Examples with MaterializedField org.apache.drill.exec.record.MaterializedField used on opensource projects

Example 41 with MaterializedField

use of org.apache.drill.exec.record.MaterializedField in project drill by apache.

the class PruneScanRule method doOnMatch.

protected void doOnMatch(RelOptRuleCall call, Filter filterRel, Project projectRel, TableScan scanRel) {
    final String pruningClassName = getClass().getName();
    logger.info("Beginning partition pruning, pruning class: {}", pruningClassName);
    Stopwatch totalPruningTime = Stopwatch.createStarted();
    final PlannerSettings settings = PrelUtil.getPlannerSettings(call.getPlanner());
    PartitionDescriptor descriptor = getPartitionDescriptor(settings, scanRel);
    final BufferAllocator allocator = optimizerContext.getAllocator();
    final Object selection = getDrillTable(scanRel).getSelection();
    MetadataContext metaContext = null;
    if (selection instanceof FormatSelection) {
        metaContext = ((FormatSelection) selection).getSelection().getMetaContext();
    }
    RexNode condition = null;
    if (projectRel == null) {
        condition = filterRel.getCondition();
    } else {
        // get the filter as if it were below the projection.
        condition = RelOptUtil.pushFilterPastProject(filterRel.getCondition(), projectRel);
    }
    RewriteAsBinaryOperators visitor = new RewriteAsBinaryOperators(true, filterRel.getCluster().getRexBuilder());
    condition = condition.accept(visitor);
    Map<Integer, String> fieldNameMap = Maps.newHashMap();
    List<String> fieldNames = scanRel.getRowType().getFieldNames();
    BitSet columnBitset = new BitSet();
    BitSet partitionColumnBitSet = new BitSet();
    Map<Integer, Integer> partitionMap = Maps.newHashMap();
    int relColIndex = 0;
    for (String field : fieldNames) {
        final Integer partitionIndex = descriptor.getIdIfValid(field);
        if (partitionIndex != null) {
            fieldNameMap.put(partitionIndex, field);
            partitionColumnBitSet.set(partitionIndex);
            columnBitset.set(relColIndex);
            // mapping between the relColIndex and partitionIndex
            partitionMap.put(relColIndex, partitionIndex);
        }
        relColIndex++;
    }
    if (partitionColumnBitSet.isEmpty()) {
        logger.info("No partition columns are projected from the scan..continue. " + "Total pruning elapsed time: {} ms", totalPruningTime.elapsed(TimeUnit.MILLISECONDS));
        setPruneStatus(metaContext, PruneStatus.NOT_PRUNED);
        return;
    }
    // stop watch to track how long we spend in different phases of pruning
    Stopwatch miscTimer = Stopwatch.createUnstarted();
    // track how long we spend building the filter tree
    miscTimer.start();
    FindPartitionConditions c = new FindPartitionConditions(columnBitset, filterRel.getCluster().getRexBuilder());
    c.analyze(condition);
    RexNode pruneCondition = c.getFinalCondition();
    BitSet referencedDirsBitSet = c.getReferencedDirs();
    logger.info("Total elapsed time to build and analyze filter tree: {} ms", miscTimer.elapsed(TimeUnit.MILLISECONDS));
    miscTimer.reset();
    if (pruneCondition == null) {
        logger.info("No conditions were found eligible for partition pruning." + "Total pruning elapsed time: {} ms", totalPruningTime.elapsed(TimeUnit.MILLISECONDS));
        setPruneStatus(metaContext, PruneStatus.NOT_PRUNED);
        return;
    }
    // set up the partitions
    List<PartitionLocation> newPartitions = Lists.newArrayList();
    // total number of partitions
    long numTotal = 0;
    int batchIndex = 0;
    PartitionLocation firstLocation = null;
    LogicalExpression materializedExpr = null;
    String[] spInfo = null;
    int maxIndex = -1;
    BitSet matchBitSet = new BitSet();
    // Outer loop: iterate over a list of batches of PartitionLocations
    for (List<PartitionLocation> partitions : descriptor) {
        numTotal += partitions.size();
        logger.debug("Evaluating partition pruning for batch {}", batchIndex);
        if (batchIndex == 0) {
            // save the first location in case everything is pruned
            firstLocation = partitions.get(0);
        }
        final NullableBitVector output = new NullableBitVector(MaterializedField.create("", Types.optional(MinorType.BIT)), allocator);
        final VectorContainer container = new VectorContainer();
        try {
            final ValueVector[] vectors = new ValueVector[descriptor.getMaxHierarchyLevel()];
            for (int partitionColumnIndex : BitSets.toIter(partitionColumnBitSet)) {
                SchemaPath column = SchemaPath.getSimplePath(fieldNameMap.get(partitionColumnIndex));
                MajorType type = descriptor.getVectorType(column, settings);
