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 package org.opendaylight.controller.cluster.raft;
   9
  10 import akka.persistence.SnapshotSelectionCriteria;
  11 import com.google.common.annotations.VisibleForTesting;
  12 import com.google.common.io.ByteSource;
  13 import java.io.IOException;
  14 import java.io.OutputStream;
  15 import java.util.List;
  16 import java.util.Optional;
  17 import java.util.function.Consumer;
  18 import org.eclipse.jdt.annotation.NonNull;
  19 import org.opendaylight.controller.cluster.io.FileBackedOutputStream;
  20 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  21 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  22 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
  23 import org.opendaylight.controller.cluster.raft.base.messages.SnapshotComplete;
  24 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  25 import org.opendaylight.controller.cluster.raft.persisted.EmptyState;
  26 import org.opendaylight.controller.cluster.raft.persisted.Snapshot;
  27 import org.slf4j.Logger;
  28
  29 /**
  30  * Manages the capturing of snapshots for a RaftActor.
  31  *
  32  * @author Moiz Raja
  33  * @author Thomas Pantelis
  34  */
  35 public class SnapshotManager implements SnapshotState {
  36
  37     @SuppressWarnings("checkstyle:MemberName")
  38     private final SnapshotState IDLE = new Idle();
  39
  40     @SuppressWarnings({"checkstyle:MemberName", "checkstyle:AbbreviationAsWordInName"})
  41     private final SnapshotState PERSISTING = new Persisting();
  42
  43     @SuppressWarnings({"checkstyle:MemberName", "checkstyle:AbbreviationAsWordInName"})
  44     private final SnapshotState CREATING = new Creating();
  45
  46     private final Logger log;
  47     private final RaftActorContext context;
  48     private final LastAppliedTermInformationReader lastAppliedTermInformationReader =
  49             new LastAppliedTermInformationReader();
  50     private final ReplicatedToAllTermInformationReader replicatedToAllTermInformationReader =
  51             new ReplicatedToAllTermInformationReader();
  52
  53
  54     private SnapshotState currentState = IDLE;
  55     private CaptureSnapshot captureSnapshot;
  56     private long lastSequenceNumber = -1;
  57
  58     private Consumer<Optional<OutputStream>> createSnapshotProcedure = null;
  59
  60     private ApplySnapshot applySnapshot;
  61     private RaftActorSnapshotCohort snapshotCohort = NoopRaftActorSnapshotCohort.INSTANCE;
  62
  63     /**
  64      * Constructs an instance.
  65      *
  66      * @param context the RaftActorContext
  67      * @param logger the Logger
  68      */
  69     public SnapshotManager(final RaftActorContext context, final Logger logger) {
  70         this.context = context;
  71         this.log = logger;
  72     }
  73
  74     public boolean isApplying() {
  75         return applySnapshot != null;
  76     }
  77
  78     @Override
  79     public boolean isCapturing() {
  80         return currentState.isCapturing();
  81     }
  82
  83     @Override
  84     public boolean captureToInstall(final ReplicatedLogEntry lastLogEntry, final long replicatedToAllIndex,
  85             final String targetFollower) {
  86         return currentState.captureToInstall(lastLogEntry, replicatedToAllIndex, targetFollower);
  87     }
  88
  89     @Override
  90     public boolean capture(final ReplicatedLogEntry lastLogEntry, final long replicatedToAllIndex) {
  91         return currentState.capture(lastLogEntry, replicatedToAllIndex);
  92     }
  93
  94     @Override
  95     public void apply(final ApplySnapshot snapshot) {
  96         currentState.apply(snapshot);
  97     }
  98
  99     @Override
 100     public void persist(final Snapshot.State state, final Optional<OutputStream> installSnapshotStream,
