use of org.apache.flink.table.types.logical.RowType in project flink by apache.
the class BatchExecSortMergeJoin method translateToPlanInternal.
@Override
@SuppressWarnings("unchecked")
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
ExecEdge leftInputEdge = getInputEdges().get(0);
ExecEdge rightInputEdge = getInputEdges().get(1);
// get input types
RowType leftType = (RowType) leftInputEdge.getOutputType();
RowType rightType = (RowType) rightInputEdge.getOutputType();
LogicalType[] keyFieldTypes = IntStream.of(leftKeys).mapToObj(leftType::getTypeAt).toArray(LogicalType[]::new);
RowType keyType = RowType.of(keyFieldTypes);
GeneratedJoinCondition condFunc = JoinUtil.generateConditionFunction(config.getTableConfig(), nonEquiCondition, leftType, rightType);
long externalBufferMemory = config.get(ExecutionConfigOptions.TABLE_EXEC_RESOURCE_EXTERNAL_BUFFER_MEMORY).getBytes();
long sortMemory = config.get(ExecutionConfigOptions.TABLE_EXEC_RESOURCE_SORT_MEMORY).getBytes();
int externalBufferNum = 1;
if (joinType == FlinkJoinType.FULL) {
externalBufferNum = 2;
}
long managedMemory = externalBufferMemory * externalBufferNum + sortMemory * 2;
SortCodeGenerator leftSortGen = newSortGen(config, leftKeys, leftType);
SortCodeGenerator rightSortGen = newSortGen(config, rightKeys, rightType);
int[] keyPositions = IntStream.range(0, leftKeys.length).toArray();
SortMergeJoinOperator operator = new SortMergeJoinOperator(1.0 * externalBufferMemory / managedMemory, joinType, leftIsSmaller, condFunc, ProjectionCodeGenerator.generateProjection(new CodeGeneratorContext(config.getTableConfig()), "SMJProjection", leftType, keyType, leftKeys), ProjectionCodeGenerator.generateProjection(new CodeGeneratorContext(config.getTableConfig()), "SMJProjection", rightType, keyType, rightKeys), leftSortGen.generateNormalizedKeyComputer("LeftComputer"), leftSortGen.generateRecordComparator("LeftComparator"), rightSortGen.generateNormalizedKeyComputer("RightComputer"), rightSortGen.generateRecordComparator("RightComparator"), newSortGen(config, keyPositions, keyType).generateRecordComparator("KeyComparator"), filterNulls);
Transformation<RowData> leftInputTransform = (Transformation<RowData>) leftInputEdge.translateToPlan(planner);
Transformation<RowData> rightInputTransform = (Transformation<RowData>) rightInputEdge.translateToPlan(planner);
return ExecNodeUtil.createTwoInputTransformation(leftInputTransform, rightInputTransform, createTransformationName(config), createTransformationDescription(config), SimpleOperatorFactory.of(operator), InternalTypeInfo.of(getOutputType()), rightInputTransform.getParallelism(), managedMemory);
}
use of org.apache.flink.table.types.logical.RowType in project flink by apache.
the class BatchExecSort method translateToPlanInternal.
@SuppressWarnings("unchecked")
@Override
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
ExecEdge inputEdge = getInputEdges().get(0);
Transformation<RowData> inputTransform = (Transformation<RowData>) inputEdge.translateToPlan(planner);
RowType inputType = (RowType) inputEdge.getOutputType();
SortCodeGenerator codeGen = new SortCodeGenerator(config.getTableConfig(), inputType, sortSpec);
SortOperator operator = new SortOperator(codeGen.generateNormalizedKeyComputer("BatchExecSortComputer"), codeGen.generateRecordComparator("BatchExecSortComparator"));
long sortMemory = config.get(ExecutionConfigOptions.TABLE_EXEC_RESOURCE_SORT_MEMORY).getBytes();
return ExecNodeUtil.createOneInputTransformation(inputTransform, createTransformationName(config), createTransformationDescription(config), SimpleOperatorFactory.of(operator), InternalTypeInfo.of((RowType) getOutputType()), inputTransform.getParallelism(), sortMemory);
}
use of org.apache.flink.table.types.logical.RowType in project flink by apache.
the class BatchExecSortLimit method translateToPlanInternal.
@SuppressWarnings("unchecked")
@Override
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
if (limitEnd == Long.MAX_VALUE) {
throw new TableException("Not support limitEnd is max value now!");
}
ExecEdge inputEdge = getInputEdges().get(0);
Transformation<RowData> inputTransform = (Transformation<RowData>) inputEdge.translateToPlan(planner);
RowType inputType = (RowType) inputEdge.getOutputType();
// generate comparator
GeneratedRecordComparator genComparator = ComparatorCodeGenerator.gen(config.getTableConfig(), "SortLimitComparator", inputType, sortSpec);
// TODO If input is ordered, there is no need to use the heap.