                MaterializedField field = MaterializedField.create(column.getAsUnescapedPath(), type);
                ValueVector v = TypeHelper.getNewVector(field, allocator);
                v.allocateNew();
                vectors[partitionColumnIndex] = v;
                container.add(v);
            }
            // track how long we spend populating partition column vectors
            miscTimer.start();
            // populate partition vectors.
            descriptor.populatePartitionVectors(vectors, partitions, partitionColumnBitSet, fieldNameMap);
            logger.info("Elapsed time to populate partitioning column vectors: {} ms within batchIndex: {}", miscTimer.elapsed(TimeUnit.MILLISECONDS), batchIndex);
            miscTimer.reset();
            // materialize the expression; only need to do this once
            if (batchIndex == 0) {
                materializedExpr = materializePruneExpr(pruneCondition, settings, scanRel, container);
                if (materializedExpr == null) {
                    // continue without partition pruning; no need to log anything here since
                    // materializePruneExpr logs it already
                    logger.info("Total pruning elapsed time: {} ms", totalPruningTime.elapsed(TimeUnit.MILLISECONDS));
                    setPruneStatus(metaContext, PruneStatus.NOT_PRUNED);
                    return;
                }
            }
            output.allocateNew(partitions.size());
            // start the timer to evaluate how long we spend in the interpreter evaluation
            miscTimer.start();
            InterpreterEvaluator.evaluate(partitions.size(), optimizerContext, container, output, materializedExpr);
            logger.info("Elapsed time in interpreter evaluation: {} ms within batchIndex: {} with # of partitions : {}", miscTimer.elapsed(TimeUnit.MILLISECONDS), batchIndex, partitions.size());
            miscTimer.reset();
            int recordCount = 0;
            int qualifiedCount = 0;
            if (descriptor.supportsMetadataCachePruning() && partitions.get(0).isCompositePartition()) /* apply single partition check only for composite partitions */
            {
                // Inner loop: within each batch iterate over the PartitionLocations
                for (PartitionLocation part : partitions) {
                    assert part.isCompositePartition();
                    if (!output.getAccessor().isNull(recordCount) && output.getAccessor().get(recordCount) == 1) {
                        newPartitions.add(part);
                        // Rather than using the PartitionLocation, get the array of partition values for the directories that are
                        // referenced by the filter since we are not interested in directory references in other parts of the query.
                        Pair<String[], Integer> p = composePartition(referencedDirsBitSet, partitionMap, vectors, recordCount);
                        String[] parts = p.getLeft();
                        int tmpIndex = p.getRight();
                        maxIndex = Math.max(maxIndex, tmpIndex);
                        if (spInfo == null) {
                            // initialization
                            spInfo = parts;
                            for (int j = 0; j <= tmpIndex; j++) {
                                if (parts[j] != null) {
                                    matchBitSet.set(j);
                                }
                            }
                        } else {
                            // compare the new partition with existing partition
                            for (int j = 0; j <= tmpIndex; j++) {
                                if (parts[j] == null || spInfo[j] == null) {
                                    // nulls don't match
                                    matchBitSet.clear(j);
                                } else {
                                    if (!parts[j].equals(spInfo[j])) {
                                        matchBitSet.clear(j);
                                    }
                                }
                            }
                        }
                        qualifiedCount++;
                    }
                    recordCount++;
                }
            } else {
                // Inner loop: within each batch iterate over the PartitionLocations
                for (PartitionLocation part : partitions) {
                    if (!output.getAccessor().isNull(recordCount) && output.getAccessor().get(recordCount) == 1) {
                        newPartitions.add(part);
                        qualifiedCount++;
                    }
                    recordCount++;
                }
            }