 101             final long totalMemory) {
 102         currentState.persist(state, installSnapshotStream, totalMemory);
 103     }
 104
 105     @Override
 106     public void commit(final long sequenceNumber, final long timeStamp) {
 107         currentState.commit(sequenceNumber, timeStamp);
 108     }
 109
 110     @Override
 111     public void rollback() {
 112         currentState.rollback();
 113     }
 114
 115     @Override
 116     public long trimLog(final long desiredTrimIndex) {
 117         return currentState.trimLog(desiredTrimIndex);
 118     }
 119
 120     @SuppressWarnings("checkstyle:hiddenField")
 121     void setCreateSnapshotConsumer(final Consumer<Optional<OutputStream>> createSnapshotProcedure) {
 122         this.createSnapshotProcedure = createSnapshotProcedure;
 123     }
 124
 125     void setSnapshotCohort(final RaftActorSnapshotCohort snapshotCohort) {
 126         this.snapshotCohort = snapshotCohort;
 127     }
 128
 129     public Snapshot.@NonNull State convertSnapshot(final ByteSource snapshotBytes) throws IOException {
 130         return snapshotCohort.deserializeSnapshot(snapshotBytes);
 131     }
 132
 133     public long getLastSequenceNumber() {
 134         return lastSequenceNumber;
 135     }
 136
 137     @VisibleForTesting
 138     public CaptureSnapshot getCaptureSnapshot() {
 139         return captureSnapshot;
 140     }
 141
 142     private boolean hasFollowers() {
 143         return context.hasFollowers();
 144     }
 145
 146     private String persistenceId() {
 147         return context.getId();
 148     }
 149
 150     /**
 151      * Constructs a CaptureSnapshot instance.
 152      *
 153      * @param lastLogEntry the last log entry for the snapshot.
 154      * @param replicatedToAllIndex the index of the last entry replicated to all followers.
 155      * @return a new CaptureSnapshot instance.
 156      */
 157     public CaptureSnapshot newCaptureSnapshot(final ReplicatedLogEntry lastLogEntry, final long replicatedToAllIndex) {
 158         TermInformationReader lastAppliedTermInfoReader =
 159                 lastAppliedTermInformationReader.init(context.getReplicatedLog(), context.getLastApplied(),
 160                         lastLogEntry, hasFollowers());
 161
 162         long lastAppliedIndex = lastAppliedTermInfoReader.getIndex();
 163         long lastAppliedTerm = lastAppliedTermInfoReader.getTerm();
 164
 165         TermInformationReader replicatedToAllTermInfoReader =
 166                 replicatedToAllTermInformationReader.init(context.getReplicatedLog(), replicatedToAllIndex);
 167
 168         long newReplicatedToAllIndex = replicatedToAllTermInfoReader.getIndex();
 169         long newReplicatedToAllTerm = replicatedToAllTermInfoReader.getTerm();
 170
 171         List<ReplicatedLogEntry> unAppliedEntries = context.getReplicatedLog().getFrom(lastAppliedIndex + 1);
 172
 173         long lastLogEntryIndex = lastAppliedIndex;
 174         long lastLogEntryTerm = lastAppliedTerm;
 175         if (lastLogEntry != null) {
 176             lastLogEntryIndex = lastLogEntry.getIndex();
 177             lastLogEntryTerm = lastLogEntry.getTerm();
 178         } else {
 179             log.debug("{}: Capturing Snapshot : lastLogEntry is null. Using lastAppliedIndex {} and "
 180                     + "lastAppliedTerm {} instead.", persistenceId(), lastAppliedIndex, lastAppliedTerm);
 181         }
 182
 183         return new CaptureSnapshot(lastLogEntryIndex, lastLogEntryTerm, lastAppliedIndex, lastAppliedTerm,
 184                 newReplicatedToAllIndex, newReplicatedToAllTerm, unAppliedEntries);
 185     }
 186
 187     private class AbstractSnapshotState implements SnapshotState {
 188
