Examples with DataSource org.apache.asterix.metadata.declared.DataSource used on opensource projects

Example 1 with DataSource

use of org.apache.asterix.metadata.declared.DataSource in project asterixdb by apache.

the class IntroduceSecondaryIndexInsertDeleteRule method rewritePost.

@Override
public boolean rewritePost(Mutable<ILogicalOperator> opRef, IOptimizationContext context) throws AlgebricksException {
    AbstractLogicalOperator op0 = (AbstractLogicalOperator) opRef.getValue();
    if (op0.getOperatorTag() != LogicalOperatorTag.DELEGATE_OPERATOR && op0.getOperatorTag() != LogicalOperatorTag.SINK) {
        return false;
    }
    if (op0.getOperatorTag() == LogicalOperatorTag.DELEGATE_OPERATOR) {
        DelegateOperator eOp = (DelegateOperator) op0;
        if (!(eOp.getDelegate() instanceof CommitOperator)) {
            return false;
        }
    }
    AbstractLogicalOperator op1 = (AbstractLogicalOperator) op0.getInputs().get(0).getValue();
    if (op1.getOperatorTag() != LogicalOperatorTag.INSERT_DELETE_UPSERT) {
        return false;
    }
    /** find the record variable */
    InsertDeleteUpsertOperator primaryIndexModificationOp = (InsertDeleteUpsertOperator) op0.getInputs().get(0).getValue();
    boolean isBulkload = primaryIndexModificationOp.isBulkload();
    ILogicalExpression newRecordExpr = primaryIndexModificationOp.getPayloadExpression().getValue();
    List<Mutable<ILogicalExpression>> newMetaExprs = primaryIndexModificationOp.getAdditionalNonFilteringExpressions();
    LogicalVariable newRecordVar;
    LogicalVariable newMetaVar = null;
    /**
         * inputOp is the assign operator which extracts primary keys from the input
         * variables (record or meta)
         */
    AbstractLogicalOperator inputOp = (AbstractLogicalOperator) primaryIndexModificationOp.getInputs().get(0).getValue();
    newRecordVar = getRecordVar(context, inputOp, newRecordExpr, 0);
    if (newMetaExprs != null && !newMetaExprs.isEmpty()) {
        if (newMetaExprs.size() > 1) {
            throw new AlgebricksException("Number of meta records can't be more than 1. Number of meta records found = " + newMetaExprs.size());
        }
        newMetaVar = getRecordVar(context, inputOp, newMetaExprs.get(0).getValue(), 1);
    }
    /*
         * At this point, we have the record variable and the insert/delete/upsert operator
         * Note: We have two operators:
         * 1. An InsertDeleteOperator (primary)
         * 2. An IndexInsertDeleteOperator (secondary)
         * The current primaryIndexModificationOp is of the first type
         */
    DataSource datasetSource = (DataSource) primaryIndexModificationOp.getDataSource();
    MetadataProvider mp = (MetadataProvider) context.getMetadataProvider();
    String dataverseName = datasetSource.getId().getDataverseName();
    String datasetName = datasetSource.getId().getDatasourceName();
    Dataset dataset = mp.findDataset(dataverseName, datasetName);
    if (dataset == null) {
        throw new AlgebricksException("Unknown dataset " + datasetName + " in dataverse " + dataverseName);
    }
    if (dataset.getDatasetType() == DatasetType.EXTERNAL) {
        return false;
    }
    // Create operators for secondary index insert / delete.
    String itemTypeName = dataset.getItemTypeName();
    IAType itemType = mp.findType(dataset.getItemTypeDataverseName(), itemTypeName);
    if (itemType.getTypeTag() != ATypeTag.OBJECT) {
        throw new AlgebricksException("Only record types can be indexed.");
    }
    ARecordType recType = (ARecordType) itemType;
    // meta type
    ARecordType metaType = null;
    if (dataset.hasMetaPart()) {
        metaType = (ARecordType) mp.findType(dataset.getMetaItemTypeDataverseName(), dataset.getMetaItemTypeName());
    }
    List<Index> indexes = mp.getDatasetIndexes(dataset.getDataverseName(), dataset.getDatasetName());
    // Set the top operator pointer to the primary IndexInsertDeleteOperator
    ILogicalOperator currentTop = primaryIndexModificationOp;
    boolean hasSecondaryIndex = false;
    // Put an n-gram or a keyword index in the later stage of index-update,
    // since TokenizeOperator needs to be involved.
    Collections.sort(indexes, (o1, o2) -> o1.getIndexType().ordinal() - o2.getIndexType().ordinal());
    // At this point, we have the data type info, and the indexes info as well
    int secondaryIndexTotalCnt = indexes.size() - 1;
    if (secondaryIndexTotalCnt > 0) {
        op0.getInputs().clear();
    } else {
        return false;
    }
    // Initialize inputs to the SINK operator Op0 (The SINK) is now without input
