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

Example 11 with AggregateOperator

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

the class SetAsterixPhysicalOperatorsRule method generateMergeAggregationExpressions.

private static void generateMergeAggregationExpressions(GroupByOperator gby, IOptimizationContext context) throws AlgebricksException {
    if (gby.getNestedPlans().size() != 1) {
        throw new AlgebricksException("External group-by currently works only for one nested plan with one root containing" + "an aggregate and a nested-tuple-source.");
    }
    ILogicalPlan p0 = gby.getNestedPlans().get(0);
    if (p0.getRoots().size() != 1) {
        throw new AlgebricksException("External group-by currently works only for one nested plan with one root containing" + "an aggregate and a nested-tuple-source.");
    }
    IMergeAggregationExpressionFactory mergeAggregationExpressionFactory = context.getMergeAggregationExpressionFactory();
    Mutable<ILogicalOperator> r0 = p0.getRoots().get(0);
    AbstractLogicalOperator r0Logical = (AbstractLogicalOperator) r0.getValue();
    if (r0Logical.getOperatorTag() != LogicalOperatorTag.AGGREGATE) {
        throw new AlgebricksException("The merge aggregation expression generation should not process a " + r0Logical.getOperatorTag() + " operator.");
    }
    AggregateOperator aggOp = (AggregateOperator) r0.getValue();
    List<Mutable<ILogicalExpression>> aggFuncRefs = aggOp.getExpressions();
    List<LogicalVariable> aggProducedVars = aggOp.getVariables();
    int n = aggOp.getExpressions().size();
    List<Mutable<ILogicalExpression>> mergeExpressionRefs = new ArrayList<Mutable<ILogicalExpression>>();
    for (int i = 0; i < n; i++) {
        ILogicalExpression mergeExpr = mergeAggregationExpressionFactory.createMergeAggregation(aggProducedVars.get(i), aggFuncRefs.get(i).getValue(), context);
        if (mergeExpr == null) {
            throw new AlgebricksException("The aggregation function " + aggFuncRefs.get(i).getValue() + " does not have a registered intermediate aggregation function.");
        }
        mergeExpressionRefs.add(new MutableObject<ILogicalExpression>(mergeExpr));
    }
    aggOp.setMergeExpressions(mergeExpressionRefs);
}

Example 12 with AggregateOperator

use of org.apache.hyracks.algebricks.core.algebra.operators.logical.AggregateOperator 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 13 with AggregateOperator

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

the class RemoveUnusedAssignAndAggregateRule method removeFromAssigns.

private int removeFromAssigns(AbstractLogicalOperator op, Set<LogicalVariable> toRemove, IOptimizationContext context) throws AlgebricksException {
    switch(op.getOperatorTag()) {
        case ASSIGN:
            AssignOperator assign = (AssignOperator) op;
            if (removeUnusedVarsAndExprs(toRemove, assign.getVariables(), assign.getExpressions())) {
                context.computeAndSetTypeEnvironmentForOperator(assign);
                isTransformed = true;
            }
            return assign.getVariables().size();
        case AGGREGATE:
            AggregateOperator agg = (AggregateOperator) op;
            if (removeUnusedVarsAndExprs(toRemove, agg.getVariables(), agg.getExpressions())) {
                context.computeAndSetTypeEnvironmentForOperator(agg);
                isTransformed = true;
            }
            return agg.getVariables().size();
        case UNNEST:
            UnnestOperator uOp = (UnnestOperator) op;
            LogicalVariable pVar = uOp.getPositionalVariable();
            if (pVar != null && toRemove != null && toRemove.contains(pVar)) {
                uOp.setPositionalVariable(null);
                assignedVarSet.remove(pVar);
                isTransformed = true;
            }
            break;
        case UNIONALL:
            UnionAllOperator unionOp = (UnionAllOperator) op;
            if (removeUnusedVarsFromUnionAll(unionOp, toRemove)) {
                context.computeAndSetTypeEnvironmentForOperator(unionOp);
                isTransformed = true;
            }
            return unionOp.getVariableMappings().size();
        case GROUP:
            GroupByOperator groupByOp = (GroupByOperator) op;
            if (removeUnusedVarsFromGroupBy(groupByOp, toRemove)) {
                context.computeAndSetTypeEnvironmentForOperator(groupByOp);
                isTransformed = true;
            }
            return groupByOp.getGroupByList().size() + groupByOp.getNestedPlans().size() + groupByOp.getDecorList().size();
        default:
            break;
    }
    return -1;
}

Example 14 with AggregateOperator

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

the class CountVarToCountOneRule method rewritePost.

