Examples with BackupManager org.apache.solr.core.backup.BackupManager used on opensource projects

Example 1 with BackupManager

use of org.apache.solr.core.backup.BackupManager in project lucene-solr by apache.

the class BackupCmd method call.

@Override
public void call(ClusterState state, ZkNodeProps message, NamedList results) throws Exception {
    String collectionName = message.getStr(COLLECTION_PROP);
    String backupName = message.getStr(NAME);
    String repo = message.getStr(CoreAdminParams.BACKUP_REPOSITORY);
    Instant startTime = Instant.now();
    CoreContainer cc = ocmh.overseer.getZkController().getCoreContainer();
    BackupRepository repository = cc.newBackupRepository(Optional.ofNullable(repo));
    BackupManager backupMgr = new BackupManager(repository, ocmh.zkStateReader);
    // Backup location
    URI location = repository.createURI(message.getStr(CoreAdminParams.BACKUP_LOCATION));
    URI backupPath = repository.resolve(location, backupName);
    //Validating if the directory already exists.
    if (repository.exists(backupPath)) {
        throw new SolrException(SolrException.ErrorCode.BAD_REQUEST, "The backup directory already exists: " + backupPath);
    }
    // Create a directory to store backup details.
    repository.createDirectory(backupPath);
    String strategy = message.getStr(CollectionAdminParams.INDEX_BACKUP_STRATEGY, CollectionAdminParams.COPY_FILES_STRATEGY);
    switch(strategy) {
        case CollectionAdminParams.COPY_FILES_STRATEGY:
            {
                copyIndexFiles(backupPath, message, results);
                break;
            }
        case CollectionAdminParams.NO_INDEX_BACKUP_STRATEGY:
            {
                break;
            }
    }
    log.info("Starting to backup ZK data for backupName={}", backupName);
    //Download the configs
    String configName = ocmh.zkStateReader.readConfigName(collectionName);
    backupMgr.downloadConfigDir(location, backupName, configName);
    //Save the collection's state. Can be part of the monolithic clusterstate.json or a individual state.json
    //Since we don't want to distinguish we extract the state and back it up as a separate json
    DocCollection collectionState = ocmh.zkStateReader.getClusterState().getCollection(collectionName);
    backupMgr.writeCollectionState(location, backupName, collectionName, collectionState);
    Properties properties = new Properties();
    properties.put(BackupManager.BACKUP_NAME_PROP, backupName);
    properties.put(BackupManager.COLLECTION_NAME_PROP, collectionName);
    properties.put(COLL_CONF, configName);
    properties.put(BackupManager.START_TIME_PROP, startTime.toString());
    properties.put(BackupManager.INDEX_VERSION_PROP, Version.LATEST.toString());
    //TODO: Add MD5 of the configset. If during restore the same name configset exists then we can compare checksums to see if they are the same.
    //if they are not the same then we can throw an error or have an 'overwriteConfig' flag
    //TODO save numDocs for the shardLeader. We can use it to sanity check the restore.
    backupMgr.writeBackupProperties(location, backupName, properties);
    log.info("Completed backing up ZK data for backupName={}", backupName);
}

Example 2 with BackupManager

use of org.apache.solr.core.backup.BackupManager in project lucene-solr by apache.

the class TestHdfsCloudBackupRestore method testConfigBackupOnly.

protected void testConfigBackupOnly(String configName, String collectionName) throws Exception {
    String backupName = "configonlybackup";
    CloudSolrClient solrClient = cluster.getSolrClient();
    CollectionAdminRequest.Backup backup = CollectionAdminRequest.backupCollection(collectionName, backupName).setRepositoryName(getBackupRepoName()).setIndexBackupStrategy(CollectionAdminParams.NO_INDEX_BACKUP_STRATEGY);
    backup.process(solrClient);
    Map<String, String> params = new HashMap<>();
    params.put("location", "/backup");
    params.put("solr.hdfs.home", hdfsUri + "/solr");
    HdfsBackupRepository repo = new HdfsBackupRepository();
    repo.init(new NamedList<>(params));
    BackupManager mgr = new BackupManager(repo, solrClient.getZkStateReader());
    URI baseLoc = repo.createURI("/backup");
    Properties props = mgr.readBackupProperties(baseLoc, backupName);
    assertNotNull(props);
    assertEquals(collectionName, props.getProperty(COLLECTION_NAME_PROP));
    assertEquals(backupName, props.getProperty(BACKUP_NAME_PROP));
    assertEquals(configName, props.getProperty(COLL_CONF));
    DocCollection collectionState = mgr.readCollectionState(baseLoc, backupName, collectionName);
    assertNotNull(collectionState);
    assertEquals(collectionName, collectionState.getName());
    URI configDirLoc = repo.resolve(baseLoc, backupName, ZK_STATE_DIR, CONFIG_STATE_DIR, configName);
    assertTrue(repo.exists(configDirLoc));
    Collection<String> expected = Arrays.asList(BACKUP_PROPS_FILE, ZK_STATE_DIR);
    URI backupLoc = repo.resolve(baseLoc, backupName);
    String[] dirs = repo.listAll(backupLoc);
    for (String d : dirs) {
        assertTrue(expected.contains(d));
    }
}

