Examples with SolrZkClient org.apache.solr.common.cloud.SolrZkClient used on opensource projects

Example 6 with SolrZkClient

use of org.apache.solr.common.cloud.SolrZkClient in project lucene-solr by apache.

the class BackupCmd method copyIndexFiles.

private void copyIndexFiles(URI backupPath, ZkNodeProps request, NamedList results) throws Exception {
    String collectionName = request.getStr(COLLECTION_PROP);
    String backupName = request.getStr(NAME);
    String asyncId = request.getStr(ASYNC);
    String repoName = request.getStr(CoreAdminParams.BACKUP_REPOSITORY);
    ShardHandler shardHandler = ocmh.shardHandlerFactory.getShardHandler();
    Map<String, String> requestMap = new HashMap<>();
    String commitName = request.getStr(CoreAdminParams.COMMIT_NAME);
    Optional<CollectionSnapshotMetaData> snapshotMeta = Optional.empty();
    if (commitName != null) {
        SolrZkClient zkClient = ocmh.overseer.getZkController().getZkClient();
        snapshotMeta = SolrSnapshotManager.getCollectionLevelSnapshot(zkClient, collectionName, commitName);
        if (!snapshotMeta.isPresent()) {
            throw new SolrException(ErrorCode.BAD_REQUEST, "Snapshot with name " + commitName + " does not exist for collection " + collectionName);
        }
        if (snapshotMeta.get().getStatus() != SnapshotStatus.Successful) {
            throw new SolrException(ErrorCode.BAD_REQUEST, "Snapshot with name " + commitName + " for collection " + collectionName + " has not completed successfully. The status is " + snapshotMeta.get().getStatus());
        }
    }
    log.info("Starting backup of collection={} with backupName={} at location={}", collectionName, backupName, backupPath);
    Collection<String> shardsToConsider = Collections.emptySet();
    if (snapshotMeta.isPresent()) {
        shardsToConsider = snapshotMeta.get().getShards();
    }
    for (Slice slice : ocmh.zkStateReader.getClusterState().getCollection(collectionName).getActiveSlices()) {
        Replica replica = null;
        if (snapshotMeta.isPresent()) {
            if (!shardsToConsider.contains(slice.getName())) {
                log.warn("Skipping the backup for shard {} since it wasn't part of the collection {} when snapshot {} was created.", slice.getName(), collectionName, snapshotMeta.get().getName());
                continue;
            }
            replica = selectReplicaWithSnapshot(snapshotMeta.get(), slice);
        } else {
            // Note - Actually this can return a null value when there is no leader for this shard.
            replica = slice.getLeader();
            if (replica == null) {
                throw new SolrException(ErrorCode.SERVER_ERROR, "No 'leader' replica available for shard " + slice.getName() + " of collection " + collectionName);
            }
        }
        String coreName = replica.getStr(CORE_NAME_PROP);
        ModifiableSolrParams params = new ModifiableSolrParams();
        params.set(CoreAdminParams.ACTION, CoreAdminParams.CoreAdminAction.BACKUPCORE.toString());
        params.set(NAME, slice.getName());
        params.set(CoreAdminParams.BACKUP_REPOSITORY, repoName);
        // note: index dir will be here then the "snapshot." + slice name
        params.set(CoreAdminParams.BACKUP_LOCATION, backupPath.toASCIIString());
        params.set(CORE_NAME_PROP, coreName);
        if (snapshotMeta.isPresent()) {
            params.set(CoreAdminParams.COMMIT_NAME, snapshotMeta.get().getName());
        }
        ocmh.sendShardRequest(replica.getNodeName(), params, shardHandler, asyncId, requestMap);
        log.debug("Sent backup request to core={} for backupName={}", coreName, backupName);
    }
    log.debug("Sent backup requests to all shard leaders for backupName={}", backupName);
    ocmh.processResponses(results, shardHandler, true, "Could not backup all replicas", asyncId, requestMap);
}

Example 7 with SolrZkClient

use of org.apache.solr.common.cloud.SolrZkClient in project lucene-solr by apache.

the class DeleteCollectionCmd method call.