            logger.debug("Within batch {}: total records: {}, qualified records: {}", batchIndex, recordCount, qualifiedCount);
            batchIndex++;
        } catch (Exception e) {
            logger.warn("Exception while trying to prune partition.", e);
            logger.info("Total pruning elapsed time: {} ms", totalPruningTime.elapsed(TimeUnit.MILLISECONDS));
            setPruneStatus(metaContext, PruneStatus.NOT_PRUNED);
            // continue without partition pruning
            return;
        } finally {
            container.clear();
            if (output != null) {
                output.clear();
            }
        }
    }
    try {
        if (newPartitions.size() == numTotal) {
            logger.info("No partitions were eligible for pruning");
            return;
        }
        // handle the case all partitions are filtered out.
        boolean canDropFilter = true;
        boolean wasAllPartitionsPruned = false;
        String cacheFileRoot = null;
        if (newPartitions.isEmpty()) {
            assert firstLocation != null;
            // Add the first non-composite partition location, since execution requires schema.
            // In such case, we should not drop filter.
            newPartitions.add(firstLocation.getPartitionLocationRecursive().get(0));
            canDropFilter = false;
            // NOTE: with DRILL-4530, the PruneScanRule may be called with only a list of
            // directories first and the non-composite partition location will still return
            // directories, not files.  So, additional processing is done depending on this flag
            wasAllPartitionsPruned = true;
            logger.info("All {} partitions were pruned; added back a single partition to allow creating a schema", numTotal);
            // set the cacheFileRoot appropriately
            if (firstLocation.isCompositePartition()) {
                cacheFileRoot = descriptor.getBaseTableLocation() + firstLocation.getCompositePartitionPath();
            }
        }
        logger.info("Pruned {} partitions down to {}", numTotal, newPartitions.size());
        List<RexNode> conjuncts = RelOptUtil.conjunctions(condition);
        List<RexNode> pruneConjuncts = RelOptUtil.conjunctions(pruneCondition);
        conjuncts.removeAll(pruneConjuncts);
        RexNode newCondition = RexUtil.composeConjunction(filterRel.getCluster().getRexBuilder(), conjuncts, false);
        RewriteCombineBinaryOperators reverseVisitor = new RewriteCombineBinaryOperators(true, filterRel.getCluster().getRexBuilder());
        condition = condition.accept(reverseVisitor);
        pruneCondition = pruneCondition.accept(reverseVisitor);
        if (descriptor.supportsMetadataCachePruning() && !wasAllPartitionsPruned) {
            // if metadata cache file could potentially be used, then assign a proper cacheFileRoot
            int index = -1;
            if (!matchBitSet.isEmpty()) {
                String path = "";
                index = matchBitSet.length() - 1;
                for (int j = 0; j < matchBitSet.length(); j++) {
                    if (!matchBitSet.get(j)) {
                        // stop at the first index with no match and use the immediate
                        // previous index
                        index = j - 1;
                        break;
                    }
                }
                for (int j = 0; j <= index; j++) {
                    path += "/" + spInfo[j];
                }
                cacheFileRoot = descriptor.getBaseTableLocation() + path;
            }
            if (index != maxIndex) {
                // if multiple partitions are being selected, we should not drop the filter
                // since we are reading the cache file at a parent/ancestor level
                canDropFilter = false;
            }
        }
        RelNode inputRel = descriptor.supportsMetadataCachePruning() ? descriptor.createTableScan(newPartitions, cacheFileRoot, wasAllPartitionsPruned, metaContext) : descriptor.createTableScan(newPartitions, wasAllPartitionsPruned);
        if (projectRel != null) {
            inputRel = projectRel.copy(projectRel.getTraitSet(), Collections.singletonList(inputRel));
        }
        if (newCondition.isAlwaysTrue() && canDropFilter) {
            call.transformTo(inputRel);
        } else {
            final RelNode newFilter = filterRel.copy(filterRel.getTraitSet(), Collections.singletonList(inputRel));
            call.transformTo(newFilter);
        }
        setPruneStatus(metaContext, PruneStatus.PRUNED);
    } catch (Exception e) {
        logger.warn("Exception while using the pruned partitions.", e);
    } finally {
        logger.info("Total pruning elapsed time: {} ms", totalPruningTime.elapsed(TimeUnit.MILLISECONDS));
    }
}

