a990b50a3222e5dbbc21ad3eaa8ee99b5279ddca
[controller.git] / opendaylight / md-sal / sal-distributed-datastore / src / main / java / org / opendaylight / controller / cluster / datastore / ShardDataTree.java
1 /*
2  * Copyright (c) 2015 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.datastore;
9
10 import akka.actor.ActorRef;
11 import akka.util.Timeout;
12 import com.google.common.annotations.VisibleForTesting;
13 import com.google.common.base.Optional;
14 import com.google.common.base.Preconditions;
15 import com.google.common.base.Stopwatch;
16 import com.google.common.base.Verify;
17 import com.google.common.collect.ImmutableList;
18 import com.google.common.collect.ImmutableMap;
19 import com.google.common.collect.ImmutableMap.Builder;
20 import com.google.common.primitives.UnsignedLong;
21 import java.io.File;
22 import java.io.IOException;
23 import java.util.AbstractMap.SimpleEntry;
24 import java.util.ArrayDeque;
25 import java.util.ArrayList;
26 import java.util.Collection;
27 import java.util.HashMap;
28 import java.util.Iterator;
29 import java.util.Map;
30 import java.util.Map.Entry;
31 import java.util.Queue;
32 import java.util.concurrent.ExecutionException;
33 import java.util.concurrent.TimeUnit;
34 import java.util.concurrent.TimeoutException;
35 import java.util.function.UnaryOperator;
36 import javax.annotation.Nonnull;
37 import javax.annotation.concurrent.NotThreadSafe;
38 import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
39 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
40 import org.opendaylight.controller.cluster.datastore.DataTreeCohortActorRegistry.CohortRegistryCommand;
41 import org.opendaylight.controller.cluster.datastore.ShardDataTreeCohort.State;
42 import org.opendaylight.controller.cluster.datastore.persisted.CommitTransactionPayload;
43 import org.opendaylight.controller.cluster.datastore.persisted.MetadataShardDataTreeSnapshot;
44 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshot;
45 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshotMetadata;
46 import org.opendaylight.controller.cluster.datastore.utils.DataTreeModificationOutput;
47 import org.opendaylight.controller.cluster.datastore.utils.PruningDataTreeModification;
48 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
49 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataBroker.DataChangeScope;
50 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataChangeListener;
51 import org.opendaylight.controller.md.sal.common.api.data.OptimisticLockFailedException;
52 import org.opendaylight.controller.md.sal.common.api.data.TransactionCommitFailedException;
53 import org.opendaylight.controller.md.sal.dom.api.DOMDataTreeChangeListener;
54 import org.opendaylight.controller.md.sal.dom.store.impl.DataChangeListenerRegistration;
55 import org.opendaylight.yangtools.concepts.Identifier;
56 import org.opendaylight.yangtools.concepts.ListenerRegistration;
57 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
58 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
59 import org.opendaylight.yangtools.yang.data.api.schema.tree.ConflictingModificationAppliedException;
60 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
61 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidateTip;
62 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidates;
63 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeModification;
64 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeSnapshot;
65 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
66 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
67 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTree;
68 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
69 import org.opendaylight.yangtools.yang.data.impl.schema.tree.InMemoryDataTreeFactory;
70 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
71 import org.slf4j.Logger;
72 import org.slf4j.LoggerFactory;
73 import scala.concurrent.duration.Duration;
74
75 /**
76  * Internal shard state, similar to a DOMStore, but optimized for use in the actor system,
77  * e.g. it does not expose public interfaces and assumes it is only ever called from a
78  * single thread.
79  *
80  * This class is not part of the API contract and is subject to change at any time.
