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