org.apache.cassandra.service.StorageProxy Java Examples

private void mutate(List<org.apache.cassandra.db.Mutation> cmds, org.apache.cassandra.db.ConsistencyLevel clvl) throws BackendException {
    try {
        schedule(DatabaseDescriptor.getRpcTimeout());
        try {
            if (atomicBatch) {
                StorageProxy.mutateAtomically(cmds, clvl);
            } else {
                StorageProxy.mutate(cmds, clvl);
            }
        } catch (RequestExecutionException e) {
            throw new TemporaryBackendException(e);
        } finally {
            release();
        }
    } catch (TimeoutException ex) {
        log.debug("Cassandra TimeoutException", ex);
        throw new TemporaryBackendException(ex);
    }
}
 

private void doInsert(ConsistencyLevel consistency_level, List<? extends IMutation> mutations, boolean mutateAtomically)
throws UnavailableException, TimedOutException, org.apache.cassandra.exceptions.InvalidRequestException
{
    org.apache.cassandra.db.ConsistencyLevel consistencyLevel = ThriftConversion.fromThrift(consistency_level);
    consistencyLevel.validateForWrite(state().getKeyspace());
    if (mutations.isEmpty())
        return;

    long timeout = Long.MAX_VALUE;
    for (IMutation m : mutations)
        timeout = Longs.min(timeout, m.getTimeout());

    schedule(timeout);
    try
    {
        StorageProxy.mutateWithTriggers(mutations, consistencyLevel, mutateAtomically);
    }
    catch (RequestExecutionException e)
    {
        ThriftConversion.rethrow(e);
    }
    finally
    {
        release();
    }
}
 

private void writeHintsForUndeliveredEndpoints(int startFrom)
{
    try
    {
        // Here we deserialize mutations 2nd time from byte buffer.
        // but this is ok, because timeout on batch direct delivery is rare
        // (it can happen only several seconds until node is marked dead)
        // so trading some cpu to keep less objects
        List<Mutation> replayingMutations = replayingMutations();
        for (int i = startFrom; i < replayHandlers.size(); i++)
        {
            Mutation undeliveredMutation = replayingMutations.get(i);
            int ttl = calculateHintTTL(replayingMutations);
            ReplayWriteResponseHandler handler = replayHandlers.get(i);

            if (ttl > 0 && handler != null)
                for (InetAddress endpoint : handler.undelivered)
                    StorageProxy.writeHintForMutation(undeliveredMutation, writtenAt, ttl, endpoint);
        }
    }
    catch (IOException e)
    {
        logger.error("Cannot schedule hints for undelivered batch", e);
    }
}
 

protected List<Row> queryNextPage(int pageSize, ConsistencyLevel consistencyLevel, boolean localQuery)
throws RequestValidationException, RequestExecutionException
{
    // For some queries, such as a DISTINCT query on static columns, the limit for slice queries will be lower
    // than the page size (in the static example, it will be 1).  We use the min here to ensure we don't fetch
    // more rows than we're supposed to.  See CASSANDRA-8108 for more details.
    SliceQueryFilter filter = command.filter.withUpdatedCount(Math.min(command.filter.count, pageSize));
    if (lastReturned != null)
        filter = filter.withUpdatedStart(lastReturned, cfm.comparator);

    logger.debug("Querying next page of slice query; new filter: {}", filter);
    ReadCommand pageCmd = command.withUpdatedFilter(filter);
    return localQuery
         ? Collections.singletonList(pageCmd.getRow(Keyspace.open(command.ksName)))
         : StorageProxy.read(Collections.singletonList(pageCmd), consistencyLevel, cstate);
}
 

protected List<Row> queryNextPage(int pageSize, ConsistencyLevel consistencyLevel, boolean localQuery)
throws RequestExecutionException
{
    SliceQueryFilter sf = (SliceQueryFilter)columnFilter;
    AbstractBounds<RowPosition> keyRange = lastReturnedKey == null ? command.keyRange : makeIncludingKeyBounds(lastReturnedKey);
    Composite start = lastReturnedName == null ? sf.start() : lastReturnedName;
    PagedRangeCommand pageCmd = new PagedRangeCommand(command.keyspace,
                                                      command.columnFamily,
                                                      command.timestamp,
                                                      keyRange,
                                                      sf,
                                                      start,
                                                      sf.finish(),
                                                      command.rowFilter,
                                                      pageSize,
                                                      command.countCQL3Rows);