81  */
82 @NotThreadSafe
83 public class ShardDataTree extends ShardDataTreeTransactionParent {
84     private static final class CommitEntry {
85         final SimpleShardDataTreeCohort cohort;
86         long lastAccess;
87
88         CommitEntry(final SimpleShardDataTreeCohort cohort, final long now) {
89             this.cohort = Preconditions.checkNotNull(cohort);
90             lastAccess = now;
91         }
92     }
93
94     private static final Timeout COMMIT_STEP_TIMEOUT = new Timeout(Duration.create(5, TimeUnit.SECONDS));
95     private static final Logger LOG = LoggerFactory.getLogger(ShardDataTree.class);
96
97     private final Map<LocalHistoryIdentifier, ShardDataTreeTransactionChain> transactionChains = new HashMap<>();
98     private final DataTreeCohortActorRegistry cohortRegistry = new DataTreeCohortActorRegistry();
99     private final Queue<CommitEntry> pendingTransactions = new ArrayDeque<>();
100     private final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher;
101     private final ShardDataChangeListenerPublisher dataChangeListenerPublisher;
102     private final Collection<ShardDataTreeMetadata<?>> metadata;
103     private final TipProducingDataTree dataTree;
104     private final String logContext;
105     private final Shard shard;
106     private Runnable runOnPendingTransactionsComplete;
107
108     private SchemaContext schemaContext;
109
110     public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TipProducingDataTree dataTree,
111             final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
112             final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
113             final ShardDataTreeMetadata<?>... metadata) {
114         this.dataTree = Preconditions.checkNotNull(dataTree);
115         updateSchemaContext(schemaContext);
116
117         this.shard = Preconditions.checkNotNull(shard);
118         this.treeChangeListenerPublisher = Preconditions.checkNotNull(treeChangeListenerPublisher);
119         this.dataChangeListenerPublisher = Preconditions.checkNotNull(dataChangeListenerPublisher);
120         this.logContext = Preconditions.checkNotNull(logContext);
121         this.metadata = ImmutableList.copyOf(metadata);
122     }
123
124     public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType,
125             final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
126             final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext) {
127         this(shard, schemaContext, InMemoryDataTreeFactory.getInstance().create(treeType),
128                 treeChangeListenerPublisher, dataChangeListenerPublisher, logContext);
129     }
130
131     @VisibleForTesting
132     public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType) {
133         this(shard, schemaContext, treeType, new DefaultShardDataTreeChangeListenerPublisher(),
134                 new DefaultShardDataChangeListenerPublisher(), "");
135     }
136
137     String logContext() {
138         return logContext;
139     }
140
141     public TipProducingDataTree getDataTree() {
142         return dataTree;
143     }
144
145     SchemaContext getSchemaContext() {
146         return schemaContext;
147     }
148
149     void updateSchemaContext(final SchemaContext schemaContext) {
150         dataTree.setSchemaContext(schemaContext);
151         this.schemaContext = Preconditions.checkNotNull(schemaContext);
152     }
153
154     /**
155      * Take a snapshot of current state for later recovery.
156      *
157      * @return A state snapshot
158      */
159     @Nonnull ShardDataTreeSnapshot takeStateSnapshot() {
160         final NormalizedNode<?, ?> rootNode = dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY).get();
161         final Builder<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> metaBuilder =
162                 ImmutableMap.builder();
163
164         for (ShardDataTreeMetadata<?> m : metadata) {
165             final ShardDataTreeSnapshotMetadata<?> meta = m.toSnapshot();
166             if (meta != null) {
167                 metaBuilder.put(meta.getType(), meta);
168             }
169         }
170
171         return new MetadataShardDataTreeSnapshot(rootNode, metaBuilder.build());
172     }
173
174     private void applySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot,
175             final UnaryOperator<DataTreeModification> wrapper) throws DataValidationFailedException {
176         final Stopwatch elapsed = Stopwatch.createStarted();
177
178         if (!pendingTransactions.isEmpty()) {
179             LOG.warn("{}: applying state snapshot with pending transactions", logContext);
180         }
181
182         final Map<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> snapshotMeta;
183         if (snapshot instanceof MetadataShardDataTreeSnapshot) {
184             snapshotMeta = ((MetadataShardDataTreeSnapshot) snapshot).getMetadata();
185         } else {
186             snapshotMeta = ImmutableMap.of();
187         }
188
189         for (ShardDataTreeMetadata<?> m : metadata) {
190             final ShardDataTreeSnapshotMetadata<?> s = snapshotMeta.get(m.getSupportedType());
191             if (s != null) {
192                 m.applySnapshot(s);
193             } else {
194                 m.reset();
195             }
196         }
197
198         final DataTreeModification mod = wrapper.apply(dataTree.takeSnapshot().newModification());
199         // delete everything first
200         mod.delete(YangInstanceIdentifier.EMPTY);
201
202         final java.util.Optional<NormalizedNode<?, ?>> maybeNode = snapshot.getRootNode();
203         if (maybeNode.isPresent()) {
204             // Add everything from the remote node back
205             mod.write(YangInstanceIdentifier.EMPTY, maybeNode.get());
206         }
207         mod.ready();
208
209         final DataTreeModification unwrapped = unwrap(mod);
210         dataTree.validate(unwrapped);
211         dataTree.commit(dataTree.prepare(unwrapped));
212         LOG.debug("{}: state snapshot applied in %s", logContext, elapsed);
213     }
214
215     private PruningDataTreeModification wrapWithPruning(final DataTreeModification delegate) {
216         return new PruningDataTreeModification(delegate, dataTree, schemaContext);
217     }
218
219     private static DataTreeModification unwrap(final DataTreeModification modification) {
220         if (modification instanceof PruningDataTreeModification) {
221             return ((PruningDataTreeModification)modification).delegate();
222         }
223         return modification;
224     }
225
226     /**
227      * Apply a snapshot coming from recovery. This method does not assume the SchemaContexts match and performs data
228      * pruning in an attempt to adjust the state to our current SchemaContext.
