
 545      */
 546     private void testFinalReplicationsAndSnapshot() throws Exception {
 547         List<ApplyState> applyStates;
 548         ApplyState applyState;
 549
 550         testLog.info("testFinalReplicationsAndSnapshot starting: replicatedToAllIndex: {}", leader.getReplicatedToAllIndex());
 551
 552         // Send another payload - a snapshot should occur.
 553         payload11 = sendPayloadData(leaderActor, "eleven");
 554
 555         // Wait for the snapshot to complete.
 556         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, SaveSnapshotSuccess.class);
 557
 558         applyState = MessageCollectorActor.expectFirstMatching(leaderCollectorActor, ApplyState.class);
 559         verifyApplyState(applyState, leaderCollectorActor, payload11.toString(), currentTerm, 11, payload11);
 560
 561         // Verify the leader's last persisted snapshot (previous ones may not be purged yet).
 562         List<Snapshot> persistedSnapshots = InMemorySnapshotStore.getSnapshots(leaderId, Snapshot.class);
 563         Snapshot persistedSnapshot = persistedSnapshots.get(persistedSnapshots.size() - 1);
 564         verifySnapshot("Persisted", persistedSnapshot, currentTerm, 10, currentTerm, 11);
 565         List<ReplicatedLogEntry> unAppliedEntry = persistedSnapshot.getUnAppliedEntries();
 566         assertEquals("Persisted Snapshot getUnAppliedEntries size", 1, unAppliedEntry.size());
 567         verifyReplicatedLogEntry(unAppliedEntry.get(0), currentTerm, 11, payload11);
 568
 569         // Send a couple more payloads.
 570         payload12 = sendPayloadData(leaderActor, "twelve");
 571         payload13 = sendPayloadData(leaderActor, "thirteen");
 572
 573         // Verify the leader applies the 2 log entries.
 574         applyStates = MessageCollectorActor.expectMatching(leaderCollectorActor, ApplyState.class, 3);
 575         verifyApplyState(applyStates.get(1), leaderCollectorActor, payload12.toString(), currentTerm, 12, payload12);
 576         verifyApplyState(applyStates.get(2), leaderCollectorActor, payload13.toString(), currentTerm, 13, payload13);
 577
 578         // Verify the leader applies a log entry for at least the last entry index.
 579         verifyApplyJournalEntries(leaderCollectorActor, 13);
 580
 581         // Ensure there's at least 1 more heartbeat to trim the log.
 582         MessageCollectorActor.clearMessages(leaderCollectorActor);
 583         MessageCollectorActor.expectFirstMatching(leaderCollectorActor, AppendEntriesReply.class);
 584
 585         // Verify the leader's final snapshot index et al.
 586         assertEquals("Leader snapshot term", currentTerm, leaderContext.getReplicatedLog().getSnapshotTerm());
 587         assertEquals("Leader snapshot index", 12, leaderContext.getReplicatedLog().getSnapshotIndex());
 588         assertEquals("Leader journal log size", 1, leaderContext.getReplicatedLog().size());
 589         assertEquals("Leader journal last index", 13, leaderContext.getReplicatedLog().lastIndex());
 590         assertEquals("Leader commit index", 13, leaderContext.getCommitIndex());
 591         assertEquals("Leader last applied", 13, leaderContext.getLastApplied());
 592         assertEquals("Leader replicatedToAllIndex", 12, leader.getReplicatedToAllIndex());
 593
 594         InMemoryJournal.dumpJournal(leaderId);
 595
 596         // Verify the leaders's persisted journal log - should only contain the last 2 ReplicatedLogEntries
 597         // added after the snapshot as the persisted journal should've been purged to the snapshot
 598         // sequence number.
 599         verifyPersistedJournal(leaderId, Arrays.asList(new ReplicatedLogImplEntry(12, currentTerm, payload12),
 600                 new ReplicatedLogImplEntry(13, currentTerm, payload13)));
 601
 602         // Verify the leaders's persisted journal contains an ApplyJournalEntries for at least the last entry index.
 603         List<ApplyJournalEntries> persistedApplyJournalEntries = InMemoryJournal.get(leaderId, ApplyJournalEntries.class);
 604         boolean found = false;
 605         for(ApplyJournalEntries entry: persistedApplyJournalEntries) {
 606             if(entry.getToIndex() == 13) {
 607                 found = true;
 608                 break;
 609             }
 610         }
 611
 612         Assert.assertTrue(String.format("ApplyJournalEntries with index %d not found in leader's persisted journal", 13), found);
 613
 614         // Verify follower 1 applies the 2 log entries.
