/**
* 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.physical.impl.sort;
import io.netty.buffer.DrillBuf;
import java.util.ArrayList;
import java.util.List;
import org.apache.drill.common.exceptions.DrillRuntimeException;
import org.apache.drill.exec.exception.SchemaChangeException;
import org.apache.drill.exec.memory.AllocationReservation;
import org.apache.drill.exec.memory.BufferAllocator;
import org.apache.drill.exec.ops.FragmentContext;
import org.apache.drill.exec.record.BatchSchema;
import org.apache.drill.exec.record.BatchSchema.SelectionVectorMode;
import org.apache.drill.exec.record.MaterializedField;
import org.apache.drill.exec.record.VectorAccessible;
import org.apache.drill.exec.record.VectorContainer;
import org.apache.drill.exec.record.VectorWrapper;
import org.apache.drill.exec.record.selection.SelectionVector2;
import org.apache.drill.exec.record.selection.SelectionVector4;
import org.apache.drill.exec.vector.ValueVector;
import com.google.common.collect.ArrayListMultimap;
import com.google.common.collect.Lists;
public class SortRecordBatchBuilder implements AutoCloseable {
static final org.slf4j.Logger logger = org.slf4j.LoggerFactory.getLogger(SortRecordBatchBuilder.class);
private final ArrayListMultimap<BatchSchema, RecordBatchData> batches = ArrayListMultimap.create();
private int recordCount;
private long runningBatches;
private SelectionVector4 sv4;
private BufferAllocator allocator;
final AllocationReservation reservation;
public SortRecordBatchBuilder(BufferAllocator a) {
this.allocator = a;
this.reservation = a.newReservation();
}
private long getSize(VectorAccessible batch) {
long bytes = 0;
for (VectorWrapper<?> v : batch) {
bytes += v.getValueVector().getBufferSize();
}
return bytes;
}
/**
* Add another record batch to the set of record batches. TODO: Refactor this and other {@link #add
* (RecordBatchData)} method into one method.
* @param batch
* @return True if the requested add completed successfully. Returns false in the case that this builder is full and cannot receive additional packages.
* @throws SchemaChangeException
*/
public boolean add(VectorAccessible batch) {
if (batch.getSchema().getSelectionVectorMode() == SelectionVectorMode.FOUR_BYTE) {
throw new UnsupportedOperationException("A sort cannot currently work against a sv4 batch.");
}
if (batch.getRecordCount() == 0 && batches.size() > 0) {
return true; // skip over empty record batches.
}
long batchBytes = getSize(batch);
if (batchBytes == 0 && batches.size() > 0) {
return true;
}
if (runningBatches >= Character.MAX_VALUE) {
return false; // allowed in batch.
}
if (!reservation.add(batch.getRecordCount() * 4)) {
return false; // sv allocation available.
}
RecordBatchData bd = new RecordBatchData(batch, allocator);
runningBatches++;
batches.put(batch.getSchema(), bd);
recordCount += bd.getRecordCount();
return true;
}
public void add(RecordBatchData rbd) {
long batchBytes = getSize(rbd.getContainer());
if (batchBytes == 0 && batches.size() > 0) {
return;
}
if(runningBatches >= Character.MAX_VALUE) {
final String errMsg = String.format("Tried to add more than %d number of batches.", (int) Character.MAX_VALUE);
logger.error(errMsg);
throw new DrillRuntimeException(errMsg);
}
if (!reservation.add(rbd.getRecordCount() * 4)) {
final String errMsg = String.format("Failed to pre-allocate memory for SV. " + "Existing recordCount*4 = %d, " +
"incoming batch recordCount*4 = %d", recordCount * 4, rbd.getRecordCount() * 4);
logger.error(errMsg);
throw new DrillRuntimeException(errMsg);
}
if (rbd.getRecordCount() == 0 && batches.size() > 0) {
rbd.getContainer().zeroVectors();
SelectionVector2 sv2 = rbd.getSv2();
if (sv2 != null) {
sv2.clear();
}
return;
}
runningBatches++;
batches.put(rbd.getContainer().getSchema(), rbd);
recordCount += rbd.getRecordCount();
}
public void canonicalize() {
for (RecordBatchData batch : batches.values()) {
batch.canonicalize();
}
}
public boolean isEmpty() {
return batches.isEmpty();
}
public void build(FragmentContext context, VectorContainer outputContainer) throws SchemaChangeException{
outputContainer.clear();
if (batches.keySet().size() > 1) {
throw new SchemaChangeException("Sort currently only supports a single schema.");
}
if (batches.size() > Character.MAX_VALUE) {
throw new SchemaChangeException("Sort cannot work on more than %d batches at a time.", (int) Character.MAX_VALUE);
}
if (batches.keys().size() < 1) {
assert false : "Invalid to have an empty set of batches with no schemas.";
}
final DrillBuf svBuffer = reservation.allocateBuffer();
if (svBuffer == null) {
throw new OutOfMemoryError("Failed to allocate direct memory for SV4 vector in SortRecordBatchBuilder.");
}
sv4 = new SelectionVector4(svBuffer, recordCount, Character.MAX_VALUE);
BatchSchema schema = batches.keySet().iterator().next();
List<RecordBatchData> data = batches.get(schema);
// now we're going to generate the sv4 pointers
switch (schema.getSelectionVectorMode()) {
case NONE: {
int index = 0;
int recordBatchId = 0;
for (RecordBatchData d : data) {
for (int i =0; i < d.getRecordCount(); i++, index++) {
sv4.set(index, recordBatchId, i);
}
recordBatchId++;
}
break;
}
case TWO_BYTE: {
int index = 0;
int recordBatchId = 0;
for (RecordBatchData d : data) {
for (int i =0; i < d.getRecordCount(); i++, index++) {
sv4.set(index, recordBatchId, (int) d.getSv2().getIndex(i));
}
// might as well drop the selection vector since we'll stop using it now.
d.getSv2().clear();
recordBatchId++;
}
break;
}
default:
throw new UnsupportedOperationException();
}
// next, we'll create lists of each of the vector types.
ArrayListMultimap<MaterializedField, ValueVector> vectors = ArrayListMultimap.create();
for (RecordBatchData rbd : batches.values()) {
for (ValueVector v : rbd.getVectors()) {
vectors.put(v.getField(), v);
}
}
for (MaterializedField f : schema) {
List<ValueVector> v = vectors.get(f);
outputContainer.addHyperList(v, false);
}
outputContainer.buildSchema(SelectionVectorMode.FOUR_BYTE);
}
public SelectionVector4 getSv4() {
return sv4;
}
public void clear() {
for (RecordBatchData d : batches.values()) {
d.container.clear();
}
if (sv4 != null) {
sv4.clear();
}
}
@Override
public void close() {
reservation.close();
}
public List<VectorContainer> getHeldRecordBatches() {
ArrayList<VectorContainer> containerList = Lists.newArrayList();
for (BatchSchema bs : batches.keySet()) {
for (RecordBatchData bd : batches.get(bs)) {
VectorContainer c = bd.getContainer();
c.setRecordCount(bd.getRecordCount());
containerList.add(c);
}
}
batches.clear();
return containerList;
}
/**
* For given record count how much memory does SortRecordBatchBuilder needs for its own purpose. This is used in
* ExternalSortBatch to make decisions about whether to spill or not.
*
* @param recordCount
* @return
*/
public static long memoryNeeded(int recordCount) {
// We need 4 bytes (SV4) for each record. Due to power-of-two allocations, the
// backing buffer might be twice this size.
return recordCount * 2 * 4;
}
}