Examples with PlannerSettings org.apache.drill.exec.planner.physical.PlannerSettings used on opensource projects

Example 21 with PlannerSettings

use of org.apache.drill.exec.planner.physical.PlannerSettings in project drill by axbaretto.

the class HivePushPartitionFilterIntoScan method getFilterOnScan.

public static final StoragePluginOptimizerRule getFilterOnScan(OptimizerRulesContext optimizerRulesContext, final String defaultPartitionValue) {
    return new PruneScanRule(RelOptHelper.some(DrillFilterRel.class, RelOptHelper.any(DrillScanRel.class)), "HivePushPartitionFilterIntoScan:Filter_On_Scan_Hive", optimizerRulesContext) {

        @Override
        public PartitionDescriptor getPartitionDescriptor(PlannerSettings settings, TableScan scanRel) {
            return new HivePartitionDescriptor(settings, (DrillScanRel) scanRel, getOptimizerRulesContext().getManagedBuffer(), defaultPartitionValue);
        }

        @Override
        public boolean matches(RelOptRuleCall call) {
            final DrillScanRel scan = (DrillScanRel) call.rel(1);
            GroupScan groupScan = scan.getGroupScan();
            // this rule is applicable only for Hive based partition pruning
            if (PrelUtil.getPlannerSettings(scan.getCluster().getPlanner()).isHepPartitionPruningEnabled()) {
                return groupScan instanceof HiveScan && groupScan.supportsPartitionFilterPushdown() && !scan.partitionFilterPushdown();
            } else {
                return groupScan instanceof HiveScan && groupScan.supportsPartitionFilterPushdown();
            }
        }

        @Override
        public void onMatch(RelOptRuleCall call) {
            final DrillFilterRel filterRel = call.rel(0);
            final DrillScanRel scanRel = call.rel(1);
            doOnMatch(call, filterRel, null, scanRel);
        }
    };
}

Example 22 with PlannerSettings

use of org.apache.drill.exec.planner.physical.PlannerSettings in project drill by axbaretto.

the class ParquetPruneScanRule method getFilterOnProjectParquet.

public static final RelOptRule getFilterOnProjectParquet(OptimizerRulesContext optimizerRulesContext) {
    return new PruneScanRule(RelOptHelper.some(DrillFilterRel.class, RelOptHelper.some(DrillProjectRel.class, RelOptHelper.any(DrillScanRel.class))), "PruneScanRule:Filter_On_Project_Parquet", optimizerRulesContext) {

        @Override
        public PartitionDescriptor getPartitionDescriptor(PlannerSettings settings, TableScan scanRel) {
            return new ParquetPartitionDescriptor(settings, (DrillScanRel) scanRel);
        }

        @Override
        public boolean matches(RelOptRuleCall call) {
            final DrillScanRel scan = call.rel(2);
            GroupScan groupScan = scan.getGroupScan();
            // this rule is applicable only for parquet based partition pruning
            if (PrelUtil.getPlannerSettings(scan.getCluster().getPlanner()).isHepPartitionPruningEnabled()) {
                return groupScan instanceof ParquetGroupScan && groupScan.supportsPartitionFilterPushdown() && !scan.partitionFilterPushdown();
            } else {
                return groupScan instanceof ParquetGroupScan && groupScan.supportsPartitionFilterPushdown();
            }
        }

        @Override
        public void onMatch(RelOptRuleCall call) {
            final DrillFilterRel filterRel = call.rel(0);
            final DrillProjectRel projectRel = call.rel(1);
            final DrillScanRel scanRel = call.rel(2);
            doOnMatch(call, filterRel, projectRel, scanRel);
        }
    };
}

Example 23 with PlannerSettings

use of org.apache.drill.exec.planner.physical.PlannerSettings 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 24 with PlannerSettings

use of org.apache.drill.exec.planner.physical.PlannerSettings in project drill by axbaretto.

the class DrillReduceAggregatesRule method reduceSum.