    return localQuery
         ? pageCmd.executeLocally()
         : StorageProxy.getRangeSlice(pageCmd, consistencyLevel);
}
 

private ResultMessage.Rows execute(Pageable command, QueryOptions options, int limit, long now, QueryState state) throws RequestValidationException, RequestExecutionException
{
    List<Row> rows;
    if (command == null)
    {
        rows = Collections.<Row>emptyList();
    }
    else
    {
        rows = command instanceof Pageable.ReadCommands
             ? StorageProxy.read(((Pageable.ReadCommands)command).commands, options.getConsistency(), state.getClientState())
             : StorageProxy.getRangeSlice((RangeSliceCommand)command, options.getConsistency());
    }

    return processResults(rows, options, limit, now);
}
 

public ResultMessage executeWithCondition(QueryState queryState, QueryOptions options)
throws RequestExecutionException, RequestValidationException
{
    List<ByteBuffer> keys = buildPartitionKeyNames(options);
    // We don't support IN for CAS operation so far
    if (keys.size() > 1)
        throw new InvalidRequestException("IN on the partition key is not supported with conditional updates");

    ByteBuffer key = keys.get(0);
    long now = options.getTimestamp(queryState);
    Composite prefix = createClusteringPrefix(options);

    CQL3CasRequest request = new CQL3CasRequest(cfm, key, false);
    addConditions(prefix, request, options);
    request.addRowUpdate(prefix, this, options, now);

    ColumnFamily result = StorageProxy.cas(keyspace(),
                                           columnFamily(),
                                           key,
                                           request,
                                           options.getSerialConsistency(),
                                           options.getConsistency(),
                                           queryState.getClientState());
    return new ResultMessage.Rows(buildCasResultSet(key, result, options));
}
 

private void executeWithoutConditions(Collection<? extends IMutation> mutations, ConsistencyLevel cl) throws RequestExecutionException, RequestValidationException
{
    // Extract each collection of cfs from it's IMutation and then lazily concatenate all of them into a single Iterable.
    Iterable<ColumnFamily> cfs = Iterables.concat(Iterables.transform(mutations, new Function<IMutation, Collection<ColumnFamily>>()
    {
        public Collection<ColumnFamily> apply(IMutation im)
        {
            return im.getColumnFamilies();
        }
    }));
    verifyBatchSize(cfs);

    boolean mutateAtomic = (type == Type.LOGGED && mutations.size() > 1);
    StorageProxy.mutateWithTriggers(mutations, cl, mutateAtomic);
}
 

private ResultMessage executeWithoutCondition(QueryState queryState, QueryOptions options)
throws RequestExecutionException, RequestValidationException
{
    ConsistencyLevel cl = options.getConsistency();
    if (isCounter())
        cl.validateCounterForWrite(cfm);
    else
        cl.validateForWrite(cfm.ksName);

    Collection<? extends IMutation> mutations = getMutations(options, false, options.getTimestamp(queryState));
    if (!mutations.isEmpty())
        StorageProxy.mutateWithTriggers(mutations, cl, false);

    return null;
}
 

protected List<Row> queryNextPage(int pageSize, ConsistencyLevel consistencyLevel, boolean localQuery)
throws RequestExecutionException
{
    AbstractRangeCommand pageCmd = command.withUpdatedLimit(pageSize);
    if (lastReturnedKey != null)
        pageCmd = pageCmd.forSubRange(makeExcludingKeyBounds(lastReturnedKey));

    return localQuery
         ? pageCmd.executeLocally()
         : StorageProxy.getRangeSlice(pageCmd, consistencyLevel);
}
 

private void retryDummyRead(String ks, String cf) throws PermanentBackendException {

        final long limit = System.currentTimeMillis() + (60L * 1000L);

