opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/SnapshotManager.java

   1 /*
   2  * Copyright (c) 2014 Cisco 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
   9 package org.opendaylight.controller.cluster.raft;
  10
  11 import akka.persistence.SnapshotSelectionCriteria;
  12 import com.google.common.annotations.VisibleForTesting;
  13 import java.util.List;
  14 import java.util.function.Consumer;
  15 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  16 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  17 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
  18 import org.opendaylight.controller.cluster.raft.base.messages.SnapshotComplete;
  19 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  20 import org.slf4j.Logger;
  21
  22 /**
  23  * Manages the capturing of snapshots for a RaftActor.
  24  *
  25  * @author Moiz Raja
  26  * @author Thomas Pantelis
  27  */
  28 public class SnapshotManager implements SnapshotState {
  29
  30     @SuppressWarnings("checkstyle:MemberName")
  31     private final SnapshotState IDLE = new Idle();
  32
  33     @SuppressWarnings({"checkstyle:MemberName", "checkstyle:AbbreviationAsWordInName"})
  34     private final SnapshotState PERSISTING = new Persisting();
  35
  36     @SuppressWarnings({"checkstyle:MemberName", "checkstyle:AbbreviationAsWordInName"})
  37     private final SnapshotState CREATING = new Creating();
  38
  39     private final Logger log;
  40     private final RaftActorContext context;
  41     private final LastAppliedTermInformationReader lastAppliedTermInformationReader =
  42             new LastAppliedTermInformationReader();
  43     private final ReplicatedToAllTermInformationReader replicatedToAllTermInformationReader =
  44             new ReplicatedToAllTermInformationReader();
  45
  46
  47     private SnapshotState currentState = IDLE;
  48     private CaptureSnapshot captureSnapshot;
  49     private long lastSequenceNumber = -1;
  50
  51     private Runnable createSnapshotProcedure;
  52
  53     private ApplySnapshot applySnapshot;
  54     private Consumer<byte[]> applySnapshotProcedure;
  55
  56     /**
  57      * Constructs an instance.
  58      *
  59      * @param context the RaftActorContext
  60      * @param logger the Logger
  61      */
  62     public SnapshotManager(RaftActorContext context, Logger logger) {
  63         this.context = context;
  64         this.log = logger;
  65     }
  66
  67     public boolean isApplying() {
  68         return applySnapshot != null;
  69     }
  70
  71     @Override
  72     public boolean isCapturing() {
  73         return currentState.isCapturing();
  74     }
  75
  76     @Override
  77     public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
  78         return currentState.captureToInstall(lastLogEntry, replicatedToAllIndex, targetFollower);
  79     }
  80
  81     @Override
  82     public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
  83         return currentState.capture(lastLogEntry, replicatedToAllIndex);
  84     }
  85
  86     @Override
  87     public void apply(ApplySnapshot snapshot) {
  88         currentState.apply(snapshot);
  89     }
  90
  91     @Override
  92     public void persist(final byte[] snapshotBytes, final long totalMemory) {
  93         currentState.persist(snapshotBytes, totalMemory);
  94     }
  95
  96     @Override
  97     public void commit(final long sequenceNumber, long timeStamp) {
  98         currentState.commit(sequenceNumber, timeStamp);
  99     }
 100
 101     @Override
 102     public void rollback() {
 103         currentState.rollback();
 104     }
 105
 106     @Override
 107     public long trimLog(final long desiredTrimIndex) {
 108         return currentState.trimLog(desiredTrimIndex);
 109     }
 110
 111     public void setCreateSnapshotRunnable(Runnable createSnapshotProcedure) {
 112         this.createSnapshotProcedure = createSnapshotProcedure;
 113     }
 114
 115     public void setApplySnapshotConsumer(Consumer<byte[]> applySnapshotProcedure) {
 116         this.applySnapshotProcedure = applySnapshotProcedure;
 117     }
 118
 119     public long getLastSequenceNumber() {
 120         return lastSequenceNumber;
 121     }
 122
 123     @VisibleForTesting
 124     public CaptureSnapshot getCaptureSnapshot() {
 125         return captureSnapshot;
 126     }
 127
 128     private boolean hasFollowers() {
 129         return context.hasFollowers();
 130     }
 131
 132     private String persistenceId() {
 133         return context.getId();
 134     }
 135
 136     /**
 137      * Constructs a CaptureSnapshot instance.