Example 3 with BackupManager

use of org.apache.solr.core.backup.BackupManager in project lucene-solr by apache.

the class RestoreCmd method call.

@Override
public void call(ClusterState state, ZkNodeProps message, NamedList results) throws Exception {
    // TODO maybe we can inherit createCollection's options/code
    String restoreCollectionName = message.getStr(COLLECTION_PROP);
    // of backup
    String backupName = message.getStr(NAME);
    ShardHandler shardHandler = ocmh.shardHandlerFactory.getShardHandler();
    String asyncId = message.getStr(ASYNC);
    String repo = message.getStr(CoreAdminParams.BACKUP_REPOSITORY);
    Map<String, String> requestMap = new HashMap<>();
    CoreContainer cc = ocmh.overseer.getZkController().getCoreContainer();
    BackupRepository repository = cc.newBackupRepository(Optional.ofNullable(repo));
    URI location = repository.createURI(message.getStr(CoreAdminParams.BACKUP_LOCATION));
    URI backupPath = repository.resolve(location, backupName);
    ZkStateReader zkStateReader = ocmh.zkStateReader;
    BackupManager backupMgr = new BackupManager(repository, zkStateReader);
    Properties properties = backupMgr.readBackupProperties(location, backupName);
    String backupCollection = properties.getProperty(BackupManager.COLLECTION_NAME_PROP);
    DocCollection backupCollectionState = backupMgr.readCollectionState(location, backupName, backupCollection);
    // Get the Solr nodes to restore a collection.
    final List<String> nodeList = OverseerCollectionMessageHandler.getLiveOrLiveAndCreateNodeSetList(zkStateReader.getClusterState().getLiveNodes(), message, RANDOM);
    int numShards = backupCollectionState.getActiveSlices().size();
    int numNrtReplicas = getInt(message, NRT_REPLICAS, backupCollectionState.getNumNrtReplicas(), 0);
    if (numNrtReplicas == 0) {
        numNrtReplicas = getInt(message, REPLICATION_FACTOR, backupCollectionState.getReplicationFactor(), 0);
    }
    int numTlogReplicas = getInt(message, TLOG_REPLICAS, backupCollectionState.getNumTlogReplicas(), 0);
    int numPullReplicas = getInt(message, PULL_REPLICAS, backupCollectionState.getNumPullReplicas(), 0);
    int totalReplicasPerShard = numNrtReplicas + numTlogReplicas + numPullReplicas;
    int maxShardsPerNode = message.getInt(MAX_SHARDS_PER_NODE, backupCollectionState.getMaxShardsPerNode());
    int availableNodeCount = nodeList.size();
    if ((numShards * totalReplicasPerShard) > (availableNodeCount * maxShardsPerNode)) {
        throw new SolrException(ErrorCode.BAD_REQUEST, String.format(Locale.ROOT, "Solr cloud with available number of nodes:%d is insufficient for" + " restoring a collection with %d shards, total replicas per shard %d and maxShardsPerNode %d." + " Consider increasing maxShardsPerNode value OR number of available nodes.", availableNodeCount, numShards, totalReplicasPerShard, maxShardsPerNode));
    }
    //Upload the configs
    String configName = (String) properties.get(COLL_CONF);
    String restoreConfigName = message.getStr(COLL_CONF, configName);
    if (zkStateReader.getConfigManager().configExists(restoreConfigName)) {
        log.info("Using existing config {}", restoreConfigName);
    //TODO add overwrite option?
    } else {
        log.info("Uploading config {}", restoreConfigName);
        backupMgr.uploadConfigDir(location, backupName, configName, restoreConfigName);
    }
    log.info("Starting restore into collection={} with backup_name={} at location={}", restoreCollectionName, backupName, location);
    //Create core-less collection
    {
        Map<String, Object> propMap = new HashMap<>();
        propMap.put(Overseer.QUEUE_OPERATION, CREATE.toString());
        // mostly true.  Prevents autoCreated=true in the collection state.
        propMap.put("fromApi", "true");
        // inherit settings from input API, defaulting to the backup's setting.  Ex: replicationFactor
        for (String collProp : COLL_PROPS.keySet()) {
            Object val = message.getProperties().getOrDefault(collProp, backupCollectionState.get(collProp));
            if (val != null) {
                propMap.put(collProp, val);
            }
        }
        propMap.put(NAME, restoreCollectionName);
        //no cores
        propMap.put(CREATE_NODE_SET, CREATE_NODE_SET_EMPTY);
        propMap.put(COLL_CONF, restoreConfigName);
        // router.*
        @SuppressWarnings("unchecked") Map<String, Object> routerProps = (Map<String, Object>) backupCollectionState.getProperties().get(DocCollection.DOC_ROUTER);
        for (Map.Entry<String, Object> pair : routerProps.entrySet()) {
            propMap.put(DocCollection.DOC_ROUTER + "." + pair.getKey(), pair.getValue());
        }
        Set<String> sliceNames = backupCollectionState.getActiveSlicesMap().keySet();
        if (backupCollectionState.getRouter() instanceof ImplicitDocRouter) {
            propMap.put(SHARDS_PROP, StrUtils.join(sliceNames, ','));
        } else {
            propMap.put(NUM_SLICES, sliceNames.size());
            // ClusterStateMutator.createCollection detects that "slices" is in fact a slice structure instead of a
            //   list of names, and if so uses this instead of building it.  We clear the replica list.
            Collection<Slice> backupSlices = backupCollectionState.getActiveSlices();
            Map<String, Slice> newSlices = new LinkedHashMap<>(backupSlices.size());
            for (Slice backupSlice : backupSlices) {
                newSlices.put(backupSlice.getName(), new Slice(backupSlice.getName(), Collections.emptyMap(), backupSlice.getProperties()));
            }
            propMap.put(SHARDS_PROP, newSlices);
        }
        ocmh.commandMap.get(CREATE).call(zkStateReader.getClusterState(), new ZkNodeProps(propMap), new NamedList());
    // note: when createCollection() returns, the collection exists (no race)
    }
    DocCollection restoreCollection = zkStateReader.getClusterState().getCollection(restoreCollectionName);
    DistributedQueue inQueue = Overseer.getStateUpdateQueue(zkStateReader.getZkClient());
    //Mark all shards in CONSTRUCTION STATE while we restore the data
    {
        //TODO might instead createCollection accept an initial state?  Is there a race?
        Map<String, Object> propMap = new HashMap<>();
        propMap.put(Overseer.QUEUE_OPERATION, OverseerAction.UPDATESHARDSTATE.toLower());
        for (Slice shard : restoreCollection.getSlices()) {
            propMap.put(shard.getName(), Slice.State.CONSTRUCTION.toString());
        }
        propMap.put(ZkStateReader.COLLECTION_PROP, restoreCollectionName);
        inQueue.offer(Utils.toJSON(new ZkNodeProps(propMap)));
    }
    // TODO how do we leverage the RULE / SNITCH logic in createCollection?
    ClusterState clusterState = zkStateReader.getClusterState();
    List<String> sliceNames = new ArrayList<>();
    restoreCollection.getSlices().forEach(x -> sliceNames.add(x.getName()));
    Map<ReplicaAssigner.Position, String> positionVsNodes = ocmh.identifyNodes(clusterState, nodeList, message, sliceNames, numNrtReplicas, numTlogReplicas, numPullReplicas);
    //Create one replica per shard and copy backed up data to it
    for (Slice slice : restoreCollection.getSlices()) {
        log.debug("Adding replica for shard={} collection={} ", slice.getName(), restoreCollection);
        HashMap<String, Object> propMap = new HashMap<>();
        propMap.put(Overseer.QUEUE_OPERATION, CREATESHARD);
        propMap.put(COLLECTION_PROP, restoreCollectionName);
        propMap.put(SHARD_ID_PROP, slice.getName());
        if (numNrtReplicas >= 1) {
            propMap.put(REPLICA_TYPE, Replica.Type.NRT.name());
        } else if (numTlogReplicas >= 1) {
            propMap.put(REPLICA_TYPE, Replica.Type.TLOG.name());
        } else {
            throw new SolrException(SolrException.ErrorCode.BAD_REQUEST, "Unexpected number of replicas, replicationFactor, " + Replica.Type.NRT + " or " + Replica.Type.TLOG + " must be greater than 0");
        }
        // Get the first node matching the shard to restore in
        String node;
        for (Map.Entry<ReplicaAssigner.Position, String> pvn : positionVsNodes.entrySet()) {
            ReplicaAssigner.Position position = pvn.getKey();
            if (position.shard == slice.getName()) {
                node = pvn.getValue();
                propMap.put(CoreAdminParams.NODE, node);
                positionVsNodes.remove(position);
                break;
            }
        }
        // add async param
        if (asyncId != null) {
            propMap.put(ASYNC, asyncId);
        }
        ocmh.addPropertyParams(message, propMap);
        ocmh.addReplica(clusterState, new ZkNodeProps(propMap), new NamedList(), null);
    }
    //refresh the location copy of collection state
    restoreCollection = zkStateReader.getClusterState().getCollection(restoreCollectionName);
    //Copy data from backed up index to each replica
    for (Slice slice : restoreCollection.getSlices()) {
        ModifiableSolrParams params = new ModifiableSolrParams();
        params.set(CoreAdminParams.ACTION, CoreAdminParams.CoreAdminAction.RESTORECORE.toString());
        params.set(NAME, "snapshot." + slice.getName());
        params.set(CoreAdminParams.BACKUP_LOCATION, backupPath.toASCIIString());
        params.set(CoreAdminParams.BACKUP_REPOSITORY, repo);
        ocmh.sliceCmd(clusterState, params, null, slice, shardHandler, asyncId, requestMap);
    }
    ocmh.processResponses(new NamedList(), shardHandler, true, "Could not restore core", asyncId, requestMap);
    //Mark all shards in ACTIVE STATE
    {
        HashMap<String, Object> propMap = new HashMap<>();
        propMap.put(Overseer.QUEUE_OPERATION, OverseerAction.UPDATESHARDSTATE.toLower());
        propMap.put(ZkStateReader.COLLECTION_PROP, restoreCollectionName);
        for (Slice shard : restoreCollection.getSlices()) {
            propMap.put(shard.getName(), Slice.State.ACTIVE.toString());
        }
        inQueue.offer(Utils.toJSON(new ZkNodeProps(propMap)));
    }
    //refresh the location copy of collection state
    restoreCollection = zkStateReader.getClusterState().getCollection(restoreCollectionName);
    if (totalReplicasPerShard > 1) {
        log.info("Adding replicas to restored collection={}", restoreCollection);
        for (Slice slice : restoreCollection.getSlices()) {
            //Add the remaining replicas for each shard, considering it's type
            int createdNrtReplicas = 0, createdTlogReplicas = 0, createdPullReplicas = 0;
            // We already created either a NRT or an TLOG replica as leader
            if (numNrtReplicas > 0) {
                createdNrtReplicas++;
            } else if (createdTlogReplicas > 0) {
                createdTlogReplicas++;
            }
            for (int i = 1; i < totalReplicasPerShard; i++) {
                Replica.Type typeToCreate;
                if (createdNrtReplicas < numNrtReplicas) {
                    createdNrtReplicas++;
                    typeToCreate = Replica.Type.NRT;
                } else if (createdTlogReplicas < numTlogReplicas) {
                    createdTlogReplicas++;
                    typeToCreate = Replica.Type.TLOG;
                } else {
                    createdPullReplicas++;
                    typeToCreate = Replica.Type.PULL;
                    assert createdPullReplicas <= numPullReplicas : "Unexpected number of replicas";
                }
                log.debug("Adding replica for shard={} collection={} of type {} ", slice.getName(), restoreCollection, typeToCreate);
                HashMap<String, Object> propMap = new HashMap<>();
                propMap.put(COLLECTION_PROP, restoreCollectionName);
                propMap.put(SHARD_ID_PROP, slice.getName());
                propMap.put(REPLICA_TYPE, typeToCreate.name());
                // Get the first node matching the shard to restore in
                String node;
                for (Map.Entry<ReplicaAssigner.Position, String> pvn : positionVsNodes.entrySet()) {
                    ReplicaAssigner.Position position = pvn.getKey();
                    if (position.shard == slice.getName()) {
                        node = pvn.getValue();
                        propMap.put(CoreAdminParams.NODE, node);
                        positionVsNodes.remove(position);
                        break;
                    }
                }
                // add async param
                if (asyncId != null) {
                    propMap.put(ASYNC, asyncId);
                }
                ocmh.addPropertyParams(message, propMap);
                ocmh.addReplica(zkStateReader.getClusterState(), new ZkNodeProps(propMap), results, null);
            }
        }
    }
    log.info("Completed restoring collection={} backupName={}", restoreCollection, backupName);
}