        while (System.currentTimeMillis() < limit) {
            try {
                SortedSet<CellName> names = new TreeSet<>(new Comparator<CellName>() {
                    // This is a singleton set.  We need to define a comparator because SimpleDenseCellName is not
                    // comparable, but it doesn't have to be a useful comparator
                    @Override
                    public int compare(CellName o1, CellName o2)
                    {
                        return 0;
                    }
                });
                names.add(CellNames.simpleDense(ByteBufferUtil.zeroByteBuffer(1)));
                NamesQueryFilter nqf = new NamesQueryFilter(names);
                SliceByNamesReadCommand cmd = new SliceByNamesReadCommand(ks, ByteBufferUtil.zeroByteBuffer(1), cf, 1L, nqf);
                StorageProxy.read(ImmutableList.<ReadCommand> of(cmd), ConsistencyLevel.QUORUM);
                log.info("Read on CF {} in KS {} succeeded", cf, ks);
                return;
            } catch (Throwable t) {
                log.warn("Failed to read CF {} in KS {} following creation", cf, ks, t);
            }

            try {
                Thread.sleep(1000L);
            } catch (InterruptedException e) {
                throw new PermanentBackendException(e);
            }
        }

        throw new PermanentBackendException("Timed out while attempting to read CF " + cf + " in KS " + ks + " following creation");
    }
 

public void doVerb(final MessageIn<CounterMutation> message, final int id)
{
    try
    {
        final CounterMutation cm = message.payload;
        logger.debug("Applying forwarded {}", cm);

        String localDataCenter = DatabaseDescriptor.getEndpointSnitch().getDatacenter(FBUtilities.getBroadcastAddress());
        // We should not wait for the result of the write in this thread,
        // otherwise we could have a distributed deadlock between replicas
        // running this VerbHandler (see #4578).
        // Instead, we use a callback to send the response. Note that the callback
        // will not be called if the request timeout, but this is ok
        // because the coordinator of the counter mutation will timeout on
        // it's own in that case.
        StorageProxy.applyCounterMutationOnLeader(cm, localDataCenter, new Runnable()
        {
            public void run()
            {
                MessagingService.instance().sendReply(new WriteResponse().createMessage(), id, message.from);
            }
        });
    }
    catch (RequestExecutionException e)
    {
        // The coordinator will timeout on it's own so ignore
        logger.debug("counter error", e);
    }
}
 

protected Map<DecoratedKey, ColumnFamily> readColumnFamily(List<ReadCommand> commands, org.apache.cassandra.db.ConsistencyLevel consistency_level, ClientState cState)
throws org.apache.cassandra.exceptions.InvalidRequestException, UnavailableException, TimedOutException
{
    // TODO - Support multiple column families per row, right now row only contains 1 column family
    Map<DecoratedKey, ColumnFamily> columnFamilyKeyMap = new HashMap<DecoratedKey, ColumnFamily>();

    List<Row> rows = null;
    try
    {
        schedule(DatabaseDescriptor.getReadRpcTimeout());
        try
        {
            rows = StorageProxy.read(commands, consistency_level, cState);
        }
        finally
        {
            release();
        }
    }
    catch (RequestExecutionException e)
    {
        ThriftConversion.rethrow(e);
    }

    for (Row row: rows)
    {
        columnFamilyKeyMap.put(row.key, row.cf);
    }
    return columnFamilyKeyMap;
}
 

public boolean shouldHint()
{
    return allowHints
        && sentMessage.verb != MessagingService.Verb.COUNTER_MUTATION
        && consistencyLevel != ConsistencyLevel.ANY
        && StorageProxy.shouldHint(target);
}
 

private void maybeTruncateSafely(RequiresColumnFamily rcf) {
  if ( rcf != null && rcf.truncateExisting() ) {
    try {
      StorageProxy.truncateBlocking(rcf.ksName(), rcf.cfName());
    } catch (Exception ex) {
      throw new RuntimeException("Could not truncate column family: " + rcf.cfName(),ex);
    }
  }
}
 

private ResultMessage executeWithConditions(BatchQueryOptions options, QueryState state)
throws RequestExecutionException, RequestValidationException
{
    long now = state.getTimestamp();
    ByteBuffer key = null;
    String ksName = null;
    String cfName = null;
    CQL3CasRequest casRequest = null;
    Set<ColumnDefinition> columnsWithConditions = new LinkedHashSet<>();