@Override
public void call(ClusterState state, ZkNodeProps message, NamedList results) throws Exception {
    ZkStateReader zkStateReader = ocmh.zkStateReader;
    final String collection = message.getStr(NAME);
    try {
        // Remove the snapshots meta-data for this collection in ZK. Deleting actual index files
        // should be taken care of as part of collection delete operation.
        SolrZkClient zkClient = zkStateReader.getZkClient();
        SolrSnapshotManager.cleanupCollectionLevelSnapshots(zkClient, collection);
        if (zkStateReader.getClusterState().getCollectionOrNull(collection) == null) {
            if (zkStateReader.getZkClient().exists(ZkStateReader.COLLECTIONS_ZKNODE + "/" + collection, true)) {
                // is not in the clusterstate
                return;
            }
        }
        ModifiableSolrParams params = new ModifiableSolrParams();
        params.set(CoreAdminParams.ACTION, CoreAdminParams.CoreAdminAction.UNLOAD.toString());
        params.set(CoreAdminParams.DELETE_INSTANCE_DIR, true);
        params.set(CoreAdminParams.DELETE_DATA_DIR, true);
        String asyncId = message.getStr(ASYNC);
        Map<String, String> requestMap = null;
        if (asyncId != null) {
            requestMap = new HashMap<>();
        }
        Set<String> okayExceptions = new HashSet<>(1);
        okayExceptions.add(NonExistentCoreException.class.getName());
        ocmh.collectionCmd(message, params, results, null, asyncId, requestMap, okayExceptions);
        ZkNodeProps m = new ZkNodeProps(Overseer.QUEUE_OPERATION, DELETE.toLower(), NAME, collection);
        Overseer.getStateUpdateQueue(zkStateReader.getZkClient()).offer(Utils.toJSON(m));
        // wait for a while until we don't see the collection
        TimeOut timeout = new TimeOut(30, TimeUnit.SECONDS);
        boolean removed = false;
        while (!timeout.hasTimedOut()) {
            Thread.sleep(100);
            removed = !zkStateReader.getClusterState().hasCollection(collection);
            if (removed) {
                // just a bit of time so it's more likely other
                Thread.sleep(500);
                // readers see on return
                break;
            }
        }
        if (!removed) {
            throw new SolrException(SolrException.ErrorCode.SERVER_ERROR, "Could not fully remove collection: " + collection);
        }
    } finally {
        try {
            if (zkStateReader.getZkClient().exists(ZkStateReader.COLLECTIONS_ZKNODE + "/" + collection, true)) {
                zkStateReader.getZkClient().clean(ZkStateReader.COLLECTIONS_ZKNODE + "/" + collection);
            }
        } catch (InterruptedException e) {
            SolrException.log(log, "Cleaning up collection in zk was interrupted:" + collection, e);
            Thread.currentThread().interrupt();
        } catch (KeeperException e) {
            SolrException.log(log, "Problem cleaning up collection in zk:" + collection, e);
        }
    }
}

Example 8 with SolrZkClient

use of org.apache.solr.common.cloud.SolrZkClient in project lucene-solr by apache.

the class DeleteSnapshotCmd method call.

