Examples with FixedWidthVector org.apache.drill.exec.vector.FixedWidthVector used on opensource projects

Example 1 with FixedWidthVector

use of org.apache.drill.exec.vector.FixedWidthVector in project drill by apache.

the class UnionAllRecordBatch method doWork.

@SuppressWarnings("resource")
private IterOutcome doWork() throws ClassTransformationException, IOException, SchemaChangeException {
    if (allocationVectors != null) {
        for (ValueVector v : allocationVectors) {
            v.clear();
        }
    }
    allocationVectors = Lists.newArrayList();
    transfers.clear();
    // If both sides of Union-All are empty
    if (unionAllInput.isBothSideEmpty()) {
        for (int i = 0; i < outputFields.size(); ++i) {
            final String colName = outputFields.get(i).getPath();
            final MajorType majorType = MajorType.newBuilder().setMinorType(MinorType.INT).setMode(DataMode.OPTIONAL).build();
            MaterializedField outputField = MaterializedField.create(colName, majorType);
            ValueVector vv = container.addOrGet(outputField, callBack);
            allocationVectors.add(vv);
        }
        container.buildSchema(BatchSchema.SelectionVectorMode.NONE);
        return IterOutcome.OK_NEW_SCHEMA;
    }
    final ClassGenerator<UnionAller> cg = CodeGenerator.getRoot(UnionAller.TEMPLATE_DEFINITION, context.getFunctionRegistry(), context.getOptions());
    cg.getCodeGenerator().plainJavaCapable(true);
    // Uncomment out this line to debug the generated code.
    //    cg.getCodeGenerator().saveCodeForDebugging(true);
    int index = 0;
    for (VectorWrapper<?> vw : current) {
        ValueVector vvIn = vw.getValueVector();
        // get the original input column names
        SchemaPath inputPath = SchemaPath.getSimplePath(vvIn.getField().getPath());
        // get the renamed column names
        SchemaPath outputPath = SchemaPath.getSimplePath(outputFields.get(index).getPath());
        final ErrorCollector collector = new ErrorCollectorImpl();
        // cast data types (Minortype or DataMode)
        if (hasSameTypeAndMode(outputFields.get(index), vw.getValueVector().getField())) {
            // Transfer column
            MajorType outputFieldType = outputFields.get(index).getType();
            MaterializedField outputField = MaterializedField.create(outputPath.getAsUnescapedPath(), outputFieldType);
            /*
          todo: Fix if condition when DRILL-4824 is merged
          If condition should be changed to:
          `if (outputFields.get(index).getPath().equals(inputPath.getAsUnescapedPath())) {`
          DRILL-5419 has changed condition to correct one but this caused regression (DRILL-5521).
          Root cause is missing indication of child column in map types when it is null.
          DRILL-4824 is re-working json reader implementation, including map types and will fix this problem.
          Reverting condition to previous one to avoid regression till DRILL-4824 is merged.
          Unit test - TestJsonReader.testKvgenWithUnionAll().
         */
            if (outputFields.get(index).getPath().equals(inputPath)) {
                ValueVector vvOut = container.addOrGet(outputField);
                TransferPair tp = vvIn.makeTransferPair(vvOut);
                transfers.add(tp);
            // Copy data in order to rename the column
            } else {
                final LogicalExpression expr = ExpressionTreeMaterializer.materialize(inputPath, current, collector, context.getFunctionRegistry());
                if (collector.hasErrors()) {
                    throw new SchemaChangeException(String.format("Failure while trying to materialize incoming schema.  Errors:\n %s.", collector.toErrorString()));
                }
                ValueVector vv = container.addOrGet(outputField, callBack);
                allocationVectors.add(vv);
                TypedFieldId fid = container.getValueVectorId(SchemaPath.getSimplePath(outputField.getPath()));
                ValueVectorWriteExpression write = new ValueVectorWriteExpression(fid, expr, true);
                cg.addExpr(write);
            }
        // Cast is necessary
        } else {
            LogicalExpression expr = ExpressionTreeMaterializer.materialize(inputPath, current, collector, context.getFunctionRegistry());
            if (collector.hasErrors()) {
                throw new SchemaChangeException(String.format("Failure while trying to materialize incoming schema.  Errors:\n %s.", collector.toErrorString()));
            }
            // cast to the one with the least restriction
            if (vvIn.getField().getType().getMode() == DataMode.REQUIRED && outputFields.get(index).getType().getMode() != DataMode.REQUIRED) {
                expr = ExpressionTreeMaterializer.convertToNullableType(expr, vvIn.getField().getType().getMinorType(), context.getFunctionRegistry(), collector);
                if (collector.hasErrors()) {
                    throw new SchemaChangeException(String.format("Failure while trying to materialize incoming schema.  Errors:\n %s.", collector.toErrorString()));
                }
            }
            // Insert a cast before the Union operation
            if (vvIn.getField().getType().getMinorType() != outputFields.get(index).getType().getMinorType()) {
                expr = ExpressionTreeMaterializer.addCastExpression(expr, outputFields.get(index).getType(), context.getFunctionRegistry(), collector);
                if (collector.hasErrors()) {
                    throw new SchemaChangeException(String.format("Failure while trying to materialize incoming schema.  Errors:\n %s.", collector.toErrorString()));
                }
            }
            final MaterializedField outputField = MaterializedField.create(outputPath.getAsUnescapedPath(), expr.getMajorType());
            ValueVector vector = container.addOrGet(outputField, callBack);
            allocationVectors.add(vector);
            TypedFieldId fid = container.getValueVectorId(SchemaPath.getSimplePath(outputField.getPath()));
            boolean useSetSafe = !(vector instanceof FixedWidthVector);
            ValueVectorWriteExpression write = new ValueVectorWriteExpression(fid, expr, useSetSafe);
            cg.addExpr(write);
        }
        ++index;
    }
    unionall = context.getImplementationClass(cg.getCodeGenerator());
    unionall.setup(context, current, this, transfers);
    if (!schemaAvailable) {
        container.buildSchema(BatchSchema.SelectionVectorMode.NONE);
        schemaAvailable = true;
    }
    if (!doAlloc()) {
        return IterOutcome.OUT_OF_MEMORY;
    }
    recordCount = unionall.unionRecords(0, current.getRecordCount(), 0);
    setValueCount(recordCount);
    return IterOutcome.OK;
}

