Examples with LeftOuterJoinOperator org.apache.hyracks.algebricks.core.algebra.operators.logical.LeftOuterJoinOperator used on opensource projects

Example 1 with LeftOuterJoinOperator

use of org.apache.hyracks.algebricks.core.algebra.operators.logical.LeftOuterJoinOperator in project asterixdb by apache.

the class RemoveLeftOuterUnnestForLeftOuterJoinRule method rewritePre.

@Override
public boolean rewritePre(Mutable<ILogicalOperator> opRef, IOptimizationContext context) throws AlgebricksException {
    ILogicalOperator op1 = opRef.getValue();
    // Checks the plan pattern.
    if (!checkOperatorPattern(op1)) {
        return false;
    }
    LeftOuterUnnestOperator outerUnnest = (LeftOuterUnnestOperator) op1;
    GroupByOperator gbyOperator = (GroupByOperator) outerUnnest.getInputs().get(0).getValue();
    LeftOuterJoinOperator lojOperator = (LeftOuterJoinOperator) gbyOperator.getInputs().get(0).getValue();
    // Checks whether the left outer unnest and the group-by operator are qualified for rewriting.
    Triple<Boolean, ILogicalExpression, ILogicalExpression> checkGbyResult = checkUnnestAndGby(outerUnnest, gbyOperator);
    // The argument for listify and not(is-missing(...)) check should be variables.
    if (!isVariableReference(checkGbyResult.second) || !isVariableReference(checkGbyResult.third)) {
        return false;
    }
    // Checks whether both the listify variable and the condition test variable are from the right input
    // branch of the left outer join.
    LogicalVariable listifyVar = ((VariableReferenceExpression) checkGbyResult.second).getVariableReference();
    LogicalVariable conditionTestVar = ((VariableReferenceExpression) checkGbyResult.third).getVariableReference();
    if (!checkListifyAndConditionVar(lojOperator, listifyVar, conditionTestVar)) {
        return false;
    }
    // Does the rewrite.
    removeGroupByAndOuterUnnest(opRef, context, outerUnnest, gbyOperator, lojOperator, listifyVar);
    return true;
}

Example 2 with LeftOuterJoinOperator

use of org.apache.hyracks.algebricks.core.algebra.operators.logical.LeftOuterJoinOperator in project asterixdb by apache.

the class SetAsterixPhysicalOperatorsRule method computeDefaultPhysicalOp.

