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