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

Example 46 with MaterializedField

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

the class DrillTestWrapper method addToCombinedVectorResults.

/**
   * Add to result vectors and compare batch schema against expected schema while iterating batches.
   * @param batches
   * @param  expectedSchema: the expected schema the batches should contain. Through SchemaChangeException
   *                       if encounter different batch schema.
   * @return
   * @throws SchemaChangeException
   * @throws UnsupportedEncodingException
   */
public static Map<String, List<Object>> addToCombinedVectorResults(Iterable<VectorAccessible> batches, BatchSchema expectedSchema) throws SchemaChangeException, UnsupportedEncodingException {
    // TODO - this does not handle schema changes
    Map<String, List<Object>> combinedVectors = new TreeMap<>();
    long totalRecords = 0;
    BatchSchema schema = null;
    for (VectorAccessible loader : batches) {
        if (expectedSchema != null) {
            if (!expectedSchema.equals(loader.getSchema())) {
                throw new SchemaChangeException(String.format("Batch schema does not match expected schema\n" + "Actual schema: %s.  Expected schema : %s", loader.getSchema(), expectedSchema));
            }
        }
        // SchemaChangeException, so check/clean throws clause above.
        if (schema == null) {
            schema = loader.getSchema();
            for (MaterializedField mf : schema) {
                combinedVectors.put(SchemaPath.getSimplePath(mf.getPath()).toExpr(), new ArrayList<Object>());
            }
        } else {
            // TODO - actually handle schema changes, this is just to get access to the SelectionVectorMode
            // of the current batch, the check for a null schema is used to only mutate the schema once
            // need to add new vectors and null fill for previous batches? distinction between null and non-existence important?
            schema = loader.getSchema();
        }
        logger.debug("reading batch with " + loader.getRecordCount() + " rows, total read so far " + totalRecords);
        totalRecords += loader.getRecordCount();
        for (VectorWrapper<?> w : loader) {
            String field = SchemaPath.getSimplePath(w.getField().getPath()).toExpr();
            ValueVector[] vectors;
            if (w.isHyper()) {
                vectors = w.getValueVectors();
            } else {
                vectors = new ValueVector[] { w.getValueVector() };
            }
            SelectionVector2 sv2 = null;
            SelectionVector4 sv4 = null;
            switch(schema.getSelectionVectorMode()) {
                case TWO_BYTE:
                    sv2 = loader.getSelectionVector2();
                    break;
                case FOUR_BYTE:
                    sv4 = loader.getSelectionVector4();
                    break;
            }
            if (sv4 != null) {
                for (int j = 0; j < sv4.getCount(); j++) {
                    int complexIndex = sv4.get(j);
                    int batchIndex = complexIndex >> 16;
                    int recordIndexInBatch = complexIndex & 65535;
                    Object obj = vectors[batchIndex].getAccessor().getObject(recordIndexInBatch);
                    if (obj != null) {
                        if (obj instanceof Text) {
                            obj = obj.toString();
                        }
                    }
                    combinedVectors.get(field).add(obj);
                }
            } else {
                for (ValueVector vv : vectors) {
                    for (int j = 0; j < loader.getRecordCount(); j++) {
                        int index;
                        if (sv2 != null) {
                            index = sv2.getIndex(j);
                        } else {
                            index = j;
                        }
                        Object obj = vv.getAccessor().getObject(index);
                        if (obj != null) {
                            if (obj instanceof Text) {
                                obj = obj.toString();
                            }
                        }
                        combinedVectors.get(field).add(obj);
                    }
                }
            }
        }
    }
    return combinedVectors;
}

Example 47 with MaterializedField

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

the class DrillTestWrapper method addToHyperVectorMap.

private Map<String, HyperVectorValueIterator> addToHyperVectorMap(final List<QueryDataBatch> records, final RecordBatchLoader loader) throws SchemaChangeException, UnsupportedEncodingException {
    // TODO - this does not handle schema changes
    Map<String, HyperVectorValueIterator> combinedVectors = new TreeMap<>();
    long totalRecords = 0;
    QueryDataBatch batch;
    int size = records.size();
    for (int i = 0; i < size; i++) {
        batch = records.get(i);
        loader.load(batch.getHeader().getDef(), batch.getData());
        logger.debug("reading batch with " + loader.getRecordCount() + " rows, total read so far " + totalRecords);
        totalRecords += loader.getRecordCount();
        for (VectorWrapper<?> w : loader) {
            String field = SchemaPath.getSimplePath(w.getField().getPath()).toExpr();
            if (!combinedVectors.containsKey(field)) {
                MaterializedField mf = w.getField();
                ValueVector[] vvList = (ValueVector[]) Array.newInstance(mf.getValueClass(), 1);
                vvList[0] = w.getValueVector();
                combinedVectors.put(field, new HyperVectorValueIterator(mf, new HyperVectorWrapper<>(mf, vvList)));
            } else {
                combinedVectors.get(field).getHyperVector().addVector(w.getValueVector());
            }
        }
    }
    for (HyperVectorValueIterator hvi : combinedVectors.values()) {
        hvi.determineTotalSize();
    }
    return combinedVectors;
}

Example 48 with MaterializedField

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

the class NestedLoopJoinBatch method setupWorker.