 189         @Override
 190         public boolean isCapturing() {
 191             return true;
 192         }
 193
 194         @Override
 195         public boolean capture(final ReplicatedLogEntry lastLogEntry, final long replicatedToAllIndex) {
 196             log.debug("capture should not be called in state {}", this);
 197             return false;
 198         }
 199
 200         @Override
 201         public boolean captureToInstall(final ReplicatedLogEntry lastLogEntry, final long replicatedToAllIndex,
 202                 final String targetFollower) {
 203             log.debug("captureToInstall should not be called in state {}", this);
 204             return false;
 205         }
 206
 207         @Override
 208         public void apply(final ApplySnapshot snapshot) {
 209             log.debug("apply should not be called in state {}", this);
 210         }
 211
 212         @Override
 213         public void persist(final Snapshot.State state, final Optional<OutputStream> installSnapshotStream,
 214                 final long totalMemory) {
 215             log.debug("persist should not be called in state {}", this);
 216         }
 217
 218         @Override
 219         public void commit(final long sequenceNumber, final long timeStamp) {
 220             log.debug("commit should not be called in state {}", this);
 221         }
 222
 223         @Override
 224         public void rollback() {
 225             log.debug("rollback should not be called in state {}", this);
 226         }
 227
 228         @Override
 229         public long trimLog(final long desiredTrimIndex) {
 230             log.debug("trimLog should not be called in state {}", this);
 231             return -1;
 232         }
 233
 234         protected long doTrimLog(final long desiredTrimIndex) {
 235             //  we would want to keep the lastApplied as its used while capturing snapshots
 236             long lastApplied = context.getLastApplied();
 237             long tempMin = Math.min(desiredTrimIndex, lastApplied > -1 ? lastApplied - 1 : -1);
 238
 239             if (log.isTraceEnabled()) {
 240                 log.trace("{}: performSnapshotWithoutCapture: desiredTrimIndex: {}, lastApplied: {}, tempMin: {}",
 241                         persistenceId(), desiredTrimIndex, lastApplied, tempMin);
 242             }
 243
 244             if (tempMin > -1 && context.getReplicatedLog().isPresent(tempMin)) {
 245                 log.debug("{}: fakeSnapshot purging log to {} for term {}", persistenceId(), tempMin,
 246                         context.getTermInformation().getCurrentTerm());
 247
 248                 //use the term of the temp-min, since we check for isPresent, entry will not be null
 249                 ReplicatedLogEntry entry = context.getReplicatedLog().get(tempMin);
 250                 context.getReplicatedLog().snapshotPreCommit(tempMin, entry.getTerm());
 251                 context.getReplicatedLog().snapshotCommit();
 252                 return tempMin;
 253             }
 254
 255             final RaftActorBehavior currentBehavior = context.getCurrentBehavior();
 256             if (tempMin > currentBehavior.getReplicatedToAllIndex()) {
 257                 // It's possible a follower was lagging and an install snapshot advanced its match index past
 258                 // the current replicatedToAllIndex. Since the follower is now caught up we should advance the
 259                 // replicatedToAllIndex (to tempMin). The fact that tempMin wasn't found in the log is likely
 260                 // due to a previous snapshot triggered by the memory threshold exceeded, in that case we
 261                 // trim the log to the last applied index even if previous entries weren't replicated to all followers.