    // Prepare filtering field information (This is the filter created using the "filter with" key word in the
    // create dataset ddl)
    List<String> filteringFields = ((InternalDatasetDetails) dataset.getDatasetDetails()).getFilterField();
    List<LogicalVariable> filteringVars;
    List<Mutable<ILogicalExpression>> filteringExpressions = null;
    if (filteringFields != null) {
        // The filter field var already exists. we can simply get it from the insert op
        filteringVars = new ArrayList<>();
        filteringExpressions = new ArrayList<>();
        for (Mutable<ILogicalExpression> filteringExpression : primaryIndexModificationOp.getAdditionalFilteringExpressions()) {
            filteringExpression.getValue().getUsedVariables(filteringVars);
            for (LogicalVariable var : filteringVars) {
                filteringExpressions.add(new MutableObject<ILogicalExpression>(new VariableReferenceExpression(var)));
            }
        }
    }
    // Replicate Operator is applied only when doing the bulk-load.
    ReplicateOperator replicateOp = null;
    if (secondaryIndexTotalCnt > 1 && primaryIndexModificationOp.isBulkload()) {
        // Split the logical plan into "each secondary index update branch"
        // to replicate each <PK,OBJECT> pair.
        replicateOp = new ReplicateOperator(secondaryIndexTotalCnt);
        replicateOp.getInputs().add(new MutableObject<ILogicalOperator>(currentTop));
        replicateOp.setExecutionMode(ExecutionMode.PARTITIONED);
        context.computeAndSetTypeEnvironmentForOperator(replicateOp);
        currentTop = replicateOp;
    }
    /*
         * The two maps are used to store variables to which [casted] field access is assigned.
         * One for the beforeOp record and the other for the new record.
         * There are two uses for these maps:
         * 1. used for shared fields in indexes with overlapping keys.
         * 2. used for setting variables of secondary keys for each secondary index operator.
         */
    Map<IndexFieldId, LogicalVariable> fieldVarsForBeforeOperation = new HashMap<>();
    Map<IndexFieldId, LogicalVariable> fieldVarsForNewRecord = new HashMap<>();
    /*
         * if the index is enforcing field types (For open indexes), We add a cast
         * operator to ensure type safety
         */
    try {
        if (primaryIndexModificationOp.getOperation() == Kind.INSERT || primaryIndexModificationOp.getOperation() == Kind.UPSERT || /* Actually, delete should not be here but it is now until issue
                     * https://issues.apache.org/jira/browse/ASTERIXDB-1507
                     * is solved
                     */
        primaryIndexModificationOp.getOperation() == Kind.DELETE) {
            injectFieldAccessesForIndexes(context, dataset, indexes, fieldVarsForNewRecord, recType, metaType, newRecordVar, newMetaVar, primaryIndexModificationOp, false);
            if (replicateOp != null) {
                context.computeAndSetTypeEnvironmentForOperator(replicateOp);
            }
        }
        if (primaryIndexModificationOp.getOperation() == Kind.UPSERT) /* Actually, delete should be here but it is not until issue
             * https://issues.apache.org/jira/browse/ASTERIXDB-1507
             * is solved
             */
        {
            List<LogicalVariable> beforeOpMetaVars = primaryIndexModificationOp.getBeforeOpAdditionalNonFilteringVars();
            LogicalVariable beforeOpMetaVar = beforeOpMetaVars == null ? null : beforeOpMetaVars.get(0);
            currentTop = injectFieldAccessesForIndexes(context, dataset, indexes, fieldVarsForBeforeOperation, recType, metaType, primaryIndexModificationOp.getBeforeOpRecordVar(), beforeOpMetaVar, currentTop, true);
        }
    } catch (AsterixException e) {
        throw new AlgebricksException(e);
    }
    // At first, op1 is the index insert op insertOp
    for (Index index : indexes) {
        if (!index.isSecondaryIndex()) {
            continue;
        }
        hasSecondaryIndex = true;
        // Get the secondary fields names and types
        List<List<String>> secondaryKeyFields = index.getKeyFieldNames();
        List<LogicalVariable> secondaryKeyVars = new ArrayList<>();
        List<Mutable<ILogicalExpression>> secondaryExpressions = new ArrayList<>();
        List<Mutable<ILogicalExpression>> beforeOpSecondaryExpressions = new ArrayList<>();
        ILogicalOperator replicateOutput;
        for (int i = 0; i < secondaryKeyFields.size(); i++) {
            IndexFieldId indexFieldId = new IndexFieldId(index.getKeyFieldSourceIndicators().get(i), secondaryKeyFields.get(i));
            LogicalVariable skVar = fieldVarsForNewRecