+ LOG.debug("{}: transaction {} ready", logContext, id);
+
+ return createReadyCohort(transaction.getIdentifier(), snapshot, participatingShardNames);
+ }
+
+ void purgeTransaction(final TransactionIdentifier id, final Runnable callback) {
+ LOG.debug("{}: purging transaction {}", logContext, id);
+ replicatePayload(id, PurgeTransactionPayload.create(
+ id, shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity()), callback);
+ }
+
+ public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
+ return dataTree.takeSnapshot().readNode(path);
+ }
+
+ DataTreeSnapshot takeSnapshot() {
+ return dataTree.takeSnapshot();
+ }
+
+ @VisibleForTesting
+ public DataTreeModification newModification() {
+ return dataTree.takeSnapshot().newModification();
+ }
+
+ public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
+ Collection<ShardDataTreeCohort> ret = new ArrayList<>(getQueueSize());
+
+ for (CommitEntry entry: pendingFinishCommits) {
+ ret.add(entry.cohort);
+ }
+
+ for (CommitEntry entry: pendingCommits) {
+ ret.add(entry.cohort);
+ }
+
+ for (CommitEntry entry: pendingTransactions) {
+ ret.add(entry.cohort);
+ }
+
+ pendingFinishCommits.clear();
+ pendingCommits.clear();
+ pendingTransactions.clear();
+ tip = dataTree;
+ return ret;
+ }
+
+ /**
+ * Called some time after {@link #processNextPendingTransaction()} decides to stop processing.
+ */
+ void resumeNextPendingTransaction() {
+ LOG.debug("{}: attempting to resume transaction processing", logContext);
+ processNextPending();
+ }
+
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ private void processNextPendingTransaction() {
+ ++currentTransactionBatch;
+ if (currentTransactionBatch > MAX_TRANSACTION_BATCH) {
+ LOG.debug("{}: Already processed {}, scheduling continuation", logContext, currentTransactionBatch);
+ shard.scheduleNextPendingTransaction();
+ return;
+ }
+
+ processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
+ final SimpleShardDataTreeCohort cohort = entry.cohort;
+ final DataTreeModification modification = cohort.getDataTreeModification();
+
+ LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
+ Exception cause;
+ try {
+ tip.validate(modification);
+ LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
+ cohort.successfulCanCommit();
+ entry.lastAccess = readTime();
+ return;
+ } catch (ConflictingModificationAppliedException e) {
+ LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
+ e.getPath());
+ cause = new OptimisticLockFailedException("Optimistic lock failed for path " + e.getPath(), e);
+ } catch (DataValidationFailedException e) {
+ LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
+ e.getPath(), e);
+
+ // For debugging purposes, allow dumping of the modification. Coupled with the above
+ // precondition log, it should allow us to understand what went on.
+ LOG.debug("{}: Store Tx {}: modifications: {}", logContext, cohort.getIdentifier(), modification);
+ LOG.trace("{}: Current tree: {}", logContext, dataTree);
+ cause = new TransactionCommitFailedException("Data did not pass validation for path " + e.getPath(), e);
+ } catch (Exception e) {
+ LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
+ cause = e;
+ }
+
+ // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
+ pendingTransactions.poll().cohort.failedCanCommit(cause);
+ });
+ }
+
+ private void processNextPending() {
+ processNextPendingCommit();
+ processNextPendingTransaction();
+ }
+
+ private void processNextPending(final Queue<CommitEntry> queue, final State allowedState,
+ final Consumer<CommitEntry> processor) {
+ while (!queue.isEmpty()) {
+ final CommitEntry entry = queue.peek();
+ final SimpleShardDataTreeCohort cohort = entry.cohort;
+
+ if (cohort.isFailed()) {
+ LOG.debug("{}: Removing failed transaction {}", logContext, cohort.getIdentifier());
+ queue.remove();
+ continue;
+ }
+
+ if (cohort.getState() == allowedState) {
+ processor.accept(entry);
+ }
+
+ break;
+ }
+
+ maybeRunOperationOnPendingTransactionsComplete();
+ }
+
+ private void processNextPendingCommit() {
+ processNextPending(pendingCommits, State.COMMIT_PENDING,
+ entry -> startCommit(entry.cohort, entry.cohort.getCandidate()));
+ }
+
+ private boolean peekNextPendingCommit() {
+ final CommitEntry first = pendingCommits.peek();
+ return first != null && first.cohort.getState() == State.COMMIT_PENDING;
+ }
+
+ void startCanCommit(final SimpleShardDataTreeCohort cohort) {
+ final CommitEntry head = pendingTransactions.peek();
+ if (head == null) {
+ LOG.warn("{}: No transactions enqueued while attempting to start canCommit on {}", logContext, cohort);
+ return;
+ }
+ if (!cohort.equals(head.cohort)) {
+ // The tx isn't at the head of the queue so we can't start canCommit at this point. Here we check if this
+ // tx should be moved ahead of other tx's in the READY state in the pendingTransactions queue. If this tx
+ // has other participating shards, it could deadlock with other tx's accessing the same shards
+ // depending on the order the tx's are readied on each shard
+ // (see https://jira.opendaylight.org/browse/CONTROLLER-1836). Therefore, if the preceding participating
+ // shard names for a preceding pending tx, call it A, in the queue matches that of this tx, then this tx
+ // is allowed to be moved ahead of tx A in the queue so it is processed first to avoid potential deadlock
+ // if tx A is behind this tx in the pendingTransactions queue for a preceding shard. In other words, since
+ // canCommmit for this tx was requested before tx A, honor that request. If this tx is moved to the head of
+ // the queue as a result, then proceed with canCommit.
