Search in sources :

Example 1 with ConsumerConfig

use of co.cask.cdap.data2.queue.ConsumerConfig in project cdap by caskdata.

the class HBaseStreamFileConsumerFactory method create.

@Override
protected StreamConsumer create(TableId tableId, StreamConfig streamConfig, ConsumerConfig consumerConfig, StreamConsumerStateStore stateStore, StreamConsumerState beginConsumerState, FileReader<StreamEventOffset, Iterable<StreamFileOffset>> reader, @Nullable ReadFilter extraFilter) throws IOException {
    int splits = cConf.getInt(Constants.Stream.CONSUMER_TABLE_PRESPLITS);
    AbstractRowKeyDistributor distributor = new RowKeyDistributorByHashPrefix(new RowKeyDistributorByHashPrefix.OneByteSimpleHash(splits));
    byte[][] splitKeys = HBaseTableUtil.getSplitKeys(splits, splits, distributor);
    TableId hBaseTableId = tableUtil.createHTableId(new NamespaceId(tableId.getNamespace()), tableId.getTableName());
    TableDescriptorBuilder tdBuilder = HBaseTableUtil.getTableDescriptorBuilder(hBaseTableId, cConf);
    ColumnFamilyDescriptorBuilder cfdBuilder = HBaseTableUtil.getColumnFamilyDescriptorBuilder(Bytes.toString(QueueEntryRow.COLUMN_FAMILY), hConf);
    tdBuilder.addColumnFamily(cfdBuilder.build());
    tdBuilder.addProperty(QueueConstants.DISTRIBUTOR_BUCKETS, Integer.toString(splits));
    try (HBaseDDLExecutor ddlExecutor = ddlExecutorFactory.get()) {
        ddlExecutor.createTableIfNotExists(tdBuilder.build(), splitKeys);
    }
    HTable hTable = tableUtil.createHTable(hConf, hBaseTableId);
    hTable.setWriteBufferSize(Constants.Stream.HBASE_WRITE_BUFFER_SIZE);
    hTable.setAutoFlushTo(false);
    return new HBaseStreamFileConsumer(cConf, streamConfig, consumerConfig, tableUtil, hTable, reader, stateStore, beginConsumerState, extraFilter, createKeyDistributor(hTable.getTableDescriptor()));
}
Also used : TableId(co.cask.cdap.data2.util.TableId) HBaseDDLExecutor(co.cask.cdap.spi.hbase.HBaseDDLExecutor) RowKeyDistributorByHashPrefix(co.cask.cdap.hbase.wd.RowKeyDistributorByHashPrefix) AbstractRowKeyDistributor(co.cask.cdap.hbase.wd.AbstractRowKeyDistributor) ColumnFamilyDescriptorBuilder(co.cask.cdap.data2.util.hbase.ColumnFamilyDescriptorBuilder) TableDescriptorBuilder(co.cask.cdap.data2.util.hbase.TableDescriptorBuilder) NamespaceId(co.cask.cdap.proto.id.NamespaceId) HTable(org.apache.hadoop.hbase.client.HTable)

Example 2 with ConsumerConfig

use of co.cask.cdap.data2.queue.ConsumerConfig in project cdap by caskdata.

the class HBaseConsumerStateStore method getState.