 262                 currentBehavior.setReplicatedToAllIndex(tempMin);
 263             }
 264             return -1;
 265         }
 266     }
 267
 268     private class Idle extends AbstractSnapshotState {
 269
 270         @Override
 271         public boolean isCapturing() {
 272             return false;
 273         }
 274
 275         @SuppressWarnings("checkstyle:IllegalCatch")
 276         private boolean capture(final ReplicatedLogEntry lastLogEntry, final long replicatedToAllIndex,
 277                 final String targetFollower) {
 278             captureSnapshot = newCaptureSnapshot(lastLogEntry, replicatedToAllIndex);
 279
 280             OutputStream installSnapshotStream = null;
 281             if (targetFollower != null) {
 282                 installSnapshotStream = context.getFileBackedOutputStreamFactory().newInstance();
 283                 log.info("{}: Initiating snapshot capture {} to install on {}",
 284                         persistenceId(), captureSnapshot, targetFollower);
 285             } else {
 286                 log.info("{}: Initiating snapshot capture {}", persistenceId(), captureSnapshot);
 287             }
 288
 289             lastSequenceNumber = context.getPersistenceProvider().getLastSequenceNumber();
 290
 291             log.debug("{}: lastSequenceNumber prior to capture: {}", persistenceId(), lastSequenceNumber);
 292
 293             SnapshotManager.this.currentState = CREATING;
 294
 295             try {
 296                 createSnapshotProcedure.accept(Optional.ofNullable(installSnapshotStream));
 297             } catch (Exception e) {
 298                 SnapshotManager.this.currentState = IDLE;
 299                 log.error("Error creating snapshot", e);
 300                 return false;
 301             }
 302
 303             return true;
 304         }
 305
 306         @Override
 307         public boolean capture(final ReplicatedLogEntry lastLogEntry, final long replicatedToAllIndex) {
 308             return capture(lastLogEntry, replicatedToAllIndex, null);
 309         }
 310
 311         @Override
 312         public boolean captureToInstall(final ReplicatedLogEntry lastLogEntry, final long replicatedToAllIndex,
 313                 final String targetFollower) {
 314             return capture(lastLogEntry, replicatedToAllIndex, targetFollower);
 315         }
 316
 317         @Override
 318         public void apply(final ApplySnapshot toApply) {
 319             SnapshotManager.this.applySnapshot = toApply;
 320
 321             lastSequenceNumber = context.getPersistenceProvider().getLastSequenceNumber();
 322
 323             log.debug("lastSequenceNumber prior to persisting applied snapshot: {}", lastSequenceNumber);
 324
 325             context.getPersistenceProvider().saveSnapshot(toApply.getSnapshot());
 326
 327             SnapshotManager.this.currentState = PERSISTING;
 328         }
 329
 330         @Override
 331         public String toString() {
 332             return "Idle";
 333         }
 334
 335         @Override
 336         public long trimLog(final long desiredTrimIndex) {
 337             return doTrimLog(desiredTrimIndex);
 338         }
 339     }
 340
 341     private class Creating extends AbstractSnapshotState {
 342
 343         @Override
 344         public void persist(final Snapshot.State snapshotState, final Optional<OutputStream> installSnapshotStream,
 345                 final long totalMemory) {
 346             // create a snapshot object from the state provided and save it
 347             // when snapshot is saved async, SaveSnapshotSuccess is raised.
 348
 349             Snapshot snapshot = Snapshot.create(snapshotState,
 350                     captureSnapshot.getUnAppliedEntries(),
 351                     captureSnapshot.getLastIndex(), captureSnapshot.getLastTerm(),
 352                     captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm(),
 353                     context.getTermInformation().getCurrentTerm(),
 354                     context.getTermInformation().getVotedFor(), context.getPeerServerInfo(true));
 355
 356             context.getPersistenceProvider().saveSnapshot(snapshot);
 357
 358             log.info("{}: Persisting of snapshot done: {}", persistenceId(), snapshot);
 359
 360             long dataThreshold = totalMemory * context.getConfigParams().getSnapshotDataThresholdPercentage() / 100;
 361             boolean dataSizeThresholdExceeded = context.getReplicatedLog().dataSize() > dataThreshold;
 362
 363             boolean logSizeExceededSnapshotBatchCount =
 364                     context.getReplicatedLog().size() >= context.getConfigParams().getSnapshotBatchCount();
 365
 366             final RaftActorBehavior currentBehavior = context.getCurrentBehavior();
 367             if (dataSizeThresholdExceeded || logSizeExceededSnapshotBatchCount) {
 368                 if (log.isDebugEnabled()) {
 369                     if (dataSizeThresholdExceeded) {
 370                         log.debug("{}: log data size {} exceeds the memory threshold {} - doing snapshotPreCommit "
 371                                 + "with index {}", context.getId(), context.getReplicatedLog().dataSize(),
 372                                 dataThreshold, captureSnapshot.getLastAppliedIndex());
 373                     } else {
 374                         log.debug("{}: log size {} exceeds the snapshot batch count {} - doing snapshotPreCommit with "
 375                                 + "index {}", context.getId(), context.getReplicatedLog().size(),
 376                                 context.getConfigParams().getSnapshotBatchCount(),
 377                                 captureSnapshot.getLastAppliedIndex());
 378                     }
 379                 }
 380
 381                 // We either exceeded the memory threshold or the log size exceeded the snapshot batch
 382                 // count so, to keep the log memory footprint in check, clear the log based on lastApplied.