229      *
230      * @param snapshot Snapshot that needs to be applied
231      * @throws DataValidationFailedException when the snapshot fails to apply
232      */
233     void applyRecoverySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
234         applySnapshot(snapshot, this::wrapWithPruning);
235     }
236
237
238     /**
239      * Apply a snapshot coming from the leader. This method assumes the leader and follower SchemaContexts match and
240      * does not perform any pruning.
241      *
242      * @param snapshot Snapshot that needs to be applied
243      * @throws DataValidationFailedException when the snapshot fails to apply
244      */
245     void applySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
246         applySnapshot(snapshot, UnaryOperator.identity());
247     }
248
249     private void applyRecoveryCandidate(final DataTreeCandidate candidate) throws DataValidationFailedException {
250         final PruningDataTreeModification mod = wrapWithPruning(dataTree.takeSnapshot().newModification());
251         DataTreeCandidates.applyToModification(mod, candidate);
252         mod.ready();
253
254         final DataTreeModification unwrapped = mod.delegate();
255         LOG.trace("{}: Applying recovery modification {}", logContext, unwrapped);
256
257         try {
258             dataTree.validate(unwrapped);
259             dataTree.commit(dataTree.prepare(unwrapped));
260         } catch (Exception e) {
261             File file = new File(System.getProperty("karaf.data", "."),
262                     "failed-recovery-payload-" + logContext + ".out");
263             DataTreeModificationOutput.toFile(file, unwrapped);
264             throw new IllegalStateException(String.format(
265                     "%s: Failed to apply recovery payload. Modification data was written to file %s",
266                     logContext, file), e);
267         }
268     }
269
270     /**
271      * Apply a payload coming from recovery. This method does not assume the SchemaContexts match and performs data
272      * pruning in an attempt to adjust the state to our current SchemaContext.
273      *
274      * @param payload Payload
275      * @throws IOException when the snapshot fails to deserialize
276      * @throws DataValidationFailedException when the snapshot fails to apply
277      */
278     void applyRecoveryPayload(final @Nonnull Payload payload) throws IOException, DataValidationFailedException {
279         if (payload instanceof CommitTransactionPayload) {
280             final Entry<TransactionIdentifier, DataTreeCandidate> e = ((CommitTransactionPayload) payload).getCandidate();
281             applyRecoveryCandidate(e.getValue());
282             allMetadataCommittedTransaction(e.getKey());
283         } else if (payload instanceof DataTreeCandidatePayload) {
284             applyRecoveryCandidate(((DataTreeCandidatePayload) payload).getCandidate());
285         } else {
286             LOG.debug("{}: ignoring unhandled payload {}", logContext, payload);
287         }
288     }
289
290     private void applyReplicatedCandidate(final Identifier identifier, final DataTreeCandidate foreign)
291             throws DataValidationFailedException {
292         LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
293
294         final DataTreeModification mod = dataTree.takeSnapshot().newModification();
295         DataTreeCandidates.applyToModification(mod, foreign);
296         mod.ready();
297
298         LOG.trace("{}: Applying foreign modification {}", logContext, mod);
299         dataTree.validate(mod);
300         final DataTreeCandidate candidate = dataTree.prepare(mod);
301         dataTree.commit(candidate);
302
303         notifyListeners(candidate);
304     }
305
306     /**
307      * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
308      * SchemaContexts match and does not perform any pruning.