/**
 * Returns the consumer state as stored in the state store for the given consumer.
 */
HBaseConsumerState getState(long groupId, int instanceId) {
    // Lookup the start row for the given instance, also search the barriers that bound the start row
    ConsumerState consumerState = getConsumerState(groupId, instanceId);
    QueueBarrier previousBarrier = consumerState.getPreviousBarrier();
    QueueBarrier nextBarrier = consumerState.getNextBarrier();
    if (previousBarrier == null && nextBarrier == null) {
        throw new IllegalStateException(String.format("Unable to find barrier information for consumer. Queue: %s, GroupId: %d, InstanceId:%d", queueName, groupId, instanceId));
    }
    // There are three possible cases:
    // 1. previousBarrier == null. It means in old compat mode. Since in old queue we didn't record the
    // consumer group config, we assume it's the same as the one recorded by
    // the nextBarrier (which was written when Flow start for the first time with new queue)
    // 2. nextBarrier == null. It means pasted the last barrier. The consumer scan is unbounded
    // 3. both not null. The scan is bounded by the nextBarrier and
    // the consumer group config is described by the previousBarrier.
    ConsumerGroupConfig groupConfig = previousBarrier != null ? previousBarrier.getGroupConfig() : nextBarrier.getGroupConfig();
    ConsumerConfig consumerConfig = new ConsumerConfig(groupConfig, instanceId);
    return new HBaseConsumerState(consumerConfig, consumerState.getConsumerStartRow(), previousBarrier == null ? null : previousBarrier.getStartRow(), nextBarrier == null ? null : nextBarrier.getStartRow());
}
Also used : ConsumerConfig(co.cask.cdap.data2.queue.ConsumerConfig) ConsumerGroupConfig(co.cask.cdap.data2.queue.ConsumerGroupConfig)

Example 3 with ConsumerConfig

use of co.cask.cdap.data2.queue.ConsumerConfig in project cdap by caskdata.

the class HBaseQueueClientFactory method createConsumer.

@Override
public QueueConsumer createConsumer(final QueueName queueName, final ConsumerConfig consumerConfig, int numGroups) throws IOException {
    final HBaseQueueAdmin admin = ensureTableExists(queueName);
    try {
        final long groupId = consumerConfig.getGroupId();
        // A callback for create a list of HBaseQueueConsumer
        // based on the current queue consumer state of the given group
        Callable<List<HBaseQueueConsumer>> consumerCreator = new Callable<List<HBaseQueueConsumer>>() {

            @Override
            public List<HBaseQueueConsumer> call() throws Exception {
                List<HBaseConsumerState> states;
                try (HBaseConsumerStateStore stateStore = admin.getConsumerStateStore(queueName)) {
                    TransactionExecutor txExecutor = Transactions.createTransactionExecutor(txExecutorFactory, stateStore);
                    // Find all consumer states for consumers that need to be created based on current state
                    states = txExecutor.execute(new Callable<List<HBaseConsumerState>>() {

                        @Override
                        public List<HBaseConsumerState> call() throws Exception {
                            List<HBaseConsumerState> consumerStates = Lists.newArrayList();
                            HBaseConsumerState state = stateStore.getState(groupId, consumerConfig.getInstanceId());
                            if (state.getPreviousBarrier() == null) {
                                // Old HBase consumer (Salted based, not sharded)
                                consumerStates.add(state);
                                return consumerStates;
                            }
                            // Find the smallest start barrier that has something to consume for this instance.
                            // It should always exists since we assume the queue is configured before this method is called
                            List<QueueBarrier> queueBarriers = stateStore.getAllBarriers(groupId);
                            if (queueBarriers.isEmpty()) {
                                throw new IllegalStateException(String.format("No consumer information available. Queue: %s, GroupId: %d, InstanceId: %d", queueName, groupId, consumerConfig.getInstanceId()));
                            }
                            QueueBarrier startBarrier = Iterables.find(Lists.reverse(queueBarriers), new Predicate<QueueBarrier>() {