 138      *
 139      * @param lastLogEntry the last log entry for the snapshot.
 140      * @param replicatedToAllIndex the index of the last entry replicated to all followers.
 141      * @param installSnapshotInitiated true if snapshot is initiated to install on a follower.
 142      * @return a new CaptureSnapshot instance.
 143      */
 144     public CaptureSnapshot newCaptureSnapshot(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex,
 145             boolean installSnapshotInitiated) {
 146         TermInformationReader lastAppliedTermInfoReader =
 147                 lastAppliedTermInformationReader.init(context.getReplicatedLog(), context.getLastApplied(),
 148                         lastLogEntry, hasFollowers());
 149
 150         long lastAppliedIndex = lastAppliedTermInfoReader.getIndex();
 151         long lastAppliedTerm = lastAppliedTermInfoReader.getTerm();
 152
 153         TermInformationReader replicatedToAllTermInfoReader =
 154                 replicatedToAllTermInformationReader.init(context.getReplicatedLog(), replicatedToAllIndex);
 155
 156         long newReplicatedToAllIndex = replicatedToAllTermInfoReader.getIndex();
 157         long newReplicatedToAllTerm = replicatedToAllTermInfoReader.getTerm();
 158
 159         List<ReplicatedLogEntry> unAppliedEntries = context.getReplicatedLog().getFrom(lastAppliedIndex + 1);
 160
 161         long lastLogEntryIndex = lastAppliedIndex;
 162         long lastLogEntryTerm = lastAppliedTerm;
 163         if (lastLogEntry != null) {
 164             lastLogEntryIndex = lastLogEntry.getIndex();
 165             lastLogEntryTerm = lastLogEntry.getTerm();
 166         } else {
 167             log.debug("{}: Capturing Snapshot : lastLogEntry is null. Using lastAppliedIndex {} and "
 168                     + "lastAppliedTerm {} instead.", persistenceId(), lastAppliedIndex, lastAppliedTerm);
 169         }
 170
 171         return new CaptureSnapshot(lastLogEntryIndex, lastLogEntryTerm, lastAppliedIndex, lastAppliedTerm,
 172                 newReplicatedToAllIndex, newReplicatedToAllTerm, unAppliedEntries, installSnapshotInitiated);
 173     }
 174
 175     private class AbstractSnapshotState implements SnapshotState {
 176
 177         @Override
 178         public boolean isCapturing() {
 179             return true;
 180         }
 181
 182         @Override
 183         public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
 184             log.debug("capture should not be called in state {}", this);
 185             return false;
 186         }
 187
 188         @Override
 189         public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex,
 190                 String targetFollower) {
 191             log.debug("captureToInstall should not be called in state {}", this);
 192             return false;
 193         }
 194
 195         @Override
 196         public void apply(ApplySnapshot snapshot) {
 197             log.debug("apply should not be called in state {}", this);
 198         }
 199
 200         @Override
 201         public void persist(final byte[] snapshotBytes, final long totalMemory) {
 202             log.debug("persist should not be called in state {}", this);
 203         }
 204
 205         @Override
 206         public void commit(final long sequenceNumber, long timeStamp) {
 207             log.debug("commit should not be called in state {}", this);
 208         }
 209
 210         @Override
 211         public void rollback() {
 212             log.debug("rollback should not be called in state {}", this);
 213         }
 214
 215         @Override
 216         public long trimLog(final long desiredTrimIndex) {
 217             log.debug("trimLog should not be called in state {}", this);
 218             return -1;
 219         }
 220
 221         protected long doTrimLog(final long desiredTrimIndex) {
 222             //  we would want to keep the lastApplied as its used while capturing snapshots
 223             long lastApplied = context.getLastApplied();
 224             long tempMin = Math.min(desiredTrimIndex, lastApplied > -1 ? lastApplied - 1 : -1);
 225
 226             if (log.isTraceEnabled()) {
 227                 log.trace("{}: performSnapshotWithoutCapture: desiredTrimIndex: {}, lastApplied: {}, tempMin: {}",
 228                         persistenceId(), desiredTrimIndex, lastApplied, tempMin);
 229             }
 230
 231             if (tempMin > -1 && context.getReplicatedLog().isPresent(tempMin)) {
 232                 log.debug("{}: fakeSnapshot purging log to {} for term {}", persistenceId(), tempMin,
 233                         context.getTermInformation().getCurrentTerm());
 234
 235                 //use the term of the temp-min, since we check for isPresent, entry will not be null
 236                 ReplicatedLogEntry entry = context.getReplicatedLog().get(tempMin);
 237                 context.getReplicatedLog().snapshotPreCommit(tempMin, entry.getTerm());
 238                 context.getReplicatedLog().snapshotCommit();
 239                 return tempMin;
 240             }
 241
 242             final RaftActorBehavior currentBehavior = context.getCurrentBehavior();
 243             if (tempMin > currentBehavior.getReplicatedToAllIndex()) {
 244                 // It's possible a follower was lagging and an install snapshot advanced its match index past
 245                 // the current replicatedToAllIndex. Since the follower is now caught up we should advance the
 246                 // replicatedToAllIndex (to tempMin). The fact that tempMin wasn't found in the log is likely
 247                 // due to a previous snapshot triggered by the memory threshold exceeded, in that case we
 248                 // trim the log to the last applied index even if previous entries weren't replicated to all followers.