309      *
310      * @param identifier Payload identifier as returned from RaftActor
311      * @param payload Payload
312      * @throws IOException when the snapshot fails to deserialize
313      * @throws DataValidationFailedException when the snapshot fails to apply
314      */
315     void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
316             DataValidationFailedException {
317         /*
318          * This is a bit more involved than it needs to be due to to the fact we do not want to be touching the payload
319          * if we are the leader and it has originated with us.
320          *
321          * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
322          * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
323          * acting in that capacity anymore.
324          *
325          * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
326          * pre-Boron state -- which limits the number of options here.
327          */
328         if (payload instanceof CommitTransactionPayload) {
329             if (identifier == null) {
330                 final Entry<TransactionIdentifier, DataTreeCandidate> e = ((CommitTransactionPayload) payload).getCandidate();
331                 applyReplicatedCandidate(e.getKey(), e.getValue());
332                 allMetadataCommittedTransaction(e.getKey());
333             } else {
334                 Verify.verify(identifier instanceof TransactionIdentifier);
335                 payloadReplicationComplete((TransactionIdentifier) identifier);
336             }
337         } else {
338             LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
339         }
340     }
341
342     private void payloadReplicationComplete(final TransactionIdentifier txId) {
343         final CommitEntry current = pendingTransactions.peek();
344         if (current == null) {
345             LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
346             return;
347         }
348
349         if (!current.cohort.getIdentifier().equals(txId)) {
350             LOG.warn("{}: Head of queue is {}, ignoring consensus on transaction {}", logContext,
351                 current.cohort.getIdentifier(), txId);
352             return;
353         }
354
355         finishCommit(current.cohort);
356     }
357
358     private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
359         for (ShardDataTreeMetadata<?> m : metadata) {
360             m.onTransactionCommitted(txId);
361         }
362     }
363
364     private ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier localHistoryIdentifier) {
365         ShardDataTreeTransactionChain chain = transactionChains.get(localHistoryIdentifier);
366         if (chain == null) {
367             chain = new ShardDataTreeTransactionChain(localHistoryIdentifier, this);
368             transactionChains.put(localHistoryIdentifier, chain);
369         }
370
371         return chain;
372     }
373
374     ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
375         if (txId.getHistoryId().getHistoryId() == 0) {
376             return new ReadOnlyShardDataTreeTransaction(txId, dataTree.takeSnapshot());
377         }
378
379         return ensureTransactionChain(txId.getHistoryId()).newReadOnlyTransaction(txId);
380     }
381
382     ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
383         if (txId.getHistoryId().getHistoryId() == 0) {
384             return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
385                     .newModification());
386         }
387
388         return ensureTransactionChain(txId.getHistoryId()).newReadWriteTransaction(txId);
389     }
390
391     @VisibleForTesting
392     public void notifyListeners(final DataTreeCandidate candidate) {
393         treeChangeListenerPublisher.publishChanges(candidate, logContext);
394         dataChangeListenerPublisher.publishChanges(candidate, logContext);
395     }
396
397     void notifyOfInitialData(final DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
398             NormalizedNode<?, ?>>> listenerReg, final Optional<DataTreeCandidate> currentState) {
399         if (currentState.isPresent()) {
400             ShardDataChangeListenerPublisher localPublisher = dataChangeListenerPublisher.newInstance();
401             localPublisher.registerDataChangeListener(listenerReg.getPath(), listenerReg.getInstance(),
402                     listenerReg.getScope());
403             localPublisher.publishChanges(currentState.get(), logContext);
404         }
405     }
406
407     void notifyOfInitialData(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
408             final Optional<DataTreeCandidate> currentState) {
409         if (currentState.isPresent()) {
410             ShardDataTreeChangeListenerPublisher localPublisher = treeChangeListenerPublisher.newInstance();
411             localPublisher.registerTreeChangeListener(path, listener);
412             localPublisher.publishChanges(currentState.get(), logContext);
413         }
414     }
415
416     void closeAllTransactionChains() {
417         for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
418             chain.close();
419         }