 383                 // This could/should only happen if one of the followers is down as normally we keep
 384                 // removing from the log as entries are replicated to all.
 385                 context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getLastAppliedIndex(),
 386                         captureSnapshot.getLastAppliedTerm());
 387
 388                 // Don't reset replicatedToAllIndex to -1 as this may prevent us from trimming the log after an
 389                 // install snapshot to a follower.
 390                 if (captureSnapshot.getReplicatedToAllIndex() >= 0) {
 391                     currentBehavior.setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
 392                 }
 393
 394             } else if (captureSnapshot.getReplicatedToAllIndex() != -1) {
 395                 // clear the log based on replicatedToAllIndex
 396                 context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getReplicatedToAllIndex(),
 397                         captureSnapshot.getReplicatedToAllTerm());
 398
 399                 currentBehavior.setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
 400             } else {
 401                 // The replicatedToAllIndex was not found in the log
 402                 // This means that replicatedToAllIndex never moved beyond -1 or that it is already in the snapshot.
 403                 // In this scenario we may need to save the snapshot to the akka persistence
 404                 // snapshot for recovery but we do not need to do the replicated log trimming.
 405                 context.getReplicatedLog().snapshotPreCommit(context.getReplicatedLog().getSnapshotIndex(),
 406                         context.getReplicatedLog().getSnapshotTerm());
 407             }
 408
 409             log.info("{}: Removed in-memory snapshotted entries, adjusted snaphsotIndex: {} and term: {}",
 410                     context.getId(), context.getReplicatedLog().getSnapshotIndex(),
 411                     context.getReplicatedLog().getSnapshotTerm());
 412
 413             if (installSnapshotStream.isPresent()) {
 414                 if (context.getId().equals(currentBehavior.getLeaderId())) {
 415                     try {
 416                         ByteSource snapshotBytes = ((FileBackedOutputStream)installSnapshotStream.get()).asByteSource();
 417                         currentBehavior.handleMessage(context.getActor(),
 418                                 new SendInstallSnapshot(snapshot, snapshotBytes));
 419                     } catch (IOException e) {
 420                         log.error("{}: Snapshot install failed due to an unrecoverable streaming error",
 421                                 context.getId(), e);
 422                     }
 423                 } else {
 424                     ((FileBackedOutputStream)installSnapshotStream.get()).cleanup();
 425                 }
 426             }
 427
 428             captureSnapshot = null;
 429             SnapshotManager.this.currentState = PERSISTING;
 430         }
 431
 432         @Override
 433         public String toString() {
 434             return "Creating";
 435         }
 436
 437     }
 438
 439     private class Persisting extends AbstractSnapshotState {
 440
 441         @Override
 442         @SuppressWarnings("checkstyle:IllegalCatch")
 443         public void commit(final long sequenceNumber, final long timeStamp) {
 444             log.debug("{}: Snapshot success -  sequence number: {}", persistenceId(), sequenceNumber);
 445
 446             if (applySnapshot != null) {
 447                 try {
 448                     Snapshot snapshot = applySnapshot.getSnapshot();
 449
 450                     //clears the followers log, sets the snapshot index to ensure adjusted-index works
 451                     context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context));
 452                     context.setLastApplied(snapshot.getLastAppliedIndex());
 453                     context.setCommitIndex(snapshot.getLastAppliedIndex());
 454                     context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor());
 455
 456                     if (snapshot.getServerConfiguration() != null) {
 457                         context.updatePeerIds(snapshot.getServerConfiguration());
 458                     }
 459
 460                     if (!(snapshot.getState() instanceof EmptyState)) {
 461                         snapshotCohort.applySnapshot(snapshot.getState());
 462                     }
 463
 464                     applySnapshot.getCallback().onSuccess();
 465                 } catch (Exception e) {
 466                     log.error("{}: Error applying snapshot", context.getId(), e);
 467                 }
 468             } else {
 469                 context.getReplicatedLog().snapshotCommit();
 470             }
 471
 472             context.getPersistenceProvider().deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(),
 473                     timeStamp - 1, 0L, 0L));
 474
 475             context.getPersistenceProvider().deleteMessages(lastSequenceNumber);
 476
 477             snapshotComplete();
 478         }
 479
 480         @Override
 481         public void rollback() {
 482             // Nothing to rollback if we're applying a snapshot from the leader.