 249                 currentBehavior.setReplicatedToAllIndex(tempMin);
 250             }
 251             return -1;
 252         }
 253     }
 254
 255     private class Idle extends AbstractSnapshotState {
 256
 257         @Override
 258         public boolean isCapturing() {
 259             return false;
 260         }
 261
 262         @SuppressWarnings("checkstyle:IllegalCatch")
 263         private boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
 264             captureSnapshot = newCaptureSnapshot(lastLogEntry, replicatedToAllIndex, targetFollower != null);
 265
 266             if (captureSnapshot.isInstallSnapshotInitiated()) {
 267                 log.info("{}: Initiating snapshot capture {} to install on {}",
 268                         persistenceId(), captureSnapshot, targetFollower);
 269             } else {
 270                 log.info("{}: Initiating snapshot capture {}", persistenceId(), captureSnapshot);
 271             }
 272
 273             lastSequenceNumber = context.getPersistenceProvider().getLastSequenceNumber();
 274
 275             log.debug("{}: lastSequenceNumber prior to capture: {}", persistenceId(), lastSequenceNumber);
 276
 277             SnapshotManager.this.currentState = CREATING;
 278
 279             try {
 280                 createSnapshotProcedure.run();
 281             } catch (Exception e) {
 282                 SnapshotManager.this.currentState = IDLE;
 283                 log.error("Error creating snapshot", e);
 284                 return false;
 285             }
 286
 287             return true;
 288         }
 289
 290         @Override
 291         public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
 292             return capture(lastLogEntry, replicatedToAllIndex, null);
 293         }
 294
 295         @Override
 296         public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex,
 297                 String targetFollower) {
 298             return capture(lastLogEntry, replicatedToAllIndex, targetFollower);
 299         }
 300
 301         @Override
 302         public void apply(ApplySnapshot toApply) {
 303             SnapshotManager.this.applySnapshot = toApply;
 304
 305             lastSequenceNumber = context.getPersistenceProvider().getLastSequenceNumber();
 306
 307             log.debug("lastSequenceNumber prior to persisting applied snapshot: {}", lastSequenceNumber);
 308
 309             context.getPersistenceProvider().saveSnapshot(toApply.getSnapshot());
 310
 311             SnapshotManager.this.currentState = PERSISTING;
 312         }
 313
 314         @Override
 315         public String toString() {
 316             return "Idle";
 317         }
 318
 319         @Override
 320         public long trimLog(final long desiredTrimIndex) {
 321             return doTrimLog(desiredTrimIndex);
 322         }
 323     }
 324
 325     private class Creating extends AbstractSnapshotState {
 326
 327         @Override
 328         public void persist(final byte[] snapshotBytes, final long totalMemory) {
 329             // create a snapshot object from the state provided and save it
 330             // when snapshot is saved async, SaveSnapshotSuccess is raised.