private RexNode reduceSum(Aggregate oldAggRel, AggregateCall oldCall, List<AggregateCall> newCalls, Map<AggregateCall, RexNode> aggCallMapping) {
    final PlannerSettings plannerSettings = (PlannerSettings) oldAggRel.getCluster().getPlanner().getContext();
    final boolean isInferenceEnabled = plannerSettings.isTypeInferenceEnabled();
    final int nGroups = oldAggRel.getGroupCount();
    RelDataTypeFactory typeFactory = oldAggRel.getCluster().getTypeFactory();
    RexBuilder rexBuilder = oldAggRel.getCluster().getRexBuilder();
    int arg = oldCall.getArgList().get(0);
    RelDataType argType = getFieldType(oldAggRel.getInput(), arg);
    final RelDataType sumType;
    final SqlAggFunction sumZeroAgg;
    if (isInferenceEnabled) {
        sumType = oldCall.getType();
    } else {
        sumType = typeFactory.createTypeWithNullability(oldCall.getType(), argType.isNullable());
    }
    sumZeroAgg = new DrillCalciteSqlAggFunctionWrapper(new SqlSumEmptyIsZeroAggFunction(), sumType);
    AggregateCall sumZeroCall = AggregateCall.create(sumZeroAgg, oldCall.isDistinct(), oldCall.isApproximate(), oldCall.getArgList(), -1, sumType, null);
    final SqlCountAggFunction countAgg = (SqlCountAggFunction) SqlStdOperatorTable.COUNT;
    final RelDataType countType = countAgg.getReturnType(typeFactory);
    AggregateCall countCall = AggregateCall.create(countAgg, oldCall.isDistinct(), oldCall.isApproximate(), oldCall.getArgList(), -1, countType, null);
    // NOTE:  these references are with respect to the output
    // of newAggRel
    RexNode sumZeroRef = rexBuilder.addAggCall(sumZeroCall, nGroups, oldAggRel.indicator, newCalls, aggCallMapping, ImmutableList.of(argType));
    if (!oldCall.getType().isNullable()) {
        // null). Therefore we translate to SUM0(x).
        return sumZeroRef;
    }
    RexNode countRef = rexBuilder.addAggCall(countCall, nGroups, oldAggRel.indicator, newCalls, aggCallMapping, ImmutableList.of(argType));
    return rexBuilder.makeCall(SqlStdOperatorTable.CASE, rexBuilder.makeCall(SqlStdOperatorTable.EQUALS, countRef, rexBuilder.makeExactLiteral(BigDecimal.ZERO)), rexBuilder.constantNull(), sumZeroRef);
}

Example 25 with PlannerSettings

use of org.apache.drill.exec.planner.physical.PlannerSettings in project drill by axbaretto.

the class PlanningBase method testSqlPlan.

protected void testSqlPlan(String sqlCommands) throws Exception {
    final DrillbitContext dbContext = mock(DrillbitContext.class);
    final QueryContext context = mock(QueryContext.class);
    final String[] sqlStrings = sqlCommands.split(";");
    final LocalPersistentStoreProvider provider = new LocalPersistentStoreProvider(config);
    provider.start();
    final ScanResult scanResult = ClassPathScanner.fromPrescan(config);
    final LogicalPlanPersistence logicalPlanPersistence = new LogicalPlanPersistence(config, scanResult);
    final SystemOptionManager systemOptions = new SystemOptionManager(logicalPlanPersistence, provider, config);
    systemOptions.init();
    @SuppressWarnings("resource") final UserSession userSession = UserSession.Builder.newBuilder().withOptionManager(systemOptions).build();
    final SessionOptionManager sessionOptions = userSession.getOptions();
    final QueryOptionManager queryOptions = new QueryOptionManager(sessionOptions);
    final ExecutionControls executionControls = new ExecutionControls(queryOptions, DrillbitEndpoint.getDefaultInstance());
    when(dbContext.getMetrics()).thenReturn(new MetricRegistry());
    when(dbContext.getAllocator()).thenReturn(allocator);
    when(dbContext.getConfig()).thenReturn(config);
    when(dbContext.getOptionManager()).thenReturn(systemOptions);
    when(dbContext.getStoreProvider()).thenReturn(provider);
    when(dbContext.getClasspathScan()).thenReturn(scanResult);
    when(dbContext.getLpPersistence()).thenReturn(logicalPlanPersistence);
    final StoragePluginRegistry registry = new StoragePluginRegistryImpl(dbContext);
    registry.init();
    final FunctionImplementationRegistry functionRegistry = new FunctionImplementationRegistry(config);
    final DrillOperatorTable table = new DrillOperatorTable(functionRegistry, systemOptions);
    final SchemaPlus root = CalciteSchema.createRootSchema(false, false).plus();
    registry.getSchemaFactory().registerSchemas(SchemaConfig.newBuilder("foo", context).build(), root);
    when(context.getNewDefaultSchema()).thenReturn(root);
    when(context.getLpPersistence()).thenReturn(new LogicalPlanPersistence(config, ClassPathScanner.fromPrescan(config)));
    when(context.getStorage()).thenReturn(registry);
    when(context.getFunctionRegistry()).thenReturn(functionRegistry);
    when(context.getSession()).thenReturn(UserSession.Builder.newBuilder().withOptionManager(sessionOptions).setSupportComplexTypes(true).build());
    when(context.getCurrentEndpoint()).thenReturn(DrillbitEndpoint.getDefaultInstance());
    when(context.getActiveEndpoints()).thenReturn(ImmutableList.of(DrillbitEndpoint.getDefaultInstance()));
    when(context.getPlannerSettings()).thenReturn(new PlannerSettings(queryOptions, functionRegistry));
    when(context.getOptions()).thenReturn(queryOptions);
    when(context.getConfig()).thenReturn(config);
    when(context.getDrillOperatorTable()).thenReturn(table);
    when(context.getAllocator()).thenReturn(allocator);
    when(context.getExecutionControls()).thenReturn(executionControls);
    when(context.getLpPersistence()).thenReturn(logicalPlanPersistence);
    for (final String sql : sqlStrings) {
        if (sql.trim().isEmpty()) {
            continue;
        }
        @SuppressWarnings("unused") final PhysicalPlan p = DrillSqlWorker.getPlan(context, sql);
    }
}