Example 2 with FixedWidthVector

use of org.apache.drill.exec.vector.FixedWidthVector in project drill by apache.

the class ProjectRecordBatch method setupNewSchema.

@Override
protected boolean setupNewSchema() throws SchemaChangeException {
    if (allocationVectors != null) {
        for (final ValueVector v : allocationVectors) {
            v.clear();
        }
    }
    this.allocationVectors = Lists.newArrayList();
    if (complexWriters != null) {
        container.clear();
    } else {
        container.zeroVectors();
    }
    final List<NamedExpression> exprs = getExpressionList();
    final ErrorCollector collector = new ErrorCollectorImpl();
    final List<TransferPair> transfers = Lists.newArrayList();
    final ClassGenerator<Projector> cg = CodeGenerator.getRoot(Projector.TEMPLATE_DEFINITION, context.getFunctionRegistry(), context.getOptions());
    cg.getCodeGenerator().plainJavaCapable(true);
    // Uncomment out this line to debug the generated code.
    //    cg.getCodeGenerator().saveCodeForDebugging(true);
    final IntHashSet transferFieldIds = new IntHashSet();
    final boolean isAnyWildcard = isAnyWildcard(exprs);
    final ClassifierResult result = new ClassifierResult();
    final boolean classify = isClassificationNeeded(exprs);
    for (int i = 0; i < exprs.size(); i++) {
        final NamedExpression namedExpression = exprs.get(i);
        result.clear();
        if (classify && namedExpression.getExpr() instanceof SchemaPath) {
            classifyExpr(namedExpression, incoming, result);
            if (result.isStar) {
                // The value indicates which wildcard we are processing now
                final Integer value = result.prefixMap.get(result.prefix);
                if (value != null && value.intValue() == 1) {
                    int k = 0;
                    for (final VectorWrapper<?> wrapper : incoming) {
                        final ValueVector vvIn = wrapper.getValueVector();
                        if (k > result.outputNames.size() - 1) {
                            assert false;
                        }
                        // get the renamed column names
                        final String name = result.outputNames.get(k++);
                        if (name == EMPTY_STRING) {
                            continue;
                        }
                        if (isImplicitFileColumn(vvIn)) {
                            continue;
                        }
                        final FieldReference ref = new FieldReference(name);
                        final ValueVector vvOut = container.addOrGet(MaterializedField.create(ref.getAsNamePart().getName(), vvIn.getField().getType()), callBack);
                        final TransferPair tp = vvIn.makeTransferPair(vvOut);
                        transfers.add(tp);
                    }
                } else if (value != null && value.intValue() > 1) {
                    // subsequent wildcards should do a copy of incoming valuevectors
                    int k = 0;
                    for (final VectorWrapper<?> wrapper : incoming) {
                        final ValueVector vvIn = wrapper.getValueVector();
                        final SchemaPath originalPath = SchemaPath.getSimplePath(vvIn.getField().getPath());
                        if (k > result.outputNames.size() - 1) {
                            assert false;
                        }
                        // get the renamed column names
                        final String name = result.outputNames.get(k++);
                        if (name == EMPTY_STRING) {
                            continue;
                        }
                        if (isImplicitFileColumn(vvIn)) {
                            continue;
                        }
                        final LogicalExpression expr = ExpressionTreeMaterializer.materialize(originalPath, incoming, collector, context.getFunctionRegistry());
                        if (collector.hasErrors()) {
                            throw new SchemaChangeException(String.format("Failure while trying to materialize incoming schema.  Errors:\n %s.", collector.toErrorString()));
                        }
                        final MaterializedField outputField = MaterializedField.create(name, expr.getMajorType());
                        final ValueVector vv = container.addOrGet(outputField, callBack);
                        allocationVectors.add(vv);
                        final TypedFieldId fid = container.getValueVectorId(SchemaPath.getSimplePath(outputField.getPath()));
                        final ValueVectorWriteExpression write = new ValueVectorWriteExpression(fid, expr, true);
                        final HoldingContainer hc = cg.addExpr(write, ClassGenerator.BlkCreateMode.TRUE_IF_BOUND);
                    }
                }
                continue;
            }
        } else {
            // For the columns which do not needed to be classified,
            // it is still necessary to ensure the output column name is unique
            result.outputNames = Lists.newArrayList();
            final String outputName = getRef(namedExpression).getRootSegment().getPath();