private static void computeDefaultPhysicalOp(AbstractLogicalOperator op, IOptimizationContext context) throws AlgebricksException {
    PhysicalOptimizationConfig physicalOptimizationConfig = context.getPhysicalOptimizationConfig();
    if (op.getOperatorTag().equals(LogicalOperatorTag.GROUP)) {
        GroupByOperator gby = (GroupByOperator) op;
        if (gby.getNestedPlans().size() == 1) {
            ILogicalPlan p0 = gby.getNestedPlans().get(0);
            if (p0.getRoots().size() == 1) {
                Mutable<ILogicalOperator> r0 = p0.getRoots().get(0);
                if (((AbstractLogicalOperator) (r0.getValue())).getOperatorTag().equals(LogicalOperatorTag.AGGREGATE)) {
                    AggregateOperator aggOp = (AggregateOperator) r0.getValue();
                    boolean serializable = true;
                    for (Mutable<ILogicalExpression> exprRef : aggOp.getExpressions()) {
                        AbstractFunctionCallExpression expr = (AbstractFunctionCallExpression) exprRef.getValue();
                        if (!BuiltinFunctions.isAggregateFunctionSerializable(expr.getFunctionIdentifier())) {
                            serializable = false;
                            break;
                        }
                    }
                    if ((gby.getAnnotations().get(OperatorAnnotations.USE_HASH_GROUP_BY) == Boolean.TRUE || gby.getAnnotations().get(OperatorAnnotations.USE_EXTERNAL_GROUP_BY) == Boolean.TRUE)) {
                        boolean setToExternalGby = false;
                        if (serializable) {
                            // if serializable, use external group-by
                            // now check whether the serialized version aggregation function has corresponding intermediate agg
                            boolean hasIntermediateAgg = true;
                            IMergeAggregationExpressionFactory mergeAggregationExpressionFactory = context.getMergeAggregationExpressionFactory();
                            List<LogicalVariable> originalVariables = aggOp.getVariables();
                            List<Mutable<ILogicalExpression>> aggExprs = aggOp.getExpressions();
                            int aggNum = aggExprs.size();
                            for (int i = 0; i < aggNum; i++) {
                                AbstractFunctionCallExpression expr = (AbstractFunctionCallExpression) aggExprs.get(i).getValue();
                                AggregateFunctionCallExpression serialAggExpr = BuiltinFunctions.makeSerializableAggregateFunctionExpression(expr.getFunctionIdentifier(), expr.getArguments());
                                if (mergeAggregationExpressionFactory.createMergeAggregation(originalVariables.get(i), serialAggExpr, context) == null) {
                                    hasIntermediateAgg = false;
                                    break;
                                }
                            }
                            // Check whether there are multiple aggregates in the sub plan.
                            // Currently, we don't support multiple aggregates in one external group-by.
                            boolean multipleAggOpsFound = false;
                            ILogicalOperator r1Logical = aggOp;
                            while (r1Logical.hasInputs()) {
                                r1Logical = r1Logical.getInputs().get(0).getValue();
                                if (r1Logical.getOperatorTag() == LogicalOperatorTag.AGGREGATE) {
                                    multipleAggOpsFound = true;
                                    break;
                                }
                            }
                            if (hasIntermediateAgg && !multipleAggOpsFound) {
                                for (int i = 0; i < aggNum; i++) {
                                    AbstractFunctionCallExpression expr = (AbstractFunctionCallExpression) aggExprs.get(i).getValue();
                                    AggregateFunctionCallExpression serialAggExpr = BuiltinFunctions.makeSerializableAggregateFunctionExpression(expr.getFunctionIdentifier(), expr.getArguments());
                                    aggOp.getExpressions().get(i).setValue(serialAggExpr);
                                }
                                ExternalGroupByPOperator externalGby = new ExternalGroupByPOperator(gby.getGroupByList(), physicalOptimizationConfig.getMaxFramesExternalGroupBy(), (long) physicalOptimizationConfig.getMaxFramesExternalGroupBy() * physicalOptimizationConfig.getFrameSize());
                                generateMergeAggregationExpressions(gby, context);
                                op.setPhysicalOperator(externalGby);
                                setToExternalGby = true;
                            }
                        }
                        if (!setToExternalGby) {
                            // if not serializable or no intermediate agg, use pre-clustered group-by
                            List<Pair<LogicalVariable, Mutable<ILogicalExpression>>> gbyList = gby.getGroupByList();
                            List<LogicalVariable> columnList = new ArrayList<LogicalVariable>(gbyList.size());
                            for (Pair<LogicalVariable, Mutable<ILogicalExpression>> p : gbyList) {
                                ILogicalExpression expr = p.second.getValue();
                                if (expr.getExpressionTag() == LogicalExpressionTag.VARIABLE) {
                                    VariableReferenceExpression varRef = (VariableReferenceExpression) expr;
                                    columnList.add(varRef.getVariableReference());
                                }
                            }
                            op.setPhysicalOperator(new PreclusteredGroupByPOperator(columnList, gby.isGroupAll()));
                        }
