opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/ShardRecoveryCoordinator.java

   1 /*
   2  * Copyright (c) 2014 Brocade Communications Systems, Inc. and others.  All rights reserved.
   3  *
   4  * This program and the accompanying materials are made available under the
   5  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
   6  * and is available at http://www.eclipse.org/legal/epl-v10.html
   7  */
   8 package org.opendaylight.controller.cluster.datastore;
   9
  10 import com.google.common.base.Preconditions;
  11 import com.google.common.base.Throwables;
  12 import java.io.File;
  13 import java.io.IOException;
  14 import java.util.Map.Entry;
  15 import java.util.Optional;
  16 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
  17 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshot;
  18 import org.opendaylight.controller.cluster.datastore.persisted.DataTreeCandidateSupplier;
  19 import org.opendaylight.controller.cluster.datastore.utils.DataTreeModificationOutput;
  20 import org.opendaylight.controller.cluster.datastore.utils.NormalizedNodeXMLOutput;
  21 import org.opendaylight.controller.cluster.datastore.utils.PruningDataTreeModification;
  22 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
  23 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
  24 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
  25 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
  26 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
  27 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidates;
  28 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
  29 import org.slf4j.Logger;
  30
  31 /**
  32  * Coordinates persistence recovery of journal log entries and snapshots for a shard. Each snapshot
  33  * and journal log entry batch are de-serialized and applied to their own write transaction
  34  * instance in parallel on a thread pool for faster recovery time. However the transactions are
  35  * committed to the data store in the order the corresponding snapshot or log batch are received
  36  * to preserve data store integrity.
  37  *
  38  * @author Thomas Pantelis
  39  */
  40 class ShardRecoveryCoordinator implements RaftActorRecoveryCohort {
  41     private final ShardDataTree store;
  42     private final String shardName;
  43     private final Logger log;
  44     private PruningDataTreeModification transaction;
  45     private int size;
  46     private final byte[] restoreFromSnapshot;
  47
  48     ShardRecoveryCoordinator(ShardDataTree store,  byte[] restoreFromSnapshot, String shardName, Logger log) {
  49         this.store = Preconditions.checkNotNull(store);
  50         this.restoreFromSnapshot = restoreFromSnapshot;
  51         this.shardName = Preconditions.checkNotNull(shardName);
  52         this.log = Preconditions.checkNotNull(log);
  53     }
  54
  55     @Override
  56     public void startLogRecoveryBatch(int maxBatchSize) {
  57         log.debug("{}: starting log recovery batch with max size {}", shardName, maxBatchSize);
  58         transaction = new PruningDataTreeModification(store.newModification(), store.getDataTree(),
  59             store.getSchemaContext());
  60         size = 0;
  61     }
  62
  63     @Override
  64     public void appendRecoveredLogEntry(Payload payload) {
  65         Preconditions.checkState(transaction != null, "call startLogRecovery before calling appendRecoveredLogEntry");
  66
  67         try {
  68             if (payload instanceof DataTreeCandidateSupplier) {
  69                 final Entry<Optional<TransactionIdentifier>, DataTreeCandidate> e =
  70                         ((DataTreeCandidateSupplier)payload).getCandidate();
  71
  72                 DataTreeCandidates.applyToModification(transaction, e.getValue());
  73                 size++;
  74
  75                 if (e.getKey().isPresent()) {
  76                     // FIXME: BUG-5280: propagate transaction state
  77                 }
  78             } else {
  79                 log.error("{}: Unknown payload {} received during recovery", shardName, payload);
  80             }
  81         } catch (IOException e) {
  82             log.error("{}: Error extracting payload", shardName, e);
  83         }
  84     }
  85
  86     private void commitTransaction(PruningDataTreeModification tx) throws DataValidationFailedException {
  87         store.commit(tx.getResultingModification());
  88     }
  89
  90     /**
  91      * Applies the current batched log entries to the data store.
  92      */
  93     @Override
  94     public void applyCurrentLogRecoveryBatch() {
  95         Preconditions.checkState(transaction != null, "call startLogRecovery before calling applyCurrentLogRecoveryBatch");
  96
  97         log.debug("{}: Applying current log recovery batch with size {}", shardName, size);
  98         try {
  99             commitTransaction(transaction);
 100         } catch (Exception e) {
 101             File file = new File(System.getProperty("karaf.data", "."),
 102                     "failed-recovery-batch-" + shardName + ".out");
 103             DataTreeModificationOutput.toFile(file, transaction.getResultingModification());
 104             throw new RuntimeException(String.format(
 105                     "%s: Failed to apply recovery batch. Modification data was written to file %s",
 106                     shardName, file), e);
 107         }
 108         transaction = null;
 109     }
 110
 111     /**
 112      * Applies a recovered snapshot to the data store.
 113      *
 114      * @param snapshotBytes the serialized snapshot
 115      */
 116     @Override
 117     public void applyRecoverySnapshot(final byte[] snapshotBytes) {
 118         log.debug("{}: Applying recovered snapshot", shardName);
 119
 120         final ShardDataTreeSnapshot snapshot;
 121         try {
 122             snapshot = ShardDataTreeSnapshot.deserialize(snapshotBytes);
 123         } catch (IOException e) {
 124             log.error("{}: failed to deserialize snapshot", e);
 125             throw Throwables.propagate(e);
 126         }
 127
 128         final PruningDataTreeModification tx = new PruningDataTreeModification(store.newModification(),
 129                 store.getDataTree(), store.getSchemaContext());
 130
 131         final NormalizedNode<?, ?> node = snapshot.getRootNode().orElse(null);
 132         tx.write(YangInstanceIdentifier.EMPTY, node);
 133
 134         try {
 135             commitTransaction(tx);
 136         } catch (Exception e) {
 137             File file = new File(System.getProperty("karaf.data", "."),
 138                     "failed-recovery-snapshot-" + shardName + ".xml");
 139             NormalizedNodeXMLOutput.toFile(file, node);
 140             throw new RuntimeException(String.format(
 141                     "%s: Failed to apply recovery snapshot. Node data was written to file %s",
 142                     shardName, file), e);
 143         }
 144     }
 145
 146     @Override
 147     public byte[] getRestoreFromSnapshot() {
 148         return restoreFromSnapshot;
 149     }
 150 }