    for (int i = 0; i < statements.size(); i++)
    {
        ModificationStatement statement = statements.get(i);
        QueryOptions statementOptions = options.forStatement(i);
        long timestamp = attrs.getTimestamp(now, statementOptions);
        List<ByteBuffer> pks = statement.buildPartitionKeyNames(statementOptions);
        if (pks.size() > 1)
            throw new IllegalArgumentException("Batch with conditions cannot span multiple partitions (you cannot use IN on the partition key)");
        if (key == null)
        {
            key = pks.get(0);
            ksName = statement.cfm.ksName;
            cfName = statement.cfm.cfName;
            casRequest = new CQL3CasRequest(statement.cfm, key, true);
        }
        else if (!key.equals(pks.get(0)))
        {
            throw new InvalidRequestException("Batch with conditions cannot span multiple partitions");
        }

        Composite clusteringPrefix = statement.createClusteringPrefix(statementOptions);
        if (statement.hasConditions())
        {
            statement.addConditions(clusteringPrefix, casRequest, statementOptions);
            // As soon as we have a ifNotExists, we set columnsWithConditions to null so that everything is in the resultSet
            if (statement.hasIfNotExistCondition() || statement.hasIfExistCondition())
                columnsWithConditions = null;
            else if (columnsWithConditions != null)
                Iterables.addAll(columnsWithConditions, statement.getColumnsWithConditions());
        }
        casRequest.addRowUpdate(clusteringPrefix, statement, statementOptions, timestamp);
    }

    ColumnFamily result = StorageProxy.cas(ksName, cfName, key, casRequest, options.getSerialConsistency(), options.getConsistency(), state.getClientState());

    return new ResultMessage.Rows(ModificationStatement.buildCasResultSet(ksName, key, cfName, result, columnsWithConditions, true, options.forStatement(0)));
}
 

private static List<org.apache.cassandra.db.Row> multiRangeSlice(CFMetaData metadata, SelectStatement select, List<ByteBuffer> variables, long now)
throws ReadTimeoutException, UnavailableException, InvalidRequestException
{
    IPartitioner p = StorageService.getPartitioner();

    AbstractType<?> keyType = Schema.instance.getCFMetaData(metadata.ksName, select.getColumnFamily()).getKeyValidator();

    ByteBuffer startKeyBytes = (select.getKeyStart() != null)
                               ? select.getKeyStart().getByteBuffer(keyType,variables)
                               : null;

    ByteBuffer finishKeyBytes = (select.getKeyFinish() != null)
                                ? select.getKeyFinish().getByteBuffer(keyType,variables)
                                : null;

    RowPosition startKey = RowPosition.ForKey.get(startKeyBytes, p), finishKey = RowPosition.ForKey.get(finishKeyBytes, p);
    if (startKey.compareTo(finishKey) > 0 && !finishKey.isMinimum(p))
    {
        if (p instanceof RandomPartitioner)
            throw new InvalidRequestException("Start key sorts after end key. This is not allowed; you probably should not specify end key at all, under RandomPartitioner");
        else
            throw new InvalidRequestException("Start key must sort before (or equal to) finish key in your partitioner!");
    }
    AbstractBounds<RowPosition> bounds = new Bounds<RowPosition>(startKey, finishKey);

    IDiskAtomFilter columnFilter = filterFromSelect(select, metadata, variables);
    validateFilter(metadata, columnFilter);

    List<Relation> columnRelations = select.getColumnRelations();
    List<IndexExpression> expressions = new ArrayList<IndexExpression>(columnRelations.size());
    for (Relation columnRelation : columnRelations)
    {
        // Left and right side of relational expression encoded according to comparator/validator.
        ByteBuffer entity = columnRelation.getEntity().getByteBuffer(metadata.comparator.asAbstractType(), variables);
        ByteBuffer value = columnRelation.getValue().getByteBuffer(metadata.getValueValidator(metadata.comparator.cellFromByteBuffer(entity)), variables);

        expressions.add(new IndexExpression(entity,
                                            Operator.valueOf(columnRelation.operator().name()),
                                            value));
    }

    int limit = select.isKeyRange() && select.getKeyStart() != null
              ? select.getNumRecords() + 1
              : select.getNumRecords();