                    }
                } else if (((AbstractLogicalOperator) (r0.getValue())).getOperatorTag().equals(LogicalOperatorTag.RUNNINGAGGREGATE)) {
                    List<Pair<LogicalVariable, Mutable<ILogicalExpression>>> gbyList = gby.getGroupByList();
                    List<LogicalVariable> columnList = new ArrayList<LogicalVariable>(gbyList.size());
                    for (Pair<LogicalVariable, Mutable<ILogicalExpression>> p : gbyList) {
                        ILogicalExpression expr = p.second.getValue();
                        if (expr.getExpressionTag() == LogicalExpressionTag.VARIABLE) {
                            VariableReferenceExpression varRef = (VariableReferenceExpression) expr;
                            columnList.add(varRef.getVariableReference());
                        }
                    }
                    op.setPhysicalOperator(new PreclusteredGroupByPOperator(columnList, gby.isGroupAll()));
                } else {
                    throw new AlgebricksException("Unsupported nested operator within a group-by: " + ((AbstractLogicalOperator) (r0.getValue())).getOperatorTag().name());
                }
            }
        }
    }
    if (op.getPhysicalOperator() == null) {
        switch(op.getOperatorTag()) {
            case INNERJOIN:
                {
                    JoinUtils.setJoinAlgorithmAndExchangeAlgo((InnerJoinOperator) op, context);
                    break;
                }
            case LEFTOUTERJOIN:
                {
                    JoinUtils.setJoinAlgorithmAndExchangeAlgo((LeftOuterJoinOperator) op, context);
                    break;
                }
            case UNNEST_MAP:
            case LEFT_OUTER_UNNEST_MAP:
                {
                    ILogicalExpression unnestExpr = null;
                    unnestExpr = ((AbstractUnnestMapOperator) op).getExpressionRef().getValue();
                    if (unnestExpr.getExpressionTag() == LogicalExpressionTag.FUNCTION_CALL) {
                        AbstractFunctionCallExpression f = (AbstractFunctionCallExpression) unnestExpr;
                        FunctionIdentifier fid = f.getFunctionIdentifier();
                        if (!fid.equals(BuiltinFunctions.INDEX_SEARCH)) {
                            throw new IllegalStateException();
                        }
                        AccessMethodJobGenParams jobGenParams = new AccessMethodJobGenParams();
                        jobGenParams.readFromFuncArgs(f.getArguments());
                        MetadataProvider mp = (MetadataProvider) context.getMetadataProvider();
                        DataSourceId dataSourceId = new DataSourceId(jobGenParams.getDataverseName(), jobGenParams.getDatasetName());
                        Dataset dataset = mp.findDataset(jobGenParams.getDataverseName(), jobGenParams.getDatasetName());
                        IDataSourceIndex<String, DataSourceId> dsi = mp.findDataSourceIndex(jobGenParams.getIndexName(), dataSourceId);
                        INodeDomain storageDomain = mp.findNodeDomain(dataset.getNodeGroupName());
                        if (dsi == null) {
                            throw new AlgebricksException("Could not find index " + jobGenParams.getIndexName() + " for dataset " + dataSourceId);
                        }
                        IndexType indexType = jobGenParams.getIndexType();
                        boolean requiresBroadcast = jobGenParams.getRequiresBroadcast();
                        switch(indexType) {
                            case BTREE:
                                {
                                    BTreeJobGenParams btreeJobGenParams = new BTreeJobGenParams();
                                    btreeJobGenParams.readFromFuncArgs(f.getArguments());
                                    op.setPhysicalOperator(new BTreeSearchPOperator(dsi, storageDomain, requiresBroadcast, btreeJobGenParams.isPrimaryIndex(), btreeJobGenParams.isEqCondition(), btreeJobGenParams.getLowKeyVarList(), btreeJobGenParams.getHighKeyVarList()));
                                    break;
                                }
                            case RTREE:
                                {
                                    op.setPhysicalOperator(new RTreeSearchPOperator(dsi, storageDomain, requiresBroadcast));
                                    break;
                                }
                            case SINGLE_PARTITION_WORD_INVIX:
                            case SINGLE_PARTITION_NGRAM_INVIX:
                                {
                                    op.setPhysicalOperator(new InvertedIndexPOperator(dsi, storageDomain, requiresBroadcast, false));
                                    break;
                                }
                            case LENGTH_PARTITIONED_WORD_INVIX:
                            case LENGTH_PARTITIONED_NGRAM_INVIX:
                                {
                                    op.setPhysicalOperator(new InvertedIndexPOperator(dsi, storageDomain, requiresBroadcast, true));
                                    break;
                                }
                            default:
                                {
                                    throw new NotImplementedException(indexType + " indexes are not implemented.");
                                }
                        }
                    }
                    break;
                }
        }
    }
    if (op.hasNestedPlans()) {
        AbstractOperatorWithNestedPlans nested = (AbstractOperatorWithNestedPlans) op;
        for (ILogicalPlan p : nested.getNestedPlans()) {
            setPhysicalOperators(p, context);
        }
    }
    for (Mutable<ILogicalOperator> opRef : op.getInputs()) {
        computeDefaultPhysicalOp((AbstractLogicalOperator) opRef.getValue(), context);
    }
}

Example 3 with LeftOuterJoinOperator

use of org.apache.hyracks.algebricks.core.algebra.operators.logical.LeftOuterJoinOperator in project asterixdb by apache.

the class IntroduceJoinAccessMethodRule method checkAndApplyJoinTransformation.

/**
     * Recursively traverse the given plan and check whether a INNERJOIN or LEFTOUTERJOIN operator exists.
     * If one is found, maintain the path from the root to the given join operator and
     * optimize the path from the given join operator to the EMPTY_TUPLE_SOURCE operator
     * if it is not already optimized.
     */
protected boolean checkAndApplyJoinTransformation(Mutable<ILogicalOperator> opRef, IOptimizationContext context) throws AlgebricksException {
    AbstractLogicalOperator op = (AbstractLogicalOperator) opRef.getValue();
    boolean joinFoundAndOptimizationApplied;
    // Check the current operator pattern to see whether it is a JOIN or not.
    boolean isThisOpInnerJoin = isInnerJoin(op);
    boolean isThisOpLeftOuterJoin = isLeftOuterJoin(op);
    boolean isParentOpGroupBy = hasGroupBy;
    Mutable<ILogicalOperator> joinRefFromThisOp = null;
    AbstractBinaryJoinOperator joinOpFromThisOp = null;
    if (isThisOpInnerJoin) {
        // Set join operator.
        joinRef = opRef;
        joinOp = (InnerJoinOperator) op;
        joinRefFromThisOp = opRef;
        joinOpFromThisOp = (InnerJoinOperator) op;
    } else if (isThisOpLeftOuterJoin) {
        // Set left-outer-join op.
        // The current operator is GROUP and the child of this op is LEFTOUERJOIN.
        joinRef = op.getInputs().get(0);
        joinOp = (LeftOuterJoinOperator) joinRef.getValue();
        joinRefFromThisOp = op.getInputs().get(0);
        joinOpFromThisOp = (LeftOuterJoinOperator) joinRefFromThisOp.getValue();
    }
    // to make sure an earlier join in the path is optimized first.
    for (Mutable<ILogicalOperator> inputOpRef : op.getInputs()) {
        joinFoundAndOptimizationApplied = checkAndApplyJoinTransformation(inputOpRef, context);
        if (joinFoundAndOptimizationApplied) {
            return true;
        }
    }
    // For a JOIN case, try to transform the given plan.
    if (isThisOpInnerJoin || isThisOpLeftOuterJoin) {
        // Restore the information from this operator since it might have been be set to null
        // if there are other join operators in the earlier path.
        joinRef = joinRefFromThisOp;
        joinOp = joinOpFromThisOp;
        boolean continueCheck = true;
        // Already checked? If not, this operator may be optimized.
        if (context.checkIfInDontApplySet(this, joinOp)) {
            continueCheck = false;
        }
        // For each access method, this contains the information about
        // whether an available index can be applicable or not.
        Map<IAccessMethod, AccessMethodAnalysisContext> analyzedAMs = null;
        if (continueCheck) {
            analyzedAMs = new HashMap<>();
        }
        // whether the given plan is truly optimizable or not.
        if (continueCheck && !checkJoinOpConditionAndInitSubTree(context)) {
            continueCheck = false;
        }
        // Analyze the condition of SELECT operator and initialize analyzedAMs.
        // Check whether the function in the SELECT operator can be truly transformed.
        boolean matchInLeftSubTree = false;
        boolean matchInRightSubTree = false;
        if (continueCheck) {
            if (leftSubTree.hasDataSource()) {
                matchInLeftSubTree = analyzeSelectOrJoinOpConditionAndUpdateAnalyzedAM(joinCond, leftSubTree.getAssignsAndUnnests(), analyzedAMs, context, typeEnvironment);
            }
            if (rightSubTree.hasDataSource()) {