SortLimitOperator operator = new SortLimitOperator(isGlobal, limitStart, limitEnd, genComparator);
return ExecNodeUtil.createOneInputTransformation(inputTransform, createTransformationName(config), createTransformationDescription(config), SimpleOperatorFactory.of(operator), InternalTypeInfo.of(inputType), inputTransform.getParallelism());
}
use of org.apache.flink.table.types.logical.RowType in project flink by apache.
the class FilterPushDownSpec method getDigests.
@Override
public String getDigests(SourceAbilityContext context) {
final List<String> expressionStrs = new ArrayList<>();
final RowType sourceRowType = context.getSourceRowType();
for (RexNode rexNode : predicates) {
expressionStrs.add(FlinkRexUtil.getExpressionString(rexNode, JavaScalaConversionUtil.toScala(sourceRowType.getFieldNames())));
}
return String.format("filter=[%s]", expressionStrs.stream().reduce((l, r) -> String.format("and(%s, %s)", l, r)).orElse(""));
}
use of org.apache.flink.table.types.logical.RowType in project flink by apache.
the class BatchExecHashJoin method translateToPlanInternal.
@Override
@SuppressWarnings("unchecked")
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
ExecEdge leftInputEdge = getInputEdges().get(0);
ExecEdge rightInputEdge = getInputEdges().get(1);
Transformation<RowData> leftInputTransform = (Transformation<RowData>) leftInputEdge.translateToPlan(planner);
Transformation<RowData> rightInputTransform = (Transformation<RowData>) rightInputEdge.translateToPlan(planner);
// get input types
RowType leftType = (RowType) leftInputEdge.getOutputType();
RowType rightType = (RowType) rightInputEdge.getOutputType();
JoinUtil.validateJoinSpec(joinSpec, leftType, rightType, false);
int[] leftKeys = joinSpec.getLeftKeys();
int[] rightKeys = joinSpec.getRightKeys();
LogicalType[] keyFieldTypes = IntStream.of(leftKeys).mapToObj(leftType::getTypeAt).toArray(LogicalType[]::new);
RowType keyType = RowType.of(keyFieldTypes);
GeneratedJoinCondition condFunc = JoinUtil.generateConditionFunction(config.getTableConfig(), joinSpec.getNonEquiCondition().orElse(null), leftType, rightType);
// projection for equals
GeneratedProjection leftProj = ProjectionCodeGenerator.generateProjection(new CodeGeneratorContext(config.getTableConfig()), "HashJoinLeftProjection", leftType, keyType, leftKeys);
GeneratedProjection rightProj = ProjectionCodeGenerator.generateProjection(new CodeGeneratorContext(config.getTableConfig()), "HashJoinRightProjection", rightType, keyType, rightKeys);
Transformation<RowData> buildTransform;
Transformation<RowData> probeTransform;
GeneratedProjection buildProj;
GeneratedProjection probeProj;
int[] buildKeys;
int[] probeKeys;
RowType buildType;
RowType probeType;
int buildRowSize;
long buildRowCount;
long probeRowCount;
boolean reverseJoin = !leftIsBuild;
if (leftIsBuild) {
buildTransform = leftInputTransform;
buildProj = leftProj;
buildType = leftType;
buildRowSize = estimatedLeftAvgRowSize;
buildRowCount = estimatedLeftRowCount;
buildKeys = leftKeys;
probeTransform = rightInputTransform;
probeProj = rightProj;
probeType = rightType;
probeRowCount = estimatedLeftRowCount;
probeKeys = rightKeys;
} else {
buildTransform = rightInputTransform;
buildProj = rightProj;
buildType = rightType;
buildRowSize = estimatedRightAvgRowSize;
buildRowCount = estimatedRightRowCount;
buildKeys = rightKeys;
probeTransform = leftInputTransform;
probeProj = leftProj;
probeType = leftType;
probeRowCount = estimatedLeftRowCount;
probeKeys = leftKeys;
}
// operator
StreamOperatorFactory<RowData> operator;
FlinkJoinType joinType = joinSpec.getJoinType();
HashJoinType hashJoinType = HashJoinType.of(leftIsBuild, joinType.isLeftOuter(), joinType.isRightOuter(), joinType == FlinkJoinType.SEMI, joinType == FlinkJoinType.ANTI);
if (LongHashJoinGenerator.support(hashJoinType, keyType, joinSpec.getFilterNulls())) {
operator = LongHashJoinGenerator.gen(config.getTableConfig(), hashJoinType, keyType, buildType, probeType, buildKeys, probeKeys, buildRowSize, buildRowCount, reverseJoin, condFunc);
} else {
operator = SimpleOperatorFactory.of(HashJoinOperator.newHashJoinOperator(hashJoinType, condFunc, reverseJoin, joinSpec.getFilterNulls(), buildProj, probeProj, tryDistinctBuildRow, buildRowSize, buildRowCount, probeRowCount, keyType));
}
long managedMemory = config.get(ExecutionConfigOptions.TABLE_EXEC_RESOURCE_HASH_JOIN_MEMORY).getBytes();
return ExecNodeUtil.createTwoInputTransformation(buildTransform, probeTransform, createTransformationName(config), createTransformationDescription(config), operator, InternalTypeInfo.of(getOutputType()), probeTransform.getParallelism(), managedMemory);
}
Aggregations