/*
* 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.join;
import java.io.IOException;
import java.util.LinkedList;
import java.util.Map;
import com.google.common.collect.ImmutableMap;
import org.apache.calcite.rel.core.JoinRelType;
import org.apache.drill.common.exceptions.DrillRuntimeException;
import org.apache.drill.common.expression.ErrorCollector;
import org.apache.drill.common.expression.ErrorCollectorImpl;
import org.apache.drill.common.expression.LogicalExpression;
import org.apache.drill.common.types.TypeProtos;
import org.apache.drill.common.types.Types;
import org.apache.drill.exec.compile.sig.GeneratorMapping;
import org.apache.drill.exec.compile.sig.MappingSet;
import org.apache.drill.exec.exception.ClassTransformationException;
import org.apache.drill.exec.exception.OutOfMemoryException;
import org.apache.drill.exec.exception.SchemaChangeException;
import org.apache.drill.exec.expr.ClassGenerator;
import org.apache.drill.exec.expr.CodeGenerator;
import org.apache.drill.exec.expr.BatchReference;
import org.apache.drill.exec.expr.ExpressionTreeMaterializer;
import org.apache.drill.exec.ops.FragmentContext;
import org.apache.drill.exec.physical.config.NestedLoopJoinPOP;
import org.apache.drill.exec.physical.impl.filter.ReturnValueExpression;
import org.apache.drill.exec.physical.impl.sort.RecordBatchData;
import org.apache.drill.exec.record.AbstractRecordBatch;
import org.apache.drill.exec.record.BatchSchema;
import org.apache.drill.exec.record.ExpandableHyperContainer;
import org.apache.drill.exec.record.MaterializedField;
import org.apache.drill.exec.record.RecordBatch;
import org.apache.drill.exec.record.TypedFieldId;
import org.apache.drill.exec.record.VectorAccessible;
import org.apache.drill.exec.record.VectorWrapper;
import org.apache.drill.exec.vector.AllocationHelper;
import com.google.common.base.Preconditions;
import com.sun.codemodel.JExpr;
import com.sun.codemodel.JExpression;
import com.sun.codemodel.JVar;
import org.apache.drill.exec.vector.ValueVector;
import org.apache.drill.exec.vector.complex.AbstractContainerVector;
/*
* RecordBatch implementation for the nested loop join operator
*/
public class NestedLoopJoinBatch extends AbstractRecordBatch<NestedLoopJoinPOP> {
private static final org.slf4j.Logger logger = org.slf4j.LoggerFactory.getLogger(NestedLoopJoinBatch.class);
// Maximum number records in the outgoing batch
protected static final int MAX_BATCH_SIZE = 4096;
// Input indexes to correctly update the stats
protected static final int LEFT_INPUT = 0;
protected static final int RIGHT_INPUT = 1;
// Left input to the nested loop join operator
private final RecordBatch left;
// Schema on the left side
private BatchSchema leftSchema = null;
// state (IterOutcome) of the left input
private IterOutcome leftUpstream = IterOutcome.NONE;
// Right input to the nested loop join operator.
private final RecordBatch right;
// Schema on the right side
private BatchSchema rightSchema = null;
// state (IterOutcome) of the right input
private IterOutcome rightUpstream = IterOutcome.NONE;
// Runtime generated class implementing the NestedLoopJoin interface
private NestedLoopJoin nljWorker = null;
// Number of output records in the current outgoing batch
private int outputRecords = 0;
// We accumulate all the batches on the right side in a hyper container.