                matchInRightSubTree = analyzeSelectOrJoinOpConditionAndUpdateAnalyzedAM(joinCond, rightSubTree.getAssignsAndUnnests(), analyzedAMs, context, typeEnvironment);
            }
        }
        // Find the dataset from the data-source and the record type of the dataset from the metadata.
        // This will be used to find an applicable index on the dataset.
        boolean checkLeftSubTreeMetadata = false;
        boolean checkRightSubTreeMetadata = false;
        if (continueCheck && (matchInLeftSubTree || matchInRightSubTree)) {
            // Set dataset and type metadata.
            if (matchInLeftSubTree) {
                checkLeftSubTreeMetadata = leftSubTree.setDatasetAndTypeMetadata(metadataProvider);
            }
            if (matchInRightSubTree) {
                checkRightSubTreeMetadata = rightSubTree.setDatasetAndTypeMetadata(metadataProvider);
            }
        }
        if (continueCheck && (checkLeftSubTreeMetadata || checkRightSubTreeMetadata)) {
            // Then find the applicable indexes for the variables used in the JOIN condition.
            if (checkLeftSubTreeMetadata) {
                fillSubTreeIndexExprs(leftSubTree, analyzedAMs, context);
            }
            if (checkRightSubTreeMetadata) {
                fillSubTreeIndexExprs(rightSubTree, analyzedAMs, context);
            }
            // Prune the access methods based on the function expression and access methods.
            pruneIndexCandidates(analyzedAMs, context, typeEnvironment);
            // If the right subtree (inner branch) has indexes, one of those indexes will be used.
            // Remove the indexes from the outer branch in the optimizer's consideration list for this rule.
            pruneIndexCandidatesFromOuterBranch(analyzedAMs);
            // We are going to use indexes from the inner branch.
            // If no index is available, then we stop here.
            Pair<IAccessMethod, Index> chosenIndex = chooseBestIndex(analyzedAMs);
            if (chosenIndex == null) {
                context.addToDontApplySet(this, joinOp);
                continueCheck = false;
            }
            if (continueCheck) {
                // Apply plan transformation using chosen index.
                AccessMethodAnalysisContext analysisCtx = analyzedAMs.get(chosenIndex.first);
                // in GroupByOp.
                if (isThisOpLeftOuterJoin && isParentOpGroupBy) {
                    analysisCtx.setLOJGroupbyOpRef(opRef);
                    ScalarFunctionCallExpression isNullFuncExpr = AccessMethodUtils.findLOJIsMissingFuncInGroupBy((GroupByOperator) opRef.getValue());
                    analysisCtx.setLOJIsNullFuncInGroupBy(isNullFuncExpr);
                }
                Dataset indexDataset = analysisCtx.getDatasetFromIndexDatasetMap(chosenIndex.second);
                // from the right subtree. The following is just a sanity check.
                if (!rightSubTree.hasDataSourceScan() && !indexDataset.getDatasetName().equals(rightSubTree.getDataset().getDatasetName())) {
                    return false;
                }
                // Finally, try to apply plan transformation using chosen index.
                boolean res = chosenIndex.first.applyJoinPlanTransformation(joinRef, leftSubTree, rightSubTree, chosenIndex.second, analysisCtx, context, isThisOpLeftOuterJoin, isParentOpGroupBy);
                // will find them.
                if (res) {
                    return res;
                }
            }
        }
        joinRef = null;
        joinOp = null;
    }
    return false;
}

Example 4 with LeftOuterJoinOperator

use of org.apache.hyracks.algebricks.core.algebra.operators.logical.LeftOuterJoinOperator in project asterixdb by apache.

the class IsomorphismOperatorVisitor method visitLeftOuterJoinOperator.