+
+ Collection<String> precedingShardNames = extractPrecedingShardNames(cohort.getParticipatingShardNames());
+ if (precedingShardNames.isEmpty()) {
+ LOG.debug("{}: Tx {} is scheduled for canCommit step", logContext, cohort.getIdentifier());
+ return;
+ }
+
+ LOG.debug("{}: Evaluating tx {} for canCommit - preceding participating shard names {}",
+ logContext, cohort.getIdentifier(), precedingShardNames);
+ final Iterator<CommitEntry> iter = pendingTransactions.iterator();
+ int index = -1;
+ int moveToIndex = -1;
+ while (iter.hasNext()) {
+ final CommitEntry entry = iter.next();
+ ++index;
+
+ if (cohort.equals(entry.cohort)) {
+ if (moveToIndex < 0) {
+ LOG.debug("{}: Not moving tx {} - cannot proceed with canCommit",
+ logContext, cohort.getIdentifier());
+ return;
+ }
+
+ LOG.debug("{}: Moving {} to index {} in the pendingTransactions queue",
+ logContext, cohort.getIdentifier(), moveToIndex);
+ iter.remove();
+ insertEntry(pendingTransactions, entry, moveToIndex);
+
+ if (!cohort.equals(pendingTransactions.peek().cohort)) {
+ LOG.debug("{}: Tx {} is not at the head of the queue - cannot proceed with canCommit",
+ logContext, cohort.getIdentifier());
+ return;
+ }
+
+ LOG.debug("{}: Tx {} is now at the head of the queue - proceeding with canCommit",
+ logContext, cohort.getIdentifier());
+ break;
+ }
+
+ if (entry.cohort.getState() != State.READY) {
+ LOG.debug("{}: Skipping pending transaction {} in state {}",
+ logContext, entry.cohort.getIdentifier(), entry.cohort.getState());
+ continue;
+ }
+
+ final Collection<String> pendingPrecedingShardNames = extractPrecedingShardNames(
+ entry.cohort.getParticipatingShardNames());
+
+ if (precedingShardNames.equals(pendingPrecedingShardNames)) {
+ if (moveToIndex < 0) {
+ LOG.debug("{}: Preceding shard names {} for pending tx {} match - saving moveToIndex {}",
+ logContext, pendingPrecedingShardNames, entry.cohort.getIdentifier(), index);
+ moveToIndex = index;
+ } else {
+ LOG.debug(
+ "{}: Preceding shard names {} for pending tx {} match but moveToIndex already set to {}",
+ logContext, pendingPrecedingShardNames, entry.cohort.getIdentifier(), moveToIndex);
+ }
+ } else {
+ LOG.debug("{}: Preceding shard names {} for pending tx {} differ - skipping",
+ logContext, pendingPrecedingShardNames, entry.cohort.getIdentifier());
+ }
+ }
+ }
+
+ processNextPendingTransaction();
+ }
+
+ private static void insertEntry(final Deque<CommitEntry> queue, final CommitEntry entry, final int atIndex) {
+ if (atIndex == 0) {
+ queue.addFirst(entry);
+ return;
+ }
+
+ LOG.trace("Inserting into Deque at index {}", atIndex);
+
+ Deque<CommitEntry> tempStack = new ArrayDeque<>(atIndex);
+ for (int i = 0; i < atIndex; i++) {
+ tempStack.push(queue.poll());
+ }
+
+ queue.addFirst(entry);
+
+ tempStack.forEach(queue::addFirst);
+ }
+
+ private Collection<String> extractPrecedingShardNames(final Optional<SortedSet<String>> participatingShardNames) {
+ return participatingShardNames.map((Function<SortedSet<String>, Collection<String>>)
+ set -> set.headSet(shard.getShardName())).orElse(Collections.<String>emptyList());
+ }
+
+ private void failPreCommit(final Throwable cause) {
+ shard.getShardMBean().incrementFailedTransactionsCount();
+ pendingTransactions.poll().cohort.failedPreCommit(cause);
+ processNextPendingTransaction();
+ }
+
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ void startPreCommit(final SimpleShardDataTreeCohort cohort) {
+ final CommitEntry entry = pendingTransactions.peek();
+ Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
+
+ final SimpleShardDataTreeCohort current = entry.cohort;
+ Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
+
+ final TransactionIdentifier currentId = current.getIdentifier();
+ LOG.debug("{}: Preparing transaction {}", logContext, currentId);
+
+ final DataTreeCandidateTip candidate;
+ try {
+ candidate = tip.prepare(cohort.getDataTreeModification());
+ LOG.debug("{}: Transaction {} candidate ready", logContext, currentId);
+ } catch (RuntimeException e) {
+ failPreCommit(e);
+ return;
+ }
+
+ cohort.userPreCommit(candidate, new FutureCallback<Void>() {
+ @Override
+ public void onSuccess(final Void noop) {
+ // Set the tip of the data tree.