private ExpandableHyperContainer rightContainer = new ExpandableHyperContainer();
// Record count of the individual batches in the right hyper container
private LinkedList<Integer> rightCounts = new LinkedList<>();
// Generator mapping for the right side
private static final GeneratorMapping EMIT_RIGHT =
GeneratorMapping.create("doSetup"/* setup method */, "emitRight" /* eval method */, null /* reset */,
null /* cleanup */);
// Generator mapping for the right side : constant
private static final GeneratorMapping EMIT_RIGHT_CONSTANT = GeneratorMapping.create("doSetup"/* setup method */,
"doSetup" /* eval method */,
null /* reset */, null /* cleanup */);
// Generator mapping for the left side : scalar
private static final GeneratorMapping EMIT_LEFT =
GeneratorMapping.create("doSetup" /* setup method */, "emitLeft" /* eval method */, null /* reset */,
null /* cleanup */);
// Generator mapping for the left side : constant
private static final GeneratorMapping EMIT_LEFT_CONSTANT = GeneratorMapping.create("doSetup" /* setup method */,
"doSetup" /* eval method */,
null /* reset */, null /* cleanup */);
// Mapping set for the right side
private static final MappingSet emitRightMapping =
new MappingSet("rightCompositeIndex" /* read index */, "outIndex" /* write index */, "rightContainer" /* read container */,
"outgoing" /* write container */, EMIT_RIGHT_CONSTANT, EMIT_RIGHT);
// Mapping set for the left side
private static final MappingSet emitLeftMapping = new MappingSet("leftIndex" /* read index */, "outIndex" /* write index */,
"leftBatch" /* read container */,
"outgoing" /* write container */,
EMIT_LEFT_CONSTANT, EMIT_LEFT);
protected NestedLoopJoinBatch(NestedLoopJoinPOP popConfig, FragmentContext context, RecordBatch left, RecordBatch right) throws OutOfMemoryException {
super(popConfig, context);
Preconditions.checkNotNull(left);
Preconditions.checkNotNull(right);
this.left = left;
this.right = right;
}
/**
* Method drains the right side input of the NLJ and accumulates the data
* in a hyper container. Once we have all the data from the right side we
* process the left side one batch at a time and produce the output batch
* @return IterOutcome state of the nested loop join batch
*/
@Override
public IterOutcome innerNext() {
// Accumulate batches on the right in a hyper container
if (state == BatchState.FIRST) {
// exit if we have an empty left batch
if (leftUpstream == IterOutcome.NONE) {
// inform upstream that we don't need anymore data and make sure we clean up any batches already in queue
killAndDrainRight();
return IterOutcome.NONE;
}
boolean drainRight = true;
while (drainRight) {
rightUpstream = next(RIGHT_INPUT, right);
switch (rightUpstream) {
case OK_NEW_SCHEMA:
if (!right.getSchema().equals(rightSchema)) {
throw new DrillRuntimeException("Nested loop join does not handle schema change. Schema change" +
" found on the right side of NLJ.");
}
// fall through
case OK:
addBatchToHyperContainer(right);
break;
case OUT_OF_MEMORY:
return IterOutcome.OUT_OF_MEMORY;
case NONE:
case STOP:
//TODO we got a STOP, shouldn't we stop immediately ?
case NOT_YET:
drainRight = false;
break;
}
}
nljWorker.setupNestedLoopJoin(context, left, rightContainer, rightCounts, this);
state = BatchState.NOT_FIRST;
}
// allocate space for the outgoing batch
allocateVectors();
// invoke the runtime generated method to emit records in the output batch
outputRecords = nljWorker.outputRecords(popConfig.getJoinType());
// Set the record count
for (final VectorWrapper<?> vw : container) {
vw.getValueVector().getMutator().setValueCount(outputRecords);
}
// Set the record count in the container
container.setRecordCount(outputRecords);
container.buildSchema(BatchSchema.SelectionVectorMode.NONE);
logger.debug("Number of records emitted: " + outputRecords);
return (outputRecords > 0) ? IterOutcome.OK : IterOutcome.NONE;
}
private void killAndDrainRight() {
if (!hasMore(rightUpstream)) {
return;
}
right.kill(true);
while (hasMore(rightUpstream)) {
for (final VectorWrapper<?> wrapper : right) {
wrapper.getValueVector().clear();
}
rightUpstream = next(HashJoinHelper.RIGHT_INPUT, right);
}
}
private boolean hasMore(IterOutcome outcome) {
return outcome == IterOutcome.OK || outcome == IterOutcome.OK_NEW_SCHEMA;
}
/**
* Method generates the runtime code needed for NLJ. Other than the setup method to set the input and output value
* vector references we implement three more methods
* 1. doEval() -> Evaluates if record from left side matches record from the right side
* 2. emitLeft() -> Project record from the left side
* 3. emitRight() -> Project record from the right side (which is a hyper container)
* @return the runtime generated class that implements the NestedLoopJoin interface
*/
private NestedLoopJoin setupWorker() throws IOException, ClassTransformationException, SchemaChangeException {
final CodeGenerator<NestedLoopJoin> nLJCodeGenerator = CodeGenerator.get(
NestedLoopJoin.TEMPLATE_DEFINITION, context.getFunctionRegistry(), context.getOptions());
nLJCodeGenerator.plainJavaCapable(true);
// Uncomment out this line to debug the generated code.