 483             if (applySnapshot == null) {
 484                 context.getReplicatedLog().snapshotRollback();
 485
 486                 log.info("{}: Replicated Log rolled back. Snapshot will be attempted in the next cycle."
 487                         + "snapshotIndex:{}, snapshotTerm:{}, log-size:{}", persistenceId(),
 488                         context.getReplicatedLog().getSnapshotIndex(),
 489                         context.getReplicatedLog().getSnapshotTerm(),
 490                         context.getReplicatedLog().size());
 491             } else {
 492                 applySnapshot.getCallback().onFailure();
 493             }
 494
 495             snapshotComplete();
 496         }
 497
 498         private void snapshotComplete() {
 499             lastSequenceNumber = -1;
 500             applySnapshot = null;
 501             SnapshotManager.this.currentState = IDLE;
 502
 503             context.getActor().tell(SnapshotComplete.INSTANCE, context.getActor());
 504         }
 505
 506         @Override
 507         public String toString() {
 508             return "Persisting";
 509         }
 510
 511     }
 512
 513     private interface TermInformationReader {
 514         long getIndex();
 515
 516         long getTerm();
 517     }
 518
 519     static class LastAppliedTermInformationReader implements TermInformationReader {
 520         private long index;
 521         private long term;
 522
 523         LastAppliedTermInformationReader init(final ReplicatedLog log, final long originalIndex,
 524                 final ReplicatedLogEntry lastLogEntry, final boolean hasFollowers) {
 525             ReplicatedLogEntry entry = log.get(originalIndex);
 526             this.index = -1L;
 527             this.term = -1L;
 528             if (!hasFollowers) {
 529                 if (lastLogEntry != null) {
 530                     // since we have persisted the last-log-entry to persistent journal before the capture,
 531                     // we would want to snapshot from this entry.
 532                     index = lastLogEntry.getIndex();
 533                     term = lastLogEntry.getTerm();
 534                 }
 535             } else if (entry != null) {
 536                 index = entry.getIndex();
 537                 term = entry.getTerm();
 538             } else if (log.getSnapshotIndex() > -1) {
 539                 index = log.getSnapshotIndex();
 540                 term = log.getSnapshotTerm();
 541             }
 542             return this;
 543         }
 544
 545         @Override
 546         public long getIndex() {
 547             return this.index;
 548         }
 549
 550         @Override
 551         public long getTerm() {
 552             return this.term;
 553         }
 554     }
 555
 556     private static class ReplicatedToAllTermInformationReader implements TermInformationReader {
 557         private long index;
 558         private long term;
 559
 560         ReplicatedToAllTermInformationReader init(final ReplicatedLog log, final long originalIndex) {
 561             ReplicatedLogEntry entry = log.get(originalIndex);
 562             this.index = -1L;
 563             this.term = -1L;
 564
 565             if (entry != null) {
 566                 index = entry.getIndex();
 567                 term = entry.getTerm();
 568             }
 569
 570             return this;
 571         }
 572
 573         @Override
 574         public long getIndex() {
 575             return this.index;
 576         }
 577
 578         @Override
 579         public long getTerm() {
 580             return this.term;
 581         }
 582     }
 583 }