// It is only for a group-by having just one aggregate which is a count.
@Override
public boolean rewritePost(Mutable<ILogicalOperator> opRef, IOptimizationContext context) throws AlgebricksException {
    AbstractLogicalOperator op1 = (AbstractLogicalOperator) opRef.getValue();
    if (op1.getOperatorTag() != LogicalOperatorTag.GROUP) {
        return false;
    }
    GroupByOperator g = (GroupByOperator) op1;
    if (g.getNestedPlans().size() != 1) {
        return false;
    }
    ILogicalPlan p = g.getNestedPlans().get(0);
    if (p.getRoots().size() != 1) {
        return false;
    }
    AbstractLogicalOperator op2 = (AbstractLogicalOperator) p.getRoots().get(0).getValue();
    if (op2.getOperatorTag() != LogicalOperatorTag.AGGREGATE) {
        return false;
    }
    AggregateOperator agg = (AggregateOperator) op2;
    if (agg.getExpressions().size() != 1) {
        return false;
    }
    ILogicalExpression exp2 = agg.getExpressions().get(0).getValue();
    if (exp2.getExpressionTag() != LogicalExpressionTag.FUNCTION_CALL) {
        return false;
    }
    AbstractFunctionCallExpression fun = (AbstractFunctionCallExpression) exp2;
    if (fun.getFunctionIdentifier() != BuiltinFunctions.COUNT) {
        return false;
    }
    ILogicalExpression exp3 = fun.getArguments().get(0).getValue();
    if (exp3.getExpressionTag() != LogicalExpressionTag.VARIABLE) {
        return false;
    }
    if (((AbstractLogicalOperator) agg.getInputs().get(0).getValue()).getOperatorTag() != LogicalOperatorTag.NESTEDTUPLESOURCE) {
        return false;
    }
    fun.getArguments().get(0).setValue(new ConstantExpression(new AsterixConstantValue(new AInt64(1L))));
    return true;
}

Example 15 with AggregateOperator

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

the class PushAggregateIntoNestedSubplanRule method removeRedundantListifies.

private void removeRedundantListifies(Map<LogicalVariable, Integer> nspAggVars, Map<LogicalVariable, AbstractOperatorWithNestedPlans> nspWithAgg, Map<LogicalVariable, Integer> nspAggVarToPlanIndex) throws AlgebricksException {
    List<Pair<AbstractOperatorWithNestedPlans, Integer>> removeList = new ArrayList<>();
    for (Map.Entry<LogicalVariable, Integer> aggVarEntry : nspAggVars.entrySet()) {
        LogicalVariable aggVar = aggVarEntry.getKey();
        int occurs = aggVarEntry.getValue();
        if (occurs == 0) {
            AbstractOperatorWithNestedPlans nspOp = nspWithAgg.get(aggVar);
            AggregateOperator aggOp = (AggregateOperator) nspOp.getNestedPlans().get(nspAggVarToPlanIndex.get(aggVar)).getRoots().get(0).getValue();
            int pos = aggOp.getVariables().indexOf(aggVar);
            if (pos >= 0) {
                aggOp.getVariables().remove(pos);
                aggOp.getExpressions().remove(pos);
                List<LogicalVariable> producedVarsAtAgg = new ArrayList<>();
                VariableUtilities.getProducedVariablesInDescendantsAndSelf(aggOp, producedVarsAtAgg);
                if (producedVarsAtAgg.isEmpty()) {
                    removeList.add(new Pair<>(nspOp, nspAggVarToPlanIndex.get(aggVar)));
                }
            }
        }
    }
    // Collects subplans that is to be removed.
    Map<AbstractOperatorWithNestedPlans, List<ILogicalPlan>> nspToSubplanListMap = new HashMap<>();
    for (Pair<AbstractOperatorWithNestedPlans, Integer> remove : removeList) {
        AbstractOperatorWithNestedPlans groupByOperator = remove.first;
        ILogicalPlan subplan = remove.first.getNestedPlans().get(remove.second);
        if (nspToSubplanListMap.containsKey(groupByOperator)) {
            List<ILogicalPlan> subplans = nspToSubplanListMap.get(groupByOperator);
            subplans.add(subplan);
        } else {
            List<ILogicalPlan> subplans = new ArrayList<>();
            subplans.add(subplan);
            nspToSubplanListMap.put(groupByOperator, subplans);
        }
    }
    // Removes subplans.
    for (Map.Entry<AbstractOperatorWithNestedPlans, List<ILogicalPlan>> entry : nspToSubplanListMap.entrySet()) {
        entry.getKey().getNestedPlans().removeAll(entry.getValue());
    }
}