@Override
public void call(ClusterState state, ZkNodeProps message, NamedList results) throws Exception {
    String collectionName = message.getStr(COLLECTION_PROP);
    String commitName = message.getStr(CoreAdminParams.COMMIT_NAME);
    String asyncId = message.getStr(ASYNC);
    Map<String, String> requestMap = new HashMap<>();
    NamedList shardRequestResults = new NamedList();
    ShardHandler shardHandler = ocmh.shardHandlerFactory.getShardHandler();
    SolrZkClient zkClient = ocmh.overseer.getZkController().getZkClient();
    Optional<CollectionSnapshotMetaData> meta = SolrSnapshotManager.getCollectionLevelSnapshot(zkClient, collectionName, commitName);
    if (!meta.isPresent()) {
        // Snapshot not found. Nothing to do.
        return;
    }
    log.info("Deleting a snapshot for collection={} with commitName={}", collectionName, commitName);
    Set<String> existingCores = new HashSet<>();
    for (Slice s : ocmh.zkStateReader.getClusterState().getCollection(collectionName).getSlices()) {
        for (Replica r : s.getReplicas()) {
            existingCores.add(r.getCoreName());
        }
    }
    Set<String> coresWithSnapshot = new HashSet<>();
    for (CoreSnapshotMetaData m : meta.get().getReplicaSnapshots()) {
        if (existingCores.contains(m.getCoreName())) {
            coresWithSnapshot.add(m.getCoreName());
        }
    }
    log.info("Existing cores with snapshot for collection={} are {}", collectionName, existingCores);
    for (Slice slice : ocmh.zkStateReader.getClusterState().getCollection(collectionName).getSlices()) {
        for (Replica replica : slice.getReplicas()) {
            if (replica.getState() == State.DOWN) {
                // Since replica is down - no point sending a request.
                continue;
            }
            // replicas to contact at this point, we try on all replicas.
            if (meta.get().getStatus() == SnapshotStatus.InProgress || coresWithSnapshot.contains(replica.getCoreName())) {
                String coreName = replica.getStr(CORE_NAME_PROP);
                ModifiableSolrParams params = new ModifiableSolrParams();
                params.set(CoreAdminParams.ACTION, CoreAdminAction.DELETESNAPSHOT.toString());
                params.set(NAME, slice.getName());
                params.set(CORE_NAME_PROP, coreName);
                params.set(CoreAdminParams.COMMIT_NAME, commitName);
                log.info("Sending deletesnapshot request to core={} with commitName={}", coreName, commitName);
                ocmh.sendShardRequest(replica.getNodeName(), params, shardHandler, asyncId, requestMap);
            }
        }
    }
    ocmh.processResponses(shardRequestResults, shardHandler, false, null, asyncId, requestMap);
    NamedList success = (NamedList) shardRequestResults.get("success");
    List<CoreSnapshotMetaData> replicas = new ArrayList<>();
    if (success != null) {
        for (int i = 0; i < success.size(); i++) {
            NamedList resp = (NamedList) success.getVal(i);
            // Unfortunately async processing logic doesn't provide the "core" name automatically.
            String coreName = (String) resp.get("core");
            coresWithSnapshot.remove(coreName);
        }
    }
    if (!coresWithSnapshot.isEmpty()) {
        // One or more failures.
        log.warn("Failed to delete a snapshot for collection {} with commitName = {}. Snapshot could not be deleted for following cores {}", collectionName, commitName, coresWithSnapshot);
        List<CoreSnapshotMetaData> replicasWithSnapshot = new ArrayList<>();
        for (CoreSnapshotMetaData m : meta.get().getReplicaSnapshots()) {
            if (coresWithSnapshot.contains(m.getCoreName())) {
                replicasWithSnapshot.add(m);
            }
        }
        // Update the ZK meta-data to include only cores with the snapshot. This will enable users to figure out
        // which cores still contain the named snapshot.
        CollectionSnapshotMetaData newResult = new CollectionSnapshotMetaData(meta.get().getName(), SnapshotStatus.Failed, meta.get().getCreationDate(), replicasWithSnapshot);
        SolrSnapshotManager.updateCollectionLevelSnapshot(zkClient, collectionName, newResult);
        log.info("Saved snapshot information for collection={} with commitName={} in Zookeeper as follows", collectionName, commitName, Utils.toJSON(newResult));
        throw new SolrException(ErrorCode.SERVER_ERROR, "Failed to delete snapshot on cores " + coresWithSnapshot);
    } else {
        // Delete the ZK path so that we eliminate the references of this snapshot from collection level meta-data.
        SolrSnapshotManager.deleteCollectionLevelSnapshot(zkClient, collectionName, commitName);
        log.info("Deleted Zookeeper snapshot metdata for collection={} with commitName={}", collectionName, commitName);
        log.info("Successfully deleted snapshot for collection={} with commitName={}", collectionName, commitName);
    }
}

Example 9 with SolrZkClient

use of org.apache.solr.common.cloud.SolrZkClient in project lucene-solr by apache.

the class OverseerCollectionMessageHandler method processReplicaDeletePropertyCommand.

private void processReplicaDeletePropertyCommand(ClusterState clusterState, ZkNodeProps message, NamedList results) throws KeeperException, InterruptedException {
    checkRequired(message, COLLECTION_PROP, SHARD_ID_PROP, REPLICA_PROP, PROPERTY_PROP);
    SolrZkClient zkClient = zkStateReader.getZkClient();
    DistributedQueue inQueue = Overseer.getStateUpdateQueue(zkClient);
    Map<String, Object> propMap = new HashMap<>();
    propMap.put(Overseer.QUEUE_OPERATION, DELETEREPLICAPROP.toLower());
    propMap.putAll(message.getProperties());
    ZkNodeProps m = new ZkNodeProps(propMap);
    inQueue.offer(Utils.toJSON(m));
}

Example 10 with SolrZkClient

use of org.apache.solr.common.cloud.SolrZkClient in project lucene-solr by apache.

the class CdcrBufferStateManager method createStateNode.

private void createStateNode() {
    SolrZkClient zkClient = core.getCoreContainer().getZkController().getZkClient();
    try {
        if (!zkClient.exists(this.getZnodePath(), true)) {
            if (!zkClient.exists(this.getZnodeBase(), true)) {
                // Should be a no-op if node exists
                zkClient.makePath(this.getZnodeBase(), null, CreateMode.PERSISTENT, null, false, true);
            }
            zkClient.create(this.getZnodePath(), DEFAULT_STATE.getBytes(), CreateMode.PERSISTENT, true);
            log.info("Created znode {}", this.getZnodePath());
        }
    } catch (KeeperException.NodeExistsException ne) {
    // Someone got in first and created the node.
    } catch (KeeperException | InterruptedException e) {
        log.warn("Failed to create CDCR buffer state node", e);
    }
}