420
421         transactionChains.clear();
422     }
423
424     void closeTransactionChain(final LocalHistoryIdentifier transactionChainId) {
425         final ShardDataTreeTransactionChain chain = transactionChains.remove(transactionChainId);
426         if (chain != null) {
427             chain.close();
428         } else {
429             LOG.debug("{}: Closing non-existent transaction chain {}", logContext, transactionChainId);
430         }
431     }
432
433     Entry<DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>,
434             Optional<DataTreeCandidate>> registerChangeListener(final YangInstanceIdentifier path,
435                     final AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener,
436                     final DataChangeScope scope) {
437         final DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>> reg =
438                 dataChangeListenerPublisher.registerDataChangeListener(path, listener, scope);
439
440         return new SimpleEntry<>(reg, readCurrentData());
441     }
442
443     private Optional<DataTreeCandidate> readCurrentData() {
444         final Optional<NormalizedNode<?, ?>> currentState = dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY);
445         return currentState.isPresent() ? Optional.of(DataTreeCandidates.fromNormalizedNode(
446             YangInstanceIdentifier.EMPTY, currentState.get())) : Optional.<DataTreeCandidate>absent();
447     }
448
449     public Entry<ListenerRegistration<DOMDataTreeChangeListener>, Optional<DataTreeCandidate>> registerTreeChangeListener(
450             final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener) {
451         final ListenerRegistration<DOMDataTreeChangeListener> reg = treeChangeListenerPublisher.registerTreeChangeListener(
452                 path, listener);
453
454         return new SimpleEntry<>(reg, readCurrentData());
455     }
456
457     int getQueueSize() {
458         return pendingTransactions.size();
459     }
460
461     @Override
462     void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction) {
463         // Intentional no-op
464     }
465
466     @Override
467     ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction) {
468         final DataTreeModification snapshot = transaction.getSnapshot();
469         snapshot.ready();
470
471         return createReadyCohort(transaction.getId(), snapshot);
472     }
473
474     public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
475         return dataTree.takeSnapshot().readNode(path);
476     }
477
478     DataTreeSnapshot takeSnapshot() {
479         return dataTree.takeSnapshot();
480     }
481
482     @VisibleForTesting
483     public DataTreeModification newModification() {
484         return dataTree.takeSnapshot().newModification();
485     }
486
487     /**
488      * @deprecated This method violates DataTree containment and will be removed.
489      */
490     @VisibleForTesting
491     @Deprecated
492     public DataTreeCandidate commit(final DataTreeModification modification) throws DataValidationFailedException {
493         modification.ready();
494         dataTree.validate(modification);
495         DataTreeCandidate candidate = dataTree.prepare(modification);
496         dataTree.commit(candidate);
497         return candidate;
498     }
499
500     public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
501         Collection<ShardDataTreeCohort> ret = new ArrayList<>(pendingTransactions.size());
502         for(CommitEntry entry: pendingTransactions) {
503             ret.add(entry.cohort);
504         }
505
506         pendingTransactions.clear();
507         return ret;
508     }
509
510     private void processNextTransaction() {
511         while (!pendingTransactions.isEmpty()) {
512             final CommitEntry entry = pendingTransactions.peek();
513             final SimpleShardDataTreeCohort cohort = entry.cohort;
514             final DataTreeModification modification = cohort.getDataTreeModification();
515
516             if(cohort.getState() != State.CAN_COMMIT_PENDING) {
517                 break;
518             }
519
520             LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
521             Exception cause;
522             try {
523                 dataTree.validate(modification);
524                 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
525                 cohort.successfulCanCommit();
526                 entry.lastAccess = shard.ticker().read();
527                 return;
528             } catch (ConflictingModificationAppliedException e) {
529                 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
530                     e.getPath());
531                 cause = new OptimisticLockFailedException("Optimistic lock failed.", e);
532             } catch (DataValidationFailedException e) {
533                 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
534                     e.getPath(), e);
535
536                 // For debugging purposes, allow dumping of the modification. Coupled with the above
537                 // precondition log, it should allow us to understand what went on.