// nLJCodeGenerator.saveCodeForDebugging(true);
final ClassGenerator<NestedLoopJoin> nLJClassGenerator = nLJCodeGenerator.getRoot();
// generate doEval
final ErrorCollector collector = new ErrorCollectorImpl();
/*
Logical expression may contain fields from left and right batches. During code generation (materialization)
we need to indicate from which input field should be taken.
Non-equality joins can belong to one of below categories. For example:
1. Join on non-equality join predicates:
select * from t1 inner join t2 on (t1.c1 between t2.c1 AND t2.c2) AND (...)
2. Join with an OR predicate:
select * from t1 inner join t2 on on t1.c1 = t2.c1 OR t1.c2 = t2.c2
*/
Map<VectorAccessible, BatchReference> batches = ImmutableMap
.<VectorAccessible, BatchReference>builder()
.put(left, new BatchReference("leftBatch", "leftIndex"))
.put(rightContainer, new BatchReference("rightContainer", "rightBatchIndex", "rightRecordIndexWithinBatch"))
.build();
LogicalExpression materialize = ExpressionTreeMaterializer.materialize(
popConfig.getCondition(),
batches,
collector,
context.getFunctionRegistry(),
false,
false);
if (collector.hasErrors()) {
throw new SchemaChangeException(String.format("Failure while trying to materialize join condition. Errors:\n %s.",
collector.toErrorString()));
}
nLJClassGenerator.addExpr(new ReturnValueExpression(materialize), ClassGenerator.BlkCreateMode.FALSE);
// generate emitLeft
nLJClassGenerator.setMappingSet(emitLeftMapping);
JExpression outIndex = JExpr.direct("outIndex");
JExpression leftIndex = JExpr.direct("leftIndex");
int fieldId = 0;
int outputFieldId = 0;
// Set the input and output value vector references corresponding to the left batch
for (MaterializedField field : leftSchema) {
final TypeProtos.MajorType fieldType = field.getType();
// Add the vector to the output container
container.addOrGet(field);
JVar inVV = nLJClassGenerator.declareVectorValueSetupAndMember("leftBatch",
new TypedFieldId(fieldType, false, fieldId));
JVar outVV = nLJClassGenerator.declareVectorValueSetupAndMember("outgoing",
new TypedFieldId(fieldType, false, outputFieldId));
nLJClassGenerator.getEvalBlock().add(outVV.invoke("copyFromSafe").arg(leftIndex).arg(outIndex).arg(inVV));
nLJClassGenerator.rotateBlock();
fieldId++;
outputFieldId++;
}
// generate emitRight
fieldId = 0;
nLJClassGenerator.setMappingSet(emitRightMapping);
JExpression batchIndex = JExpr.direct("batchIndex");
JExpression recordIndexWithinBatch = JExpr.direct("recordIndexWithinBatch");
// Set the input and output value vector references corresponding to the right batch
for (MaterializedField field : rightSchema) {
final TypeProtos.MajorType inputType = field.getType();
TypeProtos.MajorType outputType;
// if join type is LEFT, make sure right batch output fields data mode is optional
if (popConfig.getJoinType() == JoinRelType.LEFT && inputType.getMode() == TypeProtos.DataMode.REQUIRED) {
outputType = Types.overrideMode(inputType, TypeProtos.DataMode.OPTIONAL);
} else {
outputType = inputType;
}
MaterializedField newField = MaterializedField.create(field.getPath(), outputType);
container.addOrGet(newField);
JVar inVV = nLJClassGenerator.declareVectorValueSetupAndMember("rightContainer",
new TypedFieldId(inputType, true, fieldId));
JVar outVV = nLJClassGenerator.declareVectorValueSetupAndMember("outgoing",
new TypedFieldId(outputType, false, outputFieldId));
nLJClassGenerator.getEvalBlock().add(outVV.invoke("copyFromSafe")
.arg(recordIndexWithinBatch)
.arg(outIndex)
.arg(inVV.component(batchIndex)));
nLJClassGenerator.rotateBlock();
fieldId++;
outputFieldId++;
}
return context.getImplementationClass(nLJCodeGenerator);
}
/**
* Simple method to allocate space for all the vectors in the container.