                                @Override
                                public boolean apply(QueueBarrier barrier) {
                                    return barrier.getGroupConfig().getGroupSize() > consumerConfig.getInstanceId() && stateStore.isAllConsumed(consumerConfig, barrier.getStartRow());
                                }
                            }, queueBarriers.get(0));
                            int groupSize = startBarrier.getGroupConfig().getGroupSize();
                            for (int i = consumerConfig.getInstanceId(); i < groupSize; i += consumerConfig.getGroupSize()) {
                                consumerStates.add(stateStore.getState(groupId, i));
                            }
                            return consumerStates;
                        }
                    });
                }
                List<HBaseQueueConsumer> consumers = Lists.newArrayList();
                for (HBaseConsumerState state : states) {
                    QueueType queueType = (state.getPreviousBarrier() == null) ? QueueType.QUEUE : QueueType.SHARDED_QUEUE;
                    HTable hTable = createHTable(admin.getDataTableId(queueName, queueType));
                    int distributorBuckets = getDistributorBuckets(hTable.getTableDescriptor());
                    HBaseQueueStrategy strategy = (state.getPreviousBarrier() == null) ? new SaltedHBaseQueueStrategy(hBaseTableUtil, distributorBuckets) : new ShardedHBaseQueueStrategy(hBaseTableUtil, distributorBuckets);
                    consumers.add(queueUtil.getQueueConsumer(cConf, hTable, queueName, state, admin.getConsumerStateStore(queueName), strategy));
                }
                return consumers;
            }
        };
        return new SmartQueueConsumer(queueName, consumerConfig, consumerCreator);
    } catch (Exception e) {
        // If there is exception, nothing much can be done here besides propagating
        Throwables.propagateIfPossible(e);
        throw new IOException(e);
    }
}
Also used : HTable(org.apache.hadoop.hbase.client.HTable) Callable(java.util.concurrent.Callable) QueueType(co.cask.cdap.data2.transaction.queue.QueueConstants.QueueType) List(java.util.List) TransactionExecutor(org.apache.tephra.TransactionExecutor) IOException(java.io.IOException) IOException(java.io.IOException)

Example 4 with ConsumerConfig

use of co.cask.cdap.data2.queue.ConsumerConfig in project cdap by caskdata.

the class AbstractStreamFileConsumerFactory method create.

@Override
public final StreamConsumer create(StreamId streamId, String namespace, ConsumerConfig consumerConfig) throws IOException {
    StreamConfig streamConfig = StreamUtils.ensureExists(streamAdmin, streamId);
    TableId tableId = getTableId(streamId, namespace);
    StreamConsumerStateStore stateStore = stateStoreFactory.create(streamConfig);
    StreamConsumerState consumerState = stateStore.get(consumerConfig.getGroupId(), consumerConfig.getInstanceId());
    return create(tableId, streamConfig, consumerConfig, stateStore, consumerState, createReader(streamConfig, consumerState), new TTLReadFilter(streamConfig.getTTL()));
}
Also used : TableId(co.cask.cdap.data2.util.TableId) TTLReadFilter(co.cask.cdap.data.file.filter.TTLReadFilter)

Example 5 with ConsumerConfig

use of co.cask.cdap.data2.queue.ConsumerConfig in project cdap by caskdata.

the class HBaseQueueConsumer method updateStartRow.

@Override
protected void updateStartRow(byte[] startRow) {
    if (canConsume && !completed) {
        ConsumerConfig consumerConfig = getConfig();
        stateStore.updateState(consumerConfig.getGroupId(), consumerConfig.getInstanceId(), startRow);
    }
}
Also used : ConsumerConfig(co.cask.cdap.data2.queue.ConsumerConfig)

Aggregations

ConsumerConfig (co.cask.cdap.data2.queue.ConsumerConfig)33 QueueConsumer (co.cask.cdap.data2.queue.QueueConsumer)17 TransactionContext (org.apache.tephra.TransactionContext)14 Test (org.junit.Test)14 QueueName (co.cask.cdap.common.queue.QueueName)12 ConsumerGroupConfig (co.cask.cdap.data2.queue.ConsumerGroupConfig)12 QueueEntry (co.cask.cdap.data2.queue.QueueEntry)8 Transaction (org.apache.tephra.Transaction)8 StreamEvent (co.cask.cdap.api.flow.flowlet.StreamEvent)7 QueueProducer (co.cask.cdap.data2.queue.QueueProducer)7 QueueEntryRow (co.cask.cdap.data2.transaction.queue.QueueEntryRow)7 StreamId (co.cask.cdap.proto.id.StreamId)7 Cell (org.apache.hadoop.hbase.Cell)7 Filter (org.apache.hadoop.hbase.filter.Filter)7 FilterList (org.apache.hadoop.hbase.filter.FilterList)7 IOException (java.io.IOException)6 TransactionFailureException (org.apache.tephra.TransactionFailureException)6 TransactionExecutor (org.apache.tephra.TransactionExecutor)5 DequeueResult (co.cask.cdap.data2.queue.DequeueResult)4 DequeueStrategy (co.cask.cdap.data2.queue.DequeueStrategy)3