 615         applyStates = MessageCollectorActor.expectMatching(follower1CollectorActor, ApplyState.class, 3);
 616         verifyApplyState(applyStates.get(0), null, null, currentTerm, 11, payload11);
 617         verifyApplyState(applyStates.get(1), null, null, currentTerm, 12, payload12);
 618         verifyApplyState(applyStates.get(2), null, null, currentTerm, 13, payload13);
 619
 620         // Verify follower 1's log state.
 621         assertEquals("Follower 1 snapshot term", currentTerm, follower1Context.getReplicatedLog().getSnapshotTerm());
 622         assertEquals("Follower 1 snapshot index", 12, follower1Context.getReplicatedLog().getSnapshotIndex());
 623         assertEquals("Follower 1 journal log size", 1, follower1Context.getReplicatedLog().size());
 624         assertEquals("Follower 1 journal last index", 13, follower1Context.getReplicatedLog().lastIndex());
 625         assertEquals("Follower 1 commit index", 13, follower1Context.getCommitIndex());
 626         assertEquals("Follower 1 last applied", 13, follower1Context.getLastApplied());
 627         assertEquals("Follower 1 replicatedToAllIndex", 12, follower1.getReplicatedToAllIndex());
 628
 629         // Verify follower 2 applies the 2 log entries.
 630         applyStates = MessageCollectorActor.expectMatching(follower2CollectorActor, ApplyState.class, 3);
 631         verifyApplyState(applyStates.get(0), null, null, currentTerm, 11, payload11);
 632         verifyApplyState(applyStates.get(1), null, null, currentTerm, 12, payload12);
 633         verifyApplyState(applyStates.get(2), null, null, currentTerm, 13, payload13);
 634
 635         // Verify follower 2's log state.
 636         assertEquals("Follower 2 snapshot term", currentTerm, follower2Context.getReplicatedLog().getSnapshotTerm());
 637         assertEquals("Follower 2 snapshot index", 12, follower2Context.getReplicatedLog().getSnapshotIndex());
 638         assertEquals("Follower 2 journal log size", 1, follower2Context.getReplicatedLog().size());
 639         assertEquals("Follower 2 journal last index", 13, follower2Context.getReplicatedLog().lastIndex());
 640         assertEquals("Follower 2 commit index", 13, follower2Context.getCommitIndex());
 641         assertEquals("Follower 2 last applied", 13, follower2Context.getLastApplied());
 642         assertEquals("Follower 2 replicatedToAllIndex", 12, follower2.getReplicatedToAllIndex());
 643
 644         testLog.info("testFinalReplicationsAndSnapshot ending");
 645     }
 646
 647     /**
 648      * Kill the leader actor, reinstate it and verify the recovered journal.
 649      */
 650     private void testLeaderReinstatement() {
 651         testLog.info("testLeaderReinstatement starting");
 652
 653         killActor(leaderActor);
 654
 655         leaderActor = newTestRaftActor(leaderId, peerAddresses, leaderConfigParams);
 656
 657         leaderActor.underlyingActor().startDropMessages(RequestVoteReply.class);
 658
 659         leaderContext = leaderActor.underlyingActor().getRaftActorContext();
 660
 661         leaderActor.underlyingActor().waitForRecoveryComplete();
 662
 663         assertEquals("Leader snapshot term", currentTerm, leaderContext.getReplicatedLog().getSnapshotTerm());
 664         assertEquals("Leader snapshot index", 10, leaderContext.getReplicatedLog().getSnapshotIndex());
 665         assertEquals("Leader journal log size", 3, leaderContext.getReplicatedLog().size());
 666         assertEquals("Leader journal last index", 13, leaderContext.getReplicatedLog().lastIndex());
 667         assertEquals("Leader commit index", 13, leaderContext.getCommitIndex());
 668         assertEquals("Leader last applied", 13, leaderContext.getLastApplied());
 669         verifyReplicatedLogEntry(leaderContext.getReplicatedLog().get(11), currentTerm, 11, payload11);
 670         verifyReplicatedLogEntry(leaderContext.getReplicatedLog().get(12), currentTerm, 12, payload12);
 671         verifyReplicatedLogEntry(leaderContext.getReplicatedLog().get(13), currentTerm, 13, payload13);
 672
 673         testLog.info("testLeaderReinstatement ending");
 674     }
 675 }