 331
 332             Snapshot snapshot = Snapshot.create(snapshotBytes,
 333                     captureSnapshot.getUnAppliedEntries(),
 334                     captureSnapshot.getLastIndex(), captureSnapshot.getLastTerm(),
 335                     captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm(),
 336                     context.getTermInformation().getCurrentTerm(),
 337                     context.getTermInformation().getVotedFor(), context.getPeerServerInfo(true));
 338
 339             context.getPersistenceProvider().saveSnapshot(snapshot);
 340
 341             log.info("{}: Persisting of snapshot done: {}", persistenceId(), snapshot);
 342
 343             long dataThreshold = totalMemory * context.getConfigParams().getSnapshotDataThresholdPercentage() / 100;
 344             boolean dataSizeThresholdExceeded = context.getReplicatedLog().dataSize() > dataThreshold;
 345
 346             boolean logSizeExceededSnapshotBatchCount =
 347                     context.getReplicatedLog().size() >= context.getConfigParams().getSnapshotBatchCount();
 348
 349             final RaftActorBehavior currentBehavior = context.getCurrentBehavior();
 350             if (dataSizeThresholdExceeded || logSizeExceededSnapshotBatchCount) {
 351                 if (log.isDebugEnabled()) {
 352                     if (dataSizeThresholdExceeded) {
 353                         log.debug("{}: log data size {} exceeds the memory threshold {} - doing snapshotPreCommit "
 354                                 + "with index {}", context.getId(), context.getReplicatedLog().dataSize(),
 355                                 dataThreshold, captureSnapshot.getLastAppliedIndex());
 356                     } else {
 357                         log.debug("{}: log size {} exceeds the snapshot batch count {} - doing snapshotPreCommit with "
 358                                 + "index {}", context.getId(), context.getReplicatedLog().size(),
 359                                 context.getConfigParams().getSnapshotBatchCount(),
 360                                 captureSnapshot.getLastAppliedIndex());
 361                     }
 362                 }
 363
 364                 // We either exceeded the memory threshold or the log size exceeded the snapshot batch
 365                 // count so, to keep the log memory footprint in check, clear the log based on lastApplied.
 366                 // This could/should only happen if one of the followers is down as normally we keep
 367                 // removing from the log as entries are replicated to all.
 368                 context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getLastAppliedIndex(),
 369                         captureSnapshot.getLastAppliedTerm());
 370
 371                 // Don't reset replicatedToAllIndex to -1 as this may prevent us from trimming the log after an
 372                 // install snapshot to a follower.
 373                 if (captureSnapshot.getReplicatedToAllIndex() >= 0) {
 374                     currentBehavior.setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
 375                 }
 376
 377             } else if (captureSnapshot.getReplicatedToAllIndex() != -1) {
 378                 // clear the log based on replicatedToAllIndex
 379                 context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getReplicatedToAllIndex(),
 380                         captureSnapshot.getReplicatedToAllTerm());
 381
 382                 currentBehavior.setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
 383             } else {
 384                 // The replicatedToAllIndex was not found in the log
 385                 // This means that replicatedToAllIndex never moved beyond -1 or that it is already in the snapshot.
 386                 // In this scenario we may need to save the snapshot to the akka persistence
 387                 // snapshot for recovery but we do not need to do the replicated log trimming.
 388                 context.getReplicatedLog().snapshotPreCommit(context.getReplicatedLog().getSnapshotIndex(),
 389                         context.getReplicatedLog().getSnapshotTerm());
 390             }
 391
 392             log.info("{}: Removed in-memory snapshotted entries, adjusted snaphsotIndex: {} and term: {}",
 393                     context.getId(), context.getReplicatedLog().getSnapshotIndex(),
 394                     context.getReplicatedLog().getSnapshotTerm());
 395
 396             if (context.getId().equals(currentBehavior.getLeaderId())
 397                     && captureSnapshot.isInstallSnapshotInitiated()) {
 398                 // this would be call straight to the leader and won't initiate in serialization
 399                 currentBehavior.handleMessage(context.getActor(), new SendInstallSnapshot(snapshot));
 400             }
 401
 402             captureSnapshot = null;
 403             SnapshotManager.this.currentState = PERSISTING;
 404         }
 405
 406         @Override
 407         public String toString() {
 408             return "Creating";
 409         }
 410
 411     }
 412
 413     private class Persisting extends AbstractSnapshotState {
 414
 415         @Override
 416         @SuppressWarnings("checkstyle:IllegalCatch")
 417         public void commit(final long sequenceNumber, long timeStamp) {
 418             log.debug("{}: Snapshot success -  sequence number: {}", persistenceId(), sequenceNumber);
 419
 420             if (applySnapshot != null) {
 421                 try {
 422                     Snapshot snapshot = applySnapshot.getSnapshot();
 423
 424                     //clears the followers log, sets the snapshot index to ensure adjusted-index works
 425                     context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context));
 426                     context.setLastApplied(snapshot.getLastAppliedIndex());
 427                     context.setCommitIndex(snapshot.getLastAppliedIndex());
 428                     context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor());
 429
 430                     if (snapshot.getServerConfiguration() != null) {
 431                         context.updatePeerIds(snapshot.getServerConfiguration());
 432                     }
 433
 434                     if (snapshot.getState().length > 0 ) {
 435                         applySnapshotProcedure.accept(snapshot.getState());
 436                     }
 437
 438                     applySnapshot.getCallback().onSuccess();
 439                 } catch (Exception e) {
 440                     log.error("{}: Error applying snapshot", context.getId(), e);
 441                 }
 442             } else {
 443                 context.getReplicatedLog().snapshotCommit();
 444             }
 445
 446             context.getPersistenceProvider().deleteSnapshots(new SnapshotSelectionCriteria(sequenceNumber,
 447                     timeStamp - 1, 0L, 0L));
 448
 449             context.getPersistenceProvider().deleteMessages(lastSequenceNumber);
 450
 451             snapshotComplete();
 452         }
 453
 454         @Override
 455         public void rollback() {
 456             // Nothing to rollback if we're applying a snapshot from the leader.