    List<org.apache.cassandra.db.Row> rows = StorageProxy.getRangeSlice(new RangeSliceCommand(metadata.ksName,
                                                                                              select.getColumnFamily(),
                                                                                              now,
                                                                                              columnFilter,
                                                                                              bounds,
                                                                                              expressions,
                                                                                              limit),
                                                                        select.getConsistencyLevel());

    // if start key was set and relation was "greater than"
    if (select.getKeyStart() != null && !select.includeStartKey() && !rows.isEmpty())
    {
        if (rows.get(0).key.getKey().equals(startKeyBytes))
            rows.remove(0);
    }

    // if finish key was set and relation was "less than"
    if (select.getKeyFinish() != null && !select.includeFinishKey() && !rows.isEmpty())
    {
        int lastIndex = rows.size() - 1;
        if (rows.get(lastIndex).key.getKey().equals(finishKeyBytes))
            rows.remove(lastIndex);
    }

    return rows.subList(0, select.getNumRecords() < rows.size() ? select.getNumRecords() : rows.size());
}
 

/**
 * Create a RangeSliceCommand and run it against the StorageProxy.
 * <p>
 * To match the behavior of the standard Cassandra thrift API endpoint, the
 * {@code nowMillis} argument should be the number of milliseconds since the
 * UNIX Epoch (e.g. System.currentTimeMillis() or equivalent obtained
 * through a {@link TimestampProvider}). This is per
 * {@link org.apache.cassandra.thrift.CassandraServer#get_range_slices(ColumnParent, SlicePredicate, KeyRange, ConsistencyLevel)},
 * which passes the server's System.currentTimeMillis() to the
 * {@code RangeSliceCommand} constructor.
 */
private List<Row> getKeySlice(Token start,
                              Token end,
                              @Nullable SliceQuery sliceQuery,
                              int pageSize,
                              long nowMillis) throws BackendException {
    IPartitioner partitioner = StorageService.getPartitioner();

    SliceRange columnSlice = new SliceRange();
    if (sliceQuery == null) {
        columnSlice.setStart(ArrayUtils.EMPTY_BYTE_ARRAY)
                .setFinish(ArrayUtils.EMPTY_BYTE_ARRAY)
                .setCount(5);
    } else {
        columnSlice.setStart(sliceQuery.getSliceStart().asByteBuffer())
                .setFinish(sliceQuery.getSliceEnd().asByteBuffer())
                .setCount(sliceQuery.hasLimit() ? sliceQuery.getLimit() : Integer.MAX_VALUE);
    }
    /* Note: we need to fetch columns for each row as well to remove "range ghosts" */
    SlicePredicate predicate = new SlicePredicate().setSlice_range(columnSlice);

    // DAVID CASSANDRA
    // Old cassandra code did not use partitioner anyway in this call...so new code removed it as a parmaeter
    // RowPosition startPosition = start.minKeyBound(partitioner);
    RowPosition startPosition = start.minKeyBound();
    // DAVID CASSANDRA
    // RowPosition endPosition = end.minKeyBound(partitioner);
    RowPosition endPosition = end.minKeyBound();

    List<Row> rows;

    try {
        CFMetaData cfm = Schema.instance.getCFMetaData(keyspace, columnFamily);
        IDiskAtomFilter filter = ThriftValidation.asIFilter(predicate, cfm, null);

        RangeSliceCommand cmd = new RangeSliceCommand(keyspace, columnFamily, nowMillis, filter, new Bounds<RowPosition>(startPosition, endPosition), pageSize);

        rows = StorageProxy.getRangeSlice(cmd, ConsistencyLevel.QUORUM);
    } catch (Exception e) {
        throw new PermanentBackendException(e);
    }

    return rows;
}
 

public Map<String, List<String>> describe_schema_versions() throws TException, InvalidRequestException
{
    logger.debug("checking schema agreement");
    return StorageProxy.describeSchemaVersions();
}