            addToResultMaps(outputName, result, true);
        }
        String outputName = getRef(namedExpression).getRootSegment().getPath();
        if (result != null && result.outputNames != null && result.outputNames.size() > 0) {
            boolean isMatched = false;
            for (int j = 0; j < result.outputNames.size(); j++) {
                if (!result.outputNames.get(j).equals(EMPTY_STRING)) {
                    outputName = result.outputNames.get(j);
                    isMatched = true;
                    break;
                }
            }
            if (!isMatched) {
                continue;
            }
        }
        final LogicalExpression expr = ExpressionTreeMaterializer.materialize(namedExpression.getExpr(), incoming, collector, context.getFunctionRegistry(), true, unionTypeEnabled);
        final MaterializedField outputField = MaterializedField.create(outputName, expr.getMajorType());
        if (collector.hasErrors()) {
            throw new SchemaChangeException(String.format("Failure while trying to materialize incoming schema.  Errors:\n %s.", collector.toErrorString()));
        }
        // add value vector to transfer if direct reference and this is allowed, otherwise, add to evaluation stack.
        if (expr instanceof ValueVectorReadExpression && incoming.getSchema().getSelectionVectorMode() == SelectionVectorMode.NONE && !((ValueVectorReadExpression) expr).hasReadPath() && !isAnyWildcard && !transferFieldIds.contains(((ValueVectorReadExpression) expr).getFieldId().getFieldIds()[0])) {
            final ValueVectorReadExpression vectorRead = (ValueVectorReadExpression) expr;
            final TypedFieldId id = vectorRead.getFieldId();
            final ValueVector vvIn = incoming.getValueAccessorById(id.getIntermediateClass(), id.getFieldIds()).getValueVector();
            Preconditions.checkNotNull(incoming);
            final FieldReference ref = getRef(namedExpression);
            final ValueVector vvOut = container.addOrGet(MaterializedField.create(ref.getAsUnescapedPath(), vectorRead.getMajorType()), callBack);
            final TransferPair tp = vvIn.makeTransferPair(vvOut);
            transfers.add(tp);
            transferFieldIds.add(vectorRead.getFieldId().getFieldIds()[0]);
        } else if (expr instanceof DrillFuncHolderExpr && ((DrillFuncHolderExpr) expr).getHolder().isComplexWriterFuncHolder()) {
            // Lazy initialization of the list of complex writers, if not done yet.
            if (complexWriters == null) {
                complexWriters = Lists.newArrayList();
            } else {
                complexWriters.clear();
            }
            // The reference name will be passed to ComplexWriter, used as the name of the output vector from the writer.
            ((DrillFuncHolderExpr) expr).getFieldReference(namedExpression.getRef());
            cg.addExpr(expr, ClassGenerator.BlkCreateMode.TRUE_IF_BOUND);
            if (complexFieldReferencesList == null) {
                complexFieldReferencesList = Lists.newArrayList();
            }
            // save the field reference for later for getting schema when input is empty
            complexFieldReferencesList.add(namedExpression.getRef());
        } else {
            // need to do evaluation.
            final ValueVector vector = container.addOrGet(outputField, callBack);
            allocationVectors.add(vector);
            final TypedFieldId fid = container.getValueVectorId(SchemaPath.getSimplePath(outputField.getPath()));
            final boolean useSetSafe = !(vector instanceof FixedWidthVector);
            final ValueVectorWriteExpression write = new ValueVectorWriteExpression(fid, expr, useSetSafe);
            final HoldingContainer hc = cg.addExpr(write, ClassGenerator.BlkCreateMode.TRUE_IF_BOUND);
            // We cannot do multiple transfers from the same vector. However we still need to instantiate the output vector.
            if (expr instanceof ValueVectorReadExpression) {
                final ValueVectorReadExpression vectorRead = (ValueVectorReadExpression) expr;
                if (!vectorRead.hasReadPath()) {
                    final TypedFieldId id = vectorRead.getFieldId();
                    final ValueVector vvIn = incoming.getValueAccessorById(id.getIntermediateClass(), id.getFieldIds()).getValueVector();
                    vvIn.makeTransferPair(vector);
                }
            }
            logger.debug("Added eval for project expression.");
        }
    }
    try {
        CodeGenerator<Projector> codeGen = cg.getCodeGenerator();
        codeGen.plainJavaCapable(true);
        // Uncomment out this line to debug the generated code.
        //      codeGen.saveCodeForDebugging(true);
        this.projector = context.getImplementationClass(codeGen);
        projector.setup(context, incoming, this, transfers);
    } catch (ClassTransformationException | IOException e) {
        throw new SchemaChangeException("Failure while attempting to load generated class", e);
    }
    if (container.isSchemaChanged()) {
        container.buildSchema(SelectionVectorMode.NONE);
        return true;
    } else {
        return false;
    }
}

