blob: 2d1b4f2f0c2763978dc756ed10e81f915014155b [file] [log] [blame]
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.drill.exec.work.batch;
import java.io.IOException;
import java.util.Arrays;
import java.util.Map;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.locks.ReadWriteLock;
import java.util.concurrent.locks.ReentrantReadWriteLock;
import org.apache.drill.common.AutoCloseables;
import org.apache.drill.common.concurrent.AutoCloseableLock;
import org.apache.drill.exec.exception.FragmentSetupException;
import org.apache.drill.exec.memory.BufferAllocator;
import org.apache.drill.exec.ops.FragmentContext;
import org.apache.drill.exec.proto.BitControl.Collector;
import org.apache.drill.exec.proto.BitControl.PlanFragment;
import org.apache.drill.exec.record.RawFragmentBatch;
import org.apache.drill.exec.rpc.data.IncomingDataBatch;
import com.google.common.collect.ImmutableMap;
import com.google.common.collect.Maps;
/**
* Determines when a particular fragment has enough data for each of its receiving exchanges to commence execution. Also monitors whether we've collected all incoming data.
*/
public class IncomingBuffers implements AutoCloseable {
static final org.slf4j.Logger logger = org.slf4j.LoggerFactory.getLogger(IncomingBuffers.class);
private volatile boolean closed = false;
private final AtomicInteger streamsRemaining = new AtomicInteger(0);
private final AtomicInteger remainingRequired;
private final Map<Integer, DataCollector> collectorMap;
private final FragmentContext context;
/**
* Lock used to manage close and data acceptance. We should only create a local reference to incoming data in the case
* that the incoming buffers are !closed. As such, we need to make sure that we aren't in the process of closing the
* incoming buffers when data is arriving. The read lock can be shared by many incoming batches but the write lock
* must be exclusive to the close method.
*/
private final ReadWriteLock lock = new ReentrantReadWriteLock();
private final AutoCloseableLock sharedIncomingBatchLock = new AutoCloseableLock(lock.readLock());
private final AutoCloseableLock exclusiveCloseLock = new AutoCloseableLock(lock.writeLock());
public IncomingBuffers(PlanFragment fragment, FragmentContext context) {
this.context = context;
Map<Integer, DataCollector> collectors = Maps.newHashMap();
remainingRequired = new AtomicInteger(fragment.getCollectorCount());
for(int i =0; i < fragment.getCollectorCount(); i++){
Collector collector = fragment.getCollector(i);
DataCollector newCollector = collector.getSupportsOutOfOrder() ?
new MergingCollector(remainingRequired, collector, context) :
new PartitionedCollector(remainingRequired, collector, context);
collectors.put(collector.getOppositeMajorFragmentId(), newCollector);
}
logger.debug("Came up with a list of {} required fragments. Fragments {}", remainingRequired.get(), collectors);
collectorMap = ImmutableMap.copyOf(collectors);
// Determine the total number of incoming streams that will need to be completed before we are finished.
int totalStreams = 0;
for (DataCollector bc : collectorMap.values()) {
totalStreams += bc.getTotalIncomingFragments();
}
assert totalStreams >= remainingRequired.get() : String.format("Total Streams %d should be more than the minimum number of streams to commence (%d). It isn't.", totalStreams, remainingRequired.get());
streamsRemaining.set(totalStreams);
}
public boolean batchArrived(final IncomingDataBatch incomingBatch) throws FragmentSetupException, IOException {
// we want to make sure that we only generate local record batch reference in the case that we're not closed.
// Otherwise we would leak memory.
try (@SuppressWarnings("unused") AutoCloseables.Closeable lock = sharedIncomingBatchLock.open()) {
if (closed) {
return false;
}
if (incomingBatch.getHeader().getIsLastBatch()) {
streamsRemaining.decrementAndGet();
}
final int sendMajorFragmentId = incomingBatch.getHeader().getSendingMajorFragmentId();
DataCollector collector = collectorMap.get(sendMajorFragmentId);
if (collector == null) {
throw new FragmentSetupException(String.format(
"We received a major fragment id that we were not expecting. The id was %d. %s", sendMajorFragmentId,
Arrays.toString(collectorMap.values().toArray())));
}
// Use the Data Collector's buffer allocator if set, otherwise the fragment's one
BufferAllocator ownerAllocator = collector.getAllocator();
synchronized (collector) {
final RawFragmentBatch newRawFragmentBatch = incomingBatch.newRawFragmentBatch(ownerAllocator);
boolean decrementedToZero = collector
.batchArrived(incomingBatch.getHeader().getSendingMinorFragmentId(), newRawFragmentBatch);
newRawFragmentBatch.release();
// we should only return true if remaining required has been decremented and is currently equal to zero.
return decrementedToZero;
}
}
}
public int getRemainingRequired() {
int rem = remainingRequired.get();
if (rem < 0) {
return 0;
}
return rem;
}
public DataCollector getCollector(int senderMajorFragmentId) {
return collectorMap.get(senderMajorFragmentId);
}
public boolean isDone() {
return streamsRemaining.get() < 1;
}
@Override
public void close() throws Exception {
try (@SuppressWarnings("unused") AutoCloseables.Closeable lock = exclusiveCloseLock.open()) {
closed = true;
AutoCloseables.close(collectorMap.values());
}
}
}