 457             if (applySnapshot == null) {
 458                 context.getReplicatedLog().snapshotRollback();
 459
 460                 log.info("{}: Replicated Log rolled back. Snapshot will be attempted in the next cycle."
 461                         + "snapshotIndex:{}, snapshotTerm:{}, log-size:{}", persistenceId(),
 462                         context.getReplicatedLog().getSnapshotIndex(),
 463                         context.getReplicatedLog().getSnapshotTerm(),
 464                         context.getReplicatedLog().size());
 465             } else {
 466                 applySnapshot.getCallback().onFailure();
 467             }
 468
 469             snapshotComplete();
 470         }
 471
 472         private void snapshotComplete() {
 473             lastSequenceNumber = -1;
 474             applySnapshot = null;
 475             SnapshotManager.this.currentState = IDLE;
 476
 477             context.getActor().tell(SnapshotComplete.INSTANCE, context.getActor());
 478         }
 479
 480         @Override
 481         public String toString() {
 482             return "Persisting";
 483         }
 484
 485     }
 486
 487     private interface TermInformationReader {
 488         long getIndex();
 489
 490         long getTerm();
 491     }
 492
 493     static class LastAppliedTermInformationReader implements TermInformationReader {
 494         private long index;
 495         private long term;
 496
 497         LastAppliedTermInformationReader init(ReplicatedLog log, long originalIndex, ReplicatedLogEntry lastLogEntry,
 498                 boolean hasFollowers) {
 499             ReplicatedLogEntry entry = log.get(originalIndex);
 500             this.index = -1L;
 501             this.term = -1L;
 502             if (!hasFollowers) {
 503                 if (lastLogEntry != null) {
 504                     // since we have persisted the last-log-entry to persistent journal before the capture,
 505                     // we would want to snapshot from this entry.
 506                     index = lastLogEntry.getIndex();
 507                     term = lastLogEntry.getTerm();
 508                 }
 509             } else if (entry != null) {
 510                 index = entry.getIndex();
 511                 term = entry.getTerm();
 512             } else if (log.getSnapshotIndex() > -1) {
 513                 index = log.getSnapshotIndex();
 514                 term = log.getSnapshotTerm();
 515             }
 516             return this;
 517         }
 518
 519         @Override
 520         public long getIndex() {
 521             return this.index;
 522         }
 523
 524         @Override
 525         public long getTerm() {
 526             return this.term;
 527         }
 528     }
 529
 530     private static class ReplicatedToAllTermInformationReader implements TermInformationReader {
 531         private long index;
 532         private long term;
 533
 534         ReplicatedToAllTermInformationReader init(ReplicatedLog log, long originalIndex) {
 535             ReplicatedLogEntry entry = log.get(originalIndex);
 536             this.index = -1L;
 537             this.term = -1L;
 538
 539             if (entry != null) {
 540                 index = entry.getIndex();
 541                 term = entry.getTerm();
 542             }
 543
 544             return this;
 545         }
 546
 547         @Override
 548         public long getIndex() {
 549             return this.index;
 550         }
 551
 552         @Override
 553         public long getTerm() {
 554             return this.term;
 555         }
 556     }
 557 }