Example 3 with FixedWidthVector

use of org.apache.drill.exec.vector.FixedWidthVector in project drill by apache.

the class StreamingAggBatch method constructSpecialBatch.

/**
   * Method is invoked when we have a straight aggregate (no group by expression) and our input is empty.
   * In this case we construct an outgoing batch with record count as 1. For the nullable vectors we don't set anything
   * as we want the output to be NULL. For the required vectors (only for count()) we set the value to be zero since
   * we don't zero out our buffers initially while allocating them.
   */
@SuppressWarnings("resource")
private void constructSpecialBatch() {
    int exprIndex = 0;
    for (final VectorWrapper<?> vw : container) {
        final ValueVector vv = vw.getValueVector();
        AllocationHelper.allocateNew(vv, SPECIAL_BATCH_COUNT);
        vv.getMutator().setValueCount(SPECIAL_BATCH_COUNT);
        if (vv.getField().getType().getMode() == TypeProtos.DataMode.REQUIRED) {
            if (vv instanceof FixedWidthVector) {
                /*
           * The only case we should have a required vector in the aggregate is for count function whose output is
           * always a FixedWidthVector (BigIntVector). Zero out the vector.
           */
                ((FixedWidthVector) vv).zeroVector();
            } else {
                /*
           * If we are in this else block it means that we have a required vector which is of variable length. We
           * should not be here, raising an error since we have set the record count to be 1 and not cleared the
           * buffer
           */
                throw new DrillRuntimeException("FixedWidth vectors is the expected output vector type. " + "Corresponding expression: " + popConfig.getExprs().get(exprIndex).toString());
            }
        }
        exprIndex++;
    }
    container.setRecordCount(SPECIAL_BATCH_COUNT);
    recordCount = SPECIAL_BATCH_COUNT;
}