Examples with BufferAllocator org.apache.drill.exec.memory.BufferAllocator used on opensource projects

Example 26 with BufferAllocator

use of org.apache.drill.exec.memory.BufferAllocator in project drill by axbaretto.

the class ExternalSortBatch method createCopier.

private void createCopier(VectorAccessible batch, List<BatchGroup> batchGroupList, VectorContainer outputContainer, boolean spilling) throws SchemaChangeException {
    try {
        if (copier == null) {
            CodeGenerator<PriorityQueueCopier> cg = CodeGenerator.get(PriorityQueueCopier.TEMPLATE_DEFINITION, context.getOptions());
            cg.plainJavaCapable(true);
            // Uncomment out this line to debug the generated code.
            // cg.saveCodeForDebugging(true);
            ClassGenerator<PriorityQueueCopier> g = cg.getRoot();
            generateComparisons(g, batch);
            g.setMappingSet(COPIER_MAPPING_SET);
            CopyUtil.generateCopies(g, batch, true);
            g.setMappingSet(MAIN_MAPPING);
            copier = context.getImplementationClass(cg);
        } else {
            copier.close();
        }
        @SuppressWarnings("resource") BufferAllocator allocator = spilling ? copierAllocator : oAllocator;
        for (VectorWrapper<?> i : batch) {
            @SuppressWarnings("resource") ValueVector v = TypeHelper.getNewVector(i.getField(), allocator);
            outputContainer.add(v);
        }
        copier.setup(context, allocator, batch, batchGroupList, outputContainer);
    } catch (ClassTransformationException | IOException e) {
        throw new RuntimeException(e);
    }
}

Example 27 with BufferAllocator

use of org.apache.drill.exec.memory.BufferAllocator in project drill by axbaretto.

the class BufferedBatches method bufferBatch.

@SuppressWarnings("resource")
private void bufferBatch(VectorContainer convertedBatch, SelectionVector2 sv2, long netSize) {
    BufferAllocator allocator = context.getAllocator();
    RecordBatchData rbd = new RecordBatchData(convertedBatch, allocator);
    try {
        rbd.setSv2(sv2);
        bufferedBatches.add(new BatchGroup.InputBatch(rbd.getContainer(), rbd.getSv2(), allocator, netSize));
    } catch (Throwable t) {
        rbd.clear();
        throw t;
    }
}

Example 28 with BufferAllocator

use of org.apache.drill.exec.memory.BufferAllocator in project drill by axbaretto.

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;
    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.getLastSegment().getNameSegment().getPath(), 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 29 with BufferAllocator

use of org.apache.drill.exec.memory.BufferAllocator in project drill by axbaretto.

the class WebUserConnection method sendData.

@Override
public void sendData(RpcOutcomeListener<Ack> listener, QueryWritableBatch result) {
    // Check if there is any data or not. There can be overflow here but DrillBuf doesn't support allocating with
    // bytes in long. Hence we are just preserving the earlier behavior and logging debug log for the case.
    final int dataByteCount = (int) result.getByteCount();
    if (dataByteCount <= 0) {
        if (logger.isDebugEnabled()) {
            logger.debug("Either no data received in this batch or there is BufferOverflow in dataByteCount: {}", dataByteCount);
        }
        listener.success(Acks.OK, null);
        return;
    }
    // If here that means there is some data for sure. Create a ByteBuf with all the data in it.
    final int rows = result.getHeader().getRowCount();
    final BufferAllocator allocator = webSessionResources.getAllocator();
    final DrillBuf bufferWithData = allocator.buffer(dataByteCount);
    try {
        final ByteBuf[] resultDataBuffers = result.getBuffers();
        for (final ByteBuf buffer : resultDataBuffers) {
            bufferWithData.writeBytes(buffer);
            buffer.release();
        }
        final RecordBatchLoader loader = new RecordBatchLoader(allocator);
        try {
            loader.load(result.getHeader().getDef(), bufferWithData);
            // SchemaChangeException, so check/clean catch clause below.
            for (int i = 0; i < loader.getSchema().getFieldCount(); ++i) {
                columns.add(loader.getSchema().getColumn(i).getName());
            }
            for (int i = 0; i < rows; ++i) {
                final Map<String, String> record = Maps.newHashMap();
                for (VectorWrapper<?> vw : loader) {
                    final String field = vw.getValueVector().getMetadata().getNamePart().getName();
                    final Accessor accessor = vw.getValueVector().getAccessor();
                    final Object value = i < accessor.getValueCount() ? accessor.getObject(i) : null;
                    final String display = value == null ? null : value.toString();
                    record.put(field, display);
                }
                results.add(record);
            }
        } finally {
            loader.clear();
        }
    } catch (Exception e) {
        exception = UserException.systemError(e).build(logger);
    } finally {
        // Notify the listener with ACK.OK both in error/success case because data was send successfully from Drillbit.
        bufferWithData.release();
        listener.success(Acks.OK, null);
    }
}