+ tip = Verify.verifyNotNull(candidate);
+
+ entry.lastAccess = readTime();
+
+ pendingTransactions.remove();
+ pendingCommits.add(entry);
+
+ LOG.debug("{}: Transaction {} prepared", logContext, currentId);
+
+ cohort.successfulPreCommit(candidate);
+
+ processNextPendingTransaction();
+ }
+
+ @Override
+ public void onFailure(final Throwable failure) {
+ failPreCommit(failure);
+ }
+ });
+ }
+
+ private void failCommit(final Exception cause) {
+ shard.getShardMBean().incrementFailedTransactionsCount();
+ pendingFinishCommits.poll().cohort.failedCommit(cause);
+ processNextPending();
+ }
+
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ private void finishCommit(final SimpleShardDataTreeCohort cohort) {
+ final TransactionIdentifier txId = cohort.getIdentifier();
+ final DataTreeCandidate candidate = cohort.getCandidate();
+
+ LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
+
+ if (tip == candidate) {
+ // All pending candidates have been committed, reset the tip to the data tree.
+ tip = dataTree;
+ }
+
+ try {
+ dataTree.commit(candidate);
+ } catch (Exception e) {
+ LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
+ failCommit(e);
+ return;
+ }
+
+ allMetadataCommittedTransaction(txId);
+ shard.getShardMBean().incrementCommittedTransactionCount();
+ shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
+
+ // FIXME: propagate journal index
+ pendingFinishCommits.poll().cohort.successfulCommit(UnsignedLong.ZERO, () -> {
+ LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
+ notifyListeners(candidate);
+
+ processNextPending();
+ });
+ }
+
+ void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
+ final CommitEntry entry = pendingCommits.peek();
+ Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
+
+ final SimpleShardDataTreeCohort current = entry.cohort;
+ if (!cohort.equals(current)) {
+ LOG.debug("{}: Transaction {} scheduled for commit step", logContext, cohort.getIdentifier());
+ return;
+ }
+
+ LOG.debug("{}: Starting commit for transaction {}", logContext, current.getIdentifier());
+
+ final TransactionIdentifier txId = cohort.getIdentifier();
+ final Payload payload;
+ try {
+ payload = CommitTransactionPayload.create(txId, candidate,
+ shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity());
+ } catch (IOException e) {
+ LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
+ pendingCommits.poll().cohort.failedCommit(e);
+ processNextPending();
+ return;
+ }
+
+ // We process next transactions pending canCommit before we call persistPayload to possibly progress subsequent
+ // transactions to the COMMIT_PENDING state so the payloads can be batched for replication. This is done for
+ // single-shard transactions that immediately transition from canCommit to preCommit to commit. Note that
+ // if the next pending transaction is progressed to COMMIT_PENDING and this method (startCommit) is called,
+ // the next transaction will not attempt to replicate b/c the current transaction is still at the head of the
+ // pendingCommits queue.
+ processNextPendingTransaction();
+
+ // After processing next pending transactions, we can now remove the current transaction from pendingCommits.
+ // Note this must be done before the call to peekNextPendingCommit below so we check the next transaction
+ // in order to properly determine the batchHint flag for the call to persistPayload.
+ pendingCommits.remove();
+ pendingFinishCommits.add(entry);
+
+ // See if the next transaction is pending commit (ie in the COMMIT_PENDING state) so it can be batched with
+ // this transaction for replication.
+ boolean replicationBatchHint = peekNextPendingCommit();
+
+ // Once completed, we will continue via payloadReplicationComplete
+ shard.persistPayload(txId, payload, replicationBatchHint);
+
+ entry.lastAccess = shard.ticker().read();
+
+ LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
+
+ // Process the next transaction pending commit, if any. If there is one it will be batched with this
+ // transaction for replication.
+ processNextPendingCommit();