538                 LOG.debug("{}: Store Tx {}: modifications: {} tree: {}", cohort.getIdentifier(), modification, dataTree);
539                 cause = new TransactionCommitFailedException("Data did not pass validation.", e);
540             } catch (Exception e) {
541                 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
542                 cause = e;
543             }
544
545             // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
546             pendingTransactions.poll().cohort.failedCanCommit(cause);
547         }
548
549         maybeRunOperationOnPendingTransactionsComplete();
550     }
551
552     void startCanCommit(final SimpleShardDataTreeCohort cohort) {
553         final SimpleShardDataTreeCohort current = pendingTransactions.peek().cohort;
554         if (!cohort.equals(current)) {
555             LOG.debug("{}: Transaction {} scheduled for canCommit step", logContext, cohort.getIdentifier());
556             return;
557         }
558
559         processNextTransaction();
560     }
561
562     private void failPreCommit(final Exception cause) {
563         shard.getShardMBean().incrementFailedTransactionsCount();
564         pendingTransactions.poll().cohort.failedPreCommit(cause);
565         processNextTransaction();
566     }
567
568     void startPreCommit(final SimpleShardDataTreeCohort cohort) {
569         final CommitEntry entry = pendingTransactions.peek();
570         Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
571
572         final SimpleShardDataTreeCohort current = entry.cohort;
573         Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
574         final DataTreeCandidateTip candidate;
575         try {
576             candidate = dataTree.prepare(cohort.getDataTreeModification());
577         } catch (Exception e) {
578             failPreCommit(e);
579             return;
580         }
581
582         try {
583             cohort.userPreCommit(candidate);
584         } catch (ExecutionException | TimeoutException e) {
585             failPreCommit(e);
586             return;
587         }
588
589         entry.lastAccess = shard.ticker().read();
590         cohort.successfulPreCommit(candidate);
591     }
592
593     private void failCommit(final Exception cause) {
594         shard.getShardMBean().incrementFailedTransactionsCount();
595         pendingTransactions.poll().cohort.failedCommit(cause);
596         processNextTransaction();
597     }
598
599     private void finishCommit(final SimpleShardDataTreeCohort cohort) {
600         final TransactionIdentifier txId = cohort.getIdentifier();
601         final DataTreeCandidate candidate = cohort.getCandidate();
602
603         LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
604
605         try {
606             dataTree.commit(candidate);
607         } catch (Exception e) {
608             LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
609             failCommit(e);
610             return;
611         }
612
613         shard.getShardMBean().incrementCommittedTransactionCount();
614         shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
615
616         // FIXME: propagate journal index
617         pendingTransactions.poll().cohort.successfulCommit(UnsignedLong.ZERO);
618
619         LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
620         notifyListeners(candidate);
621
622         processNextTransaction();
623     }
624
625     void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
626         final CommitEntry entry = pendingTransactions.peek();
627         Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
628
629         final SimpleShardDataTreeCohort current = entry.cohort;
630         Verify.verify(cohort.equals(current), "Attempted to commit %s while %s is pending", cohort, current);
631
632         if (shard.canSkipPayload() || candidate.getRootNode().getModificationType() == ModificationType.UNMODIFIED) {
633             LOG.debug("{}: No replication required, proceeding to finish commit", logContext);
634             finishCommit(cohort);
635             return;
636         }
637
638         final TransactionIdentifier txId = cohort.getIdentifier();
639         final Payload payload;
640         try {
641             payload = CommitTransactionPayload.create(txId, candidate);
642         } catch (IOException e) {
643             LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
644             pendingTransactions.poll().cohort.failedCommit(e);
645             return;
646         }
647
648         // Once completed, we will continue via payloadReplicationComplete
649         entry.lastAccess = shard.ticker().read();
650         shard.persistPayload(txId, payload);
651         LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
652     }
653
654     void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
655         cohortRegistry.process(sender, message);
656     }
657
658     ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId,
659             final DataTreeModification modification) {
660         SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, modification, txId,
661                 cohortRegistry.createCohort(schemaContext, txId, COMMIT_STEP_TIMEOUT));
662         pendingTransactions.add(new CommitEntry(cohort, shard.ticker().read()));
663         return cohort;
664     }
665
666     void checkForExpiredTransactions(final long transactionCommitTimeoutMillis) {
667         final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
668         final long now = shard.ticker().read();