Example 30 with BufferAllocator

use of org.apache.drill.exec.memory.BufferAllocator in project drill by axbaretto.

the class TestLoad method testMapSchemaChange.

@Test
public void testMapSchemaChange() throws SchemaChangeException {
    final BufferAllocator allocator = RootAllocatorFactory.newRoot(drillConfig);
    final RecordBatchLoader batchLoader = new RecordBatchLoader(allocator);
    // Initial schema: a: INT, m: MAP{}
    BatchSchema schema1 = new SchemaBuilder().add("a", MinorType.INT).addMap("m").resumeSchema().build();
    {
        assertTrue(loadBatch(allocator, batchLoader, schema1));
        assertTrue(schema1.isEquivalent(batchLoader.getSchema()));
        batchLoader.getContainer().zeroVectors();
    }
    // Same schema
    // Schema change: No
    {
        assertFalse(loadBatch(allocator, batchLoader, schema1));
        assertTrue(schema1.isEquivalent(batchLoader.getSchema()));
        batchLoader.getContainer().zeroVectors();
    }
    // Add column to map: a: INT, m: MAP{b: VARCHAR}
    // Schema change: Yes
    BatchSchema schema2 = new SchemaBuilder().add("a", MinorType.INT).addMap("m").add("b", MinorType.VARCHAR).resumeSchema().build();
    {
        assertTrue(loadBatch(allocator, batchLoader, schema2));
        assertTrue(schema2.isEquivalent(batchLoader.getSchema()));
        batchLoader.getContainer().zeroVectors();
    }
    // Same schema
    // Schema change: No
    {
        assertFalse(loadBatch(allocator, batchLoader, schema2));
        assertTrue(schema2.isEquivalent(batchLoader.getSchema()));
        batchLoader.getContainer().zeroVectors();
    }
    // Add column:  a: INT, m: MAP{b: VARCHAR, c: INT}
    // Schema change: Yes
    {
        BatchSchema schema = new SchemaBuilder().add("a", MinorType.INT).addMap("m").add("b", MinorType.VARCHAR).add("c", MinorType.INT).resumeSchema().build();
        assertTrue(loadBatch(allocator, batchLoader, schema));
        assertTrue(schema.isEquivalent(batchLoader.getSchema()));
        batchLoader.getContainer().zeroVectors();
    }
    // Drop a column:  a: INT, m: MAP{b: VARCHAR}
    // Schema change: Yes
    {
        BatchSchema schema = new SchemaBuilder().add("a", MinorType.INT).addMap("m").add("b", MinorType.VARCHAR).resumeSchema().build();
        assertTrue(loadBatch(allocator, batchLoader, schema));
        assertTrue(schema.isEquivalent(batchLoader.getSchema()));
        batchLoader.getContainer().zeroVectors();
    }
    // Change type:  a: INT, m: MAP{b: INT}
    // Schema change: Yes
    {
        BatchSchema schema = new SchemaBuilder().add("a", MinorType.INT).addMap("m").add("b", MinorType.INT).resumeSchema().build();
        assertTrue(loadBatch(allocator, batchLoader, schema));
        assertTrue(schema.isEquivalent(batchLoader.getSchema()));
        batchLoader.getContainer().zeroVectors();
    }
    // Empty map: a: INT, m: MAP{}
    {
        assertTrue(loadBatch(allocator, batchLoader, schema1));
        assertTrue(schema1.isEquivalent(batchLoader.getSchema()));
        batchLoader.getContainer().zeroVectors();
    }
    // Drop map: a: INT
    {
        BatchSchema schema = new SchemaBuilder().add("a", MinorType.INT).build();
        assertTrue(loadBatch(allocator, batchLoader, schema));
        assertTrue(schema.isEquivalent(batchLoader.getSchema()));
        batchLoader.getContainer().zeroVectors();
    }
    batchLoader.clear();
    allocator.close();
}