Example 42 with MaterializedField

use of org.apache.drill.exec.record.MaterializedField in project drill by apache.

the class MapUtility method writeToMapFromReader.

/*
   * Function to read a value from the field reader, detect the type, construct the appropriate value holder
   * and use the value holder to write to the Map.
   */
// TODO : This should be templatized and generated using freemarker
public static void writeToMapFromReader(FieldReader fieldReader, BaseWriter.MapWriter mapWriter) {
    try {
        MajorType valueMajorType = fieldReader.getType();
        MinorType valueMinorType = valueMajorType.getMinorType();
        boolean repeated = false;
        if (valueMajorType.getMode() == TypeProtos.DataMode.REPEATED) {
            repeated = true;
        }
        switch(valueMinorType) {
            case TINYINT:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).tinyInt());
                } else {
                    fieldReader.copyAsValue(mapWriter.tinyInt(MappifyUtility.fieldValue));
                }
                break;
            case SMALLINT:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).smallInt());
                } else {
                    fieldReader.copyAsValue(mapWriter.smallInt(MappifyUtility.fieldValue));
                }
                break;
            case BIGINT:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).bigInt());
                } else {
                    fieldReader.copyAsValue(mapWriter.bigInt(MappifyUtility.fieldValue));
                }
                break;
            case INT:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).integer());
                } else {
                    fieldReader.copyAsValue(mapWriter.integer(MappifyUtility.fieldValue));
                }
                break;
            case UINT1:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).uInt1());
                } else {
                    fieldReader.copyAsValue(mapWriter.uInt1(MappifyUtility.fieldValue));
                }
                break;
            case UINT2:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).uInt2());
                } else {
                    fieldReader.copyAsValue(mapWriter.uInt2(MappifyUtility.fieldValue));
                }
                break;
            case UINT4:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).uInt4());
                } else {
                    fieldReader.copyAsValue(mapWriter.uInt4(MappifyUtility.fieldValue));
                }
                break;
            case UINT8:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).uInt8());
                } else {
                    fieldReader.copyAsValue(mapWriter.uInt8(MappifyUtility.fieldValue));
                }
                break;
            case DECIMAL9:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).decimal9());
                } else {
                    fieldReader.copyAsValue(mapWriter.decimal9(MappifyUtility.fieldValue));
                }
                break;
            case DECIMAL18:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).decimal18());
                } else {
                    fieldReader.copyAsValue(mapWriter.decimal18(MappifyUtility.fieldValue));
                }
                break;
            case DECIMAL28SPARSE:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).decimal28Sparse());
                } else {
                    fieldReader.copyAsValue(mapWriter.decimal28Sparse(MappifyUtility.fieldValue));
                }
                break;
            case DECIMAL38SPARSE:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).decimal38Sparse());
                } else {
                    fieldReader.copyAsValue(mapWriter.decimal38Sparse(MappifyUtility.fieldValue));
                }
                break;
            case DATE:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).date());
                } else {
                    fieldReader.copyAsValue(mapWriter.date(MappifyUtility.fieldValue));
                }
                break;
            case TIME:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).time());
                } else {
                    fieldReader.copyAsValue(mapWriter.time(MappifyUtility.fieldValue));
                }
                break;
            case TIMESTAMP:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).timeStamp());
                } else {
                    fieldReader.copyAsValue(mapWriter.timeStamp(MappifyUtility.fieldValue));
                }
                break;
            case INTERVAL:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).interval());
                } else {
                    fieldReader.copyAsValue(mapWriter.interval(MappifyUtility.fieldValue));
                }
                break;
            case INTERVALDAY:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).intervalDay());
                } else {
                    fieldReader.copyAsValue(mapWriter.intervalDay(MappifyUtility.fieldValue));
                }
                break;
            case INTERVALYEAR:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).intervalYear());
                } else {
                    fieldReader.copyAsValue(mapWriter.intervalYear(MappifyUtility.fieldValue));
                }
                break;
            case FLOAT4:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).float4());
                } else {
                    fieldReader.copyAsValue(mapWriter.float4(MappifyUtility.fieldValue));
                }
                break;
            case FLOAT8:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).float8());
                } else {
                    fieldReader.copyAsValue(mapWriter.float8(MappifyUtility.fieldValue));
                }
                break;
            case BIT:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).bit());
                } else {
                    fieldReader.copyAsValue(mapWriter.bit(MappifyUtility.fieldValue));
                }
                break;
            case VARCHAR:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).varChar());
                } else {
                    fieldReader.copyAsValue(mapWriter.varChar(MappifyUtility.fieldValue));
                }
                break;
            case VARBINARY:
                if (repeated) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).varBinary());
                } else {
                    fieldReader.copyAsValue(mapWriter.varBinary(MappifyUtility.fieldValue));
                }
                break;
            case MAP:
                if (valueMajorType.getMode() == TypeProtos.DataMode.REPEATED) {
                    fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).map());
                } else {
                    fieldReader.copyAsValue(mapWriter.map(MappifyUtility.fieldValue));
                }
                break;
            case LIST:
                fieldReader.copyAsValue(mapWriter.list(MappifyUtility.fieldValue).list());
                break;
            default:
                throw new DrillRuntimeException(String.format("kvgen does not support input of type: %s", valueMinorType));
        }
    } catch (ClassCastException e) {
        final MaterializedField field = fieldReader.getField();
        throw new DrillRuntimeException(String.format(TYPE_MISMATCH_ERROR, field.getPath(), field.getType()));
    }
}

Example 43 with MaterializedField

use of org.apache.drill.exec.record.MaterializedField 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 44 with MaterializedField

use of org.apache.drill.exec.record.MaterializedField in project drill by apache.

the class ProjectRecordBatch method getExpressionList.

private List<NamedExpression> getExpressionList() {
    if (popConfig.getExprs() != null) {
        return popConfig.getExprs();
    }
    final List<NamedExpression> exprs = Lists.newArrayList();
    for (final MaterializedField field : incoming.getSchema()) {
        if (Types.isComplex(field.getType()) || Types.isRepeated(field.getType())) {
            final LogicalExpression convertToJson = FunctionCallFactory.createConvert(ConvertExpression.CONVERT_TO, "JSON", SchemaPath.getSimplePath(field.getPath()), ExpressionPosition.UNKNOWN);
            final String castFuncName = CastFunctions.getCastFunc(MinorType.VARCHAR);
            final List<LogicalExpression> castArgs = Lists.newArrayList();
            //input_expr
            castArgs.add(convertToJson);
            // implicitly casting to varchar, since we don't know actual source length, cast to undefined length, which will preserve source length
            castArgs.add(new ValueExpressions.LongExpression(Types.MAX_VARCHAR_LENGTH, null));
            final FunctionCall castCall = new FunctionCall(castFuncName, castArgs, ExpressionPosition.UNKNOWN);
            exprs.add(new NamedExpression(castCall, new FieldReference(field.getPath())));
        } else {
            exprs.add(new NamedExpression(SchemaPath.getSimplePath(field.getPath()), new FieldReference(field.getPath())));
        }
    }
    return exprs;
}

Example 45 with MaterializedField

use of org.apache.drill.exec.record.MaterializedField in project drill by apache.

the class OrderedPartitionRecordBatch method getCopier.

/**
   * Creates a copier that does a project for every Nth record from a VectorContainer incoming into VectorContainer
   * outgoing. Each Ordering in orderings generates a column, and evaluation of the expression associated with each
   * Ordering determines the value of each column. These records will later be sorted based on the values in each
   * column, in the same order as the orderings.
   *
   * @param sv4
   * @param incoming
   * @param outgoing
   * @param orderings
   * @return
   * @throws SchemaChangeException
   */
private SampleCopier getCopier(SelectionVector4 sv4, VectorContainer incoming, VectorContainer outgoing, List<Ordering> orderings, List<ValueVector> localAllocationVectors) throws SchemaChangeException {
    final ErrorCollector collector = new ErrorCollectorImpl();
    final ClassGenerator<SampleCopier> cg = CodeGenerator.getRoot(SampleCopier.TEMPLATE_DEFINITION, context.getFunctionRegistry(), context.getOptions());
    // Note: disabled for now. This may require some debugging:
    // no tests are available for this operator.
    //    cg.getCodeGenerator().plainOldJavaCapable(true);
    // Uncomment out this line to debug the generated code.
    //    cg.getCodeGenerator().saveCodeForDebugging(true);
    int i = 0;
    for (Ordering od : orderings) {
        final LogicalExpression expr = ExpressionTreeMaterializer.materialize(od.getExpr(), incoming, collector, context.getFunctionRegistry());
        SchemaPath schemaPath = SchemaPath.getSimplePath("f" + i++);
        TypeProtos.MajorType.Builder builder = TypeProtos.MajorType.newBuilder().mergeFrom(expr.getMajorType()).clearMode().setMode(TypeProtos.DataMode.REQUIRED);
        TypeProtos.MajorType newType = builder.build();
        MaterializedField outputField = MaterializedField.create(schemaPath.getAsUnescapedPath(), newType);
        if (collector.hasErrors()) {
            throw new SchemaChangeException(String.format("Failure while trying to materialize incoming schema.  Errors:\n %s.", collector.toErrorString()));
        }
        @SuppressWarnings("resource") ValueVector vector = TypeHelper.getNewVector(outputField, oContext.getAllocator());
        localAllocationVectors.add(vector);
        TypedFieldId fid = outgoing.add(vector);
        ValueVectorWriteExpression write = new ValueVectorWriteExpression(fid, expr, true);
        HoldingContainer hc = cg.addExpr(write);
        cg.getEvalBlock()._if(hc.getValue().eq(JExpr.lit(0)))._then()._return(JExpr.FALSE);
    }
    cg.rotateBlock();
    cg.getEvalBlock()._return(JExpr.TRUE);
    outgoing.buildSchema(BatchSchema.SelectionVectorMode.NONE);
    try {
        SampleCopier sampleCopier = context.getImplementationClass(cg);
        sampleCopier.setupCopier(context, sv4, incoming, outgoing);
        return sampleCopier;
    } catch (ClassTransformationException | IOException e) {
        throw new SchemaChangeException(e);
    }
}