669         final CommitEntry currentTx = pendingTransactions.peek();
670         if (currentTx != null && currentTx.lastAccess + timeout < now) {
671             LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
672                     currentTx.cohort.getIdentifier(), transactionCommitTimeoutMillis, currentTx.cohort.getState());
673             boolean processNext = true;
674             switch (currentTx.cohort.getState()) {
675                 case CAN_COMMIT_PENDING:
676                     pendingTransactions.poll().cohort.failedCanCommit(new TimeoutException());
677                     break;
678                 case CAN_COMMIT_COMPLETE:
679                     pendingTransactions.poll().cohort.reportFailure(new TimeoutException());
680                     break;
681                 case PRE_COMMIT_PENDING:
682                     pendingTransactions.poll().cohort.failedPreCommit(new TimeoutException());
683                     break;
684                 case PRE_COMMIT_COMPLETE:
685                     // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
686                     //        are ready we should commit the transaction, not abort it. Our current software stack does
687                     //        not allow us to do that consistently, because we persist at the time of commit, hence
688                     //        we can end up in a state where we have pre-committed a transaction, then a leader failover
689                     //        occurred ... the new leader does not see the pre-committed transaction and does not have
690                     //        a running timer. To fix this we really need two persistence events.
691                     //
692                     //        The first one, done at pre-commit time will hold the transaction payload. When consensus
693                     //        is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
694                     //        apply the state in this event.
695                     //
696                     //        The second one, done at commit (or abort) time holds only the transaction identifier and
697                     //        signals to followers that the state should (or should not) be applied.
698                     //
699                     //        In order to make the pre-commit timer working across failovers, though, we need
700                     //        a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
701                     //        restart the timer.
702                     pendingTransactions.poll().cohort.reportFailure(new TimeoutException());
703                     break;
704                 case COMMIT_PENDING:
705                     LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
706                         currentTx.cohort.getIdentifier());
707                     currentTx.lastAccess = now;
708                     processNext = false;
709                     return;
710                 case ABORTED:
711                 case COMMITTED:
712                 case FAILED:
713                 case READY:
714                 default:
715                     pendingTransactions.poll();
716             }
717
718             if (processNext) {
719                 processNextTransaction();
720             }
721         }
722     }
723
724     void startAbort(final SimpleShardDataTreeCohort cohort) {
725         final Iterator<CommitEntry> it = pendingTransactions.iterator();
726         if (!it.hasNext()) {
727             LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
728             return;
729         }
730
731         // First entry is special, as it may already be committing
732         final CommitEntry first = it.next();
733         if (cohort.equals(first.cohort)) {
734             if (cohort.getState() != State.COMMIT_PENDING) {
735                 LOG.debug("{}: aborted head of queue {} in state {}", logContext, cohort.getIdentifier(),
736                     cohort.getIdentifier());
737                 pendingTransactions.poll();
738                 processNextTransaction();
739             } else {
740                 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
741             }
742
743             return;
744         }
745
746         while (it.hasNext()) {
747             final CommitEntry e = it.next();
748             if (cohort.equals(e.cohort)) {
749                 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
750                 it.remove();
751                 return;
752             }
753         }
754
755         LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
756     }
757
758     void setRunOnPendingTransactionsComplete(final Runnable operation) {
759         runOnPendingTransactionsComplete = operation;
760         maybeRunOperationOnPendingTransactionsComplete();
761     }
762
763     private void maybeRunOperationOnPendingTransactionsComplete() {
764       if (runOnPendingTransactionsComplete != null && pendingTransactions.isEmpty()) {
765           LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
766                   runOnPendingTransactionsComplete);
767
768           runOnPendingTransactionsComplete.run();
769           runOnPendingTransactionsComplete = null;
770       }
771   }
772 }

©2013 OpenDaylight, A Linux Foundation Collaborative Project. All Rights Reserved.
OpenDaylight is a registered trademark of The OpenDaylight Project, Inc.
Linux Foundation and OpenDaylight are registered trademarks of the Linux Foundation.
Linux is a registered trademark of Linus Torvalds.