*/
private void allocateVectors() {
for (final VectorWrapper<?> vw : container) {
AllocationHelper.allocateNew(vw.getValueVector(), MAX_BATCH_SIZE);
}
}
/**
* Builds the output container's schema. Goes over the left and the right
* batch and adds the corresponding vectors to the output container.
* @throws SchemaChangeException if batch schema was changed during execution
*/
@Override
protected void buildSchema() throws SchemaChangeException {
try {
leftUpstream = next(LEFT_INPUT, left);
rightUpstream = next(RIGHT_INPUT, right);
if (leftUpstream == IterOutcome.STOP || rightUpstream == IterOutcome.STOP) {
state = BatchState.STOP;
return;
}
if (leftUpstream == IterOutcome.OUT_OF_MEMORY || rightUpstream == IterOutcome.OUT_OF_MEMORY) {
state = BatchState.OUT_OF_MEMORY;
return;
}
if (leftUpstream != IterOutcome.NONE) {
leftSchema = left.getSchema();
for (final VectorWrapper<?> vw : left) {
container.addOrGet(vw.getField());
}
}
if (rightUpstream != IterOutcome.NONE) {
// make right input schema optional if we have LEFT join
for (final VectorWrapper<?> vectorWrapper : right) {
TypeProtos.MajorType inputType = vectorWrapper.getField().getType();
TypeProtos.MajorType outputType;
if (popConfig.getJoinType() == JoinRelType.LEFT && inputType.getMode() == TypeProtos.DataMode.REQUIRED) {
outputType = Types.overrideMode(inputType, TypeProtos.DataMode.OPTIONAL);
} else {
outputType = inputType;
}
MaterializedField newField = MaterializedField.create(vectorWrapper.getField().getPath(), outputType);
ValueVector valueVector = container.addOrGet(newField);
if (valueVector instanceof AbstractContainerVector) {
vectorWrapper.getValueVector().makeTransferPair(valueVector);
valueVector.clear();
}
}
rightSchema = right.getSchema();
addBatchToHyperContainer(right);
}
allocateVectors();
nljWorker = setupWorker();
// if left batch is empty, fetch next
if (leftUpstream != IterOutcome.NONE && left.getRecordCount() == 0) {
leftUpstream = next(LEFT_INPUT, left);
}
container.setRecordCount(0);
container.buildSchema(BatchSchema.SelectionVectorMode.NONE);
} catch (ClassTransformationException | IOException e) {
throw new SchemaChangeException(e);
}
}
private void addBatchToHyperContainer(RecordBatch inputBatch) {
final RecordBatchData batchCopy = new RecordBatchData(inputBatch, oContext.getAllocator());
boolean success = false;
try {
rightCounts.addLast(inputBatch.getRecordCount());
rightContainer.addBatch(batchCopy.getContainer());
success = true;
} finally {
if (!success) {
batchCopy.clear();
}
}
}
@Override
public void close() {
rightContainer.clear();
rightCounts.clear();
super.close();
}
@Override
protected void killIncoming(boolean sendUpstream) {
this.left.kill(sendUpstream);
this.right.kill(sendUpstream);
}
@Override
public int getRecordCount() {
return outputRecords;
}
}