Examples with SchemaPath org.apache.drill.common.expression.SchemaPath used on opensource projects

Example 51 with SchemaPath

use of org.apache.drill.common.expression.SchemaPath 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);
    }
}

Example 52 with SchemaPath

use of org.apache.drill.common.expression.SchemaPath in project drill by apache.

the class OrderedPartitionRecordBatch method buildTable.

private void buildTable() throws SchemaChangeException, ClassTransformationException, IOException {
    // Get all samples from distributed map
    @SuppressWarnings("resource") SortRecordBatchBuilder containerBuilder = new SortRecordBatchBuilder(context.getAllocator());
    final VectorContainer allSamplesContainer = new VectorContainer();
    final VectorContainer candidatePartitionTable = new VectorContainer();
    CachedVectorContainer wrap = null;
    try {
        for (CachedVectorContainer w : mmap.get(mapKey)) {
            containerBuilder.add(w.get());
        }
        containerBuilder.build(context, allSamplesContainer);
        List<Ordering> orderDefs = Lists.newArrayList();
        int i = 0;
        for (Ordering od : popConfig.getOrderings()) {
            SchemaPath sp = SchemaPath.getSimplePath("f" + i++);
            orderDefs.add(new Ordering(od.getDirection(), new FieldReference(sp)));
        }
        // sort the data incoming samples.
        @SuppressWarnings("resource") SelectionVector4 newSv4 = containerBuilder.getSv4();
        Sorter sorter = SortBatch.createNewSorter(context, orderDefs, allSamplesContainer);
        sorter.setup(context, newSv4, allSamplesContainer);
        sorter.sort(newSv4, allSamplesContainer);
        // Copy every Nth record from the samples into a candidate partition table, where N = totalSampledRecords/partitions
        // Attempt to push this to the distributed map. Only the first candidate to get pushed will be used.
        SampleCopier copier = null;
        List<ValueVector> localAllocationVectors = Lists.newArrayList();
        copier = getCopier(newSv4, allSamplesContainer, candidatePartitionTable, orderDefs, localAllocationVectors);
        int allocationSize = 50;
        while (true) {
            for (ValueVector vv : localAllocationVectors) {
                AllocationHelper.allocate(vv, samplingFactor * partitions, allocationSize);
            }
            int skipRecords = containerBuilder.getSv4().getTotalCount() / partitions;
            if (copier.copyRecords(skipRecords, skipRecords, partitions - 1)) {
                assert copier.getOutputRecords() == partitions - 1 : String.format("output records: %d partitions: %d", copier.getOutputRecords(), partitions);
                for (VectorWrapper<?> vw : candidatePartitionTable) {
                    vw.getValueVector().getMutator().setValueCount(copier.getOutputRecords());
                }
                break;
            } else {
                candidatePartitionTable.zeroVectors();
                allocationSize *= 2;
            }
        }
        candidatePartitionTable.setRecordCount(copier.getOutputRecords());
        @SuppressWarnings("resource") WritableBatch batch = WritableBatch.getBatchNoHVWrap(candidatePartitionTable.getRecordCount(), candidatePartitionTable, false);
        wrap = new CachedVectorContainer(batch, context.getDrillbitContext().getAllocator());
        tableMap.putIfAbsent(mapKey + "final", wrap, 1, TimeUnit.MINUTES);
    } finally {
        candidatePartitionTable.clear();
        allSamplesContainer.clear();
        containerBuilder.clear();
        containerBuilder.close();
        if (wrap != null) {
            wrap.clear();
        }
    }
}

Example 53 with SchemaPath

use of org.apache.drill.common.expression.SchemaPath in project drill by apache.

the class KuduScanBatchCreator method getBatch.

@Override
public ScanBatch getBatch(FragmentContext context, KuduSubScan subScan, List<RecordBatch> children) throws ExecutionSetupException {
    Preconditions.checkArgument(children.isEmpty());
    List<RecordReader> readers = Lists.newArrayList();
    List<SchemaPath> columns = null;
    for (KuduSubScan.KuduSubScanSpec scanSpec : subScan.getTabletScanSpecList()) {
        try {
            if ((columns = subScan.getColumns()) == null) {
                columns = GroupScan.ALL_COLUMNS;
            }
            readers.add(new KuduRecordReader(subScan.getStorageEngine().getClient(), scanSpec, columns, context));
        } catch (Exception e1) {
            throw new ExecutionSetupException(e1);
        }
    }
    return new ScanBatch(subScan, context, readers.iterator());
}

Example 54 with SchemaPath

use of org.apache.drill.common.expression.SchemaPath in project drill by apache.

the class MongoRecordReader method transformColumns.

@Override
protected Collection<SchemaPath> transformColumns(Collection<SchemaPath> projectedColumns) {
    Set<SchemaPath> transformed = Sets.newLinkedHashSet();
    if (!isStarQuery()) {
        for (SchemaPath column : projectedColumns) {
            String fieldName = column.getRootSegment().getPath();
            transformed.add(column);
            this.fields.put(fieldName, Integer.valueOf(1));
        }
    } else {
        // Tale all the fields including the _id
        this.fields.remove(DrillMongoConstants.ID);
        transformed.add(AbstractRecordReader.STAR_COLUMN);
    }
    return transformed;
}

Example 55 with SchemaPath

use of org.apache.drill.common.expression.SchemaPath in project drill by apache.

the class ParquetGroupScan method checkForPartitionColumn.

/**
   * When reading the very first footer, any column is a potential partition column. So for the first footer, we check
   * every column to see if it is single valued, and if so, add it to the list of potential partition columns. For the
   * remaining footers, we will not find any new partition columns, but we may discover that what was previously a
   * potential partition column now no longer qualifies, so it needs to be removed from the list.
   * @return whether column is a potential partition column
   */
private boolean checkForPartitionColumn(ColumnMetadata columnMetadata, boolean first) {
    SchemaPath schemaPath = SchemaPath.getCompoundPath(columnMetadata.getName());
    final PrimitiveTypeName primitiveType;
    final OriginalType originalType;
    if (this.parquetTableMetadata.hasColumnMetadata()) {
        primitiveType = this.parquetTableMetadata.getPrimitiveType(columnMetadata.getName());
        originalType = this.parquetTableMetadata.getOriginalType(columnMetadata.getName());
    } else {
        primitiveType = columnMetadata.getPrimitiveType();
        originalType = columnMetadata.getOriginalType();
    }
    if (first) {
        if (hasSingleValue(columnMetadata)) {
            partitionColTypeMap.put(schemaPath, getType(primitiveType, originalType));
            return true;
        } else {
            return false;
        }
    } else {
        if (!partitionColTypeMap.keySet().contains(schemaPath)) {
            return false;
        } else {
            if (!hasSingleValue(columnMetadata)) {
                partitionColTypeMap.remove(schemaPath);
                return false;
            }
            if (!getType(primitiveType, originalType).equals(partitionColTypeMap.get(schemaPath))) {
                partitionColTypeMap.remove(schemaPath);
                return false;
            }
        }
    }
    return true;
}