/**
   * 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);
}

Example 49 with MaterializedField

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

the class HashAggBatch method createAggregatorInternal.

private HashAggregator createAggregatorInternal() throws SchemaChangeException, ClassTransformationException, IOException {
    CodeGenerator<HashAggregator> top = CodeGenerator.get(HashAggregator.TEMPLATE_DEFINITION, context.getFunctionRegistry(), context.getOptions());
    ClassGenerator<HashAggregator> cg = top.getRoot();
    ClassGenerator<HashAggregator> cgInner = cg.getInnerGenerator("BatchHolder");
    top.plainJavaCapable(true);
    // Uncomment out this line to debug the generated code.
    //    top.saveCodeForDebugging(true);
    container.clear();
    int numGroupByExprs = (popConfig.getGroupByExprs() != null) ? popConfig.getGroupByExprs().size() : 0;
    int numAggrExprs = (popConfig.getAggrExprs() != null) ? popConfig.getAggrExprs().size() : 0;
    aggrExprs = new LogicalExpression[numAggrExprs];
    groupByOutFieldIds = new TypedFieldId[numGroupByExprs];
    aggrOutFieldIds = new TypedFieldId[numAggrExprs];
    ErrorCollector collector = new ErrorCollectorImpl();
    int i;
    for (i = 0; i < numGroupByExprs; i++) {
        NamedExpression ne = popConfig.getGroupByExprs().get(i);
        final LogicalExpression expr = ExpressionTreeMaterializer.materialize(ne.getExpr(), incoming, collector, context.getFunctionRegistry());
        if (expr == null) {
            continue;
        }
        final MaterializedField outputField = MaterializedField.create(ne.getRef().getAsNamePart().getName(), expr.getMajorType());
        @SuppressWarnings("resource") ValueVector vv = TypeHelper.getNewVector(outputField, oContext.getAllocator());
        // add this group-by vector to the output container
        groupByOutFieldIds[i] = container.add(vv);
    }
    for (i = 0; i < numAggrExprs; i++) {
        NamedExpression ne = popConfig.getAggrExprs().get(i);
        final LogicalExpression expr = ExpressionTreeMaterializer.materialize(ne.getExpr(), incoming, collector, context.getFunctionRegistry());
        if (expr instanceof IfExpression) {
            throw UserException.unsupportedError(new UnsupportedOperationException("Union type not supported in aggregate functions")).build(logger);
        }
        if (collector.hasErrors()) {
            throw new SchemaChangeException("Failure while materializing expression. " + collector.toErrorString());
        }
        if (expr == null) {
            continue;
        }
        final MaterializedField outputField = MaterializedField.create(ne.getRef().getAsNamePart().getName(), expr.getMajorType());
        @SuppressWarnings("resource") ValueVector vv = TypeHelper.getNewVector(outputField, oContext.getAllocator());
        aggrOutFieldIds[i] = container.add(vv);
        aggrExprs[i] = new ValueVectorWriteExpression(aggrOutFieldIds[i], expr, true);
    }
    setupUpdateAggrValues(cgInner);
    setupGetIndex(cg);
    cg.getBlock("resetValues")._return(JExpr.TRUE);
    container.buildSchema(SelectionVectorMode.NONE);
    HashAggregator agg = context.getImplementationClass(top);
    HashTableConfig htConfig = // TODO - fix the validator on this option
    new HashTableConfig((int) context.getOptions().getOption(ExecConstants.MIN_HASH_TABLE_SIZE), HashTable.DEFAULT_LOAD_FACTOR, popConfig.getGroupByExprs(), null, /* no probe exprs */
    comparators);
    agg.setup(popConfig, htConfig, context, this.stats, oContext.getAllocator(), incoming, this, aggrExprs, cgInner.getWorkspaceTypes(), groupByOutFieldIds, this.container);
    return agg;
}

Example 50 with MaterializedField

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

the class PriorityQueueTemplate method resetQueue.

@Override
public void resetQueue(VectorContainer container, SelectionVector4 v4) throws SchemaChangeException {
    assert container.getSchema().getSelectionVectorMode() == BatchSchema.SelectionVectorMode.FOUR_BYTE;
    BatchSchema schema = container.getSchema();
    VectorContainer newContainer = new VectorContainer();
    for (MaterializedField field : schema) {
        int[] ids = container.getValueVectorId(SchemaPath.getSimplePath(field.getPath())).getFieldIds();
        newContainer.add(container.getValueAccessorById(field.getValueClass(), ids).getValueVectors());
    }
    newContainer.buildSchema(BatchSchema.SelectionVectorMode.FOUR_BYTE);
    // Cleanup before recreating hyperbatch and sv4.
    cleanup();
    hyperBatch = new ExpandableHyperContainer(newContainer);
    batchCount = hyperBatch.iterator().next().getValueVectors().length;
    @SuppressWarnings("resource") final DrillBuf drillBuf = allocator.buffer(4 * (limit + 1));
    heapSv4 = new SelectionVector4(drillBuf, limit, Character.MAX_VALUE);
    // Reset queue size (most likely to be set to limit).
    queueSize = 0;
    for (int i = 0; i < v4.getTotalCount(); i++) {
        heapSv4.set(i, v4.get(i));
        ++queueSize;
    }
    v4.clear();
    doSetup(context, hyperBatch, null);
}