@Override
public Boolean visitLeftOuterJoinOperator(LeftOuterJoinOperator op, ILogicalOperator arg) throws AlgebricksException {
    AbstractLogicalOperator aop = (AbstractLogicalOperator) arg;
    if (aop.getOperatorTag() != LogicalOperatorTag.LEFTOUTERJOIN) {
        return Boolean.FALSE;
    }
    LeftOuterJoinOperator joinOpArg = (LeftOuterJoinOperator) copyAndSubstituteVar(op, arg);
    boolean isomorphic = op.getCondition().getValue().equals(joinOpArg.getCondition().getValue());
    return isomorphic;
}

Example 5 with LeftOuterJoinOperator

use of org.apache.hyracks.algebricks.core.algebra.operators.logical.LeftOuterJoinOperator in project asterixdb by apache.

the class IntroduceLeftOuterJoinForSubplanRule method rewritePost.

@Override
public boolean rewritePost(Mutable<ILogicalOperator> opRef, IOptimizationContext context) throws AlgebricksException {
    AbstractLogicalOperator op0 = (AbstractLogicalOperator) opRef.getValue();
    if (op0.getOperatorTag() != LogicalOperatorTag.SUBPLAN) {
        return false;
    }
    SubplanOperator subplan = (SubplanOperator) op0;
    Iterator<ILogicalPlan> plansIter = subplan.getNestedPlans().iterator();
    ILogicalPlan p = null;
    while (plansIter.hasNext()) {
        p = plansIter.next();
    }
    if (p == null) {
        return false;
    }
    if (p.getRoots().size() != 1) {
        return false;
    }
    Mutable<ILogicalOperator> subplanRoot = p.getRoots().get(0);
    AbstractLogicalOperator op1 = (AbstractLogicalOperator) subplanRoot.getValue();
    Mutable<ILogicalOperator> opUnder = subplan.getInputs().get(0);
    if (OperatorPropertiesUtil.isMissingTest((AbstractLogicalOperator) opUnder.getValue())) {
        return false;
    }
    switch(op1.getOperatorTag()) {
        case INNERJOIN:
            {
                InnerJoinOperator join = (InnerJoinOperator) op1;
                Mutable<ILogicalOperator> leftRef = join.getInputs().get(0);
                Mutable<ILogicalOperator> rightRef = join.getInputs().get(1);
                Mutable<ILogicalOperator> ntsRef = getNtsAtEndOfPipeline(leftRef);
                if (ntsRef == null) {
                    ntsRef = getNtsAtEndOfPipeline(rightRef);
                    if (ntsRef == null) {
                        return false;
                    } else {
                        Mutable<ILogicalOperator> t = leftRef;
                        leftRef = rightRef;
                        rightRef = t;
                    }
                }
                ntsRef.setValue(opUnder.getValue());
                LeftOuterJoinOperator loj = new LeftOuterJoinOperator(join.getCondition());
                loj.getInputs().add(leftRef);
                loj.getInputs().add(rightRef);
                opRef.setValue(loj);
                context.computeAndSetTypeEnvironmentForOperator(loj);
                return true;
            }
        case LEFTOUTERJOIN:
            {
                LeftOuterJoinOperator join = (LeftOuterJoinOperator) op1;
                Mutable<ILogicalOperator> leftRef = join.getInputs().get(0);
                Mutable<ILogicalOperator> ntsRef = getNtsAtEndOfPipeline(leftRef);
                if (ntsRef == null) {
                    return false;
                }
                ntsRef.setValue(opUnder.getValue());
                opRef.setValue(join);
                context.computeAndSetTypeEnvironmentForOperator(join);
                return true;
            }
        default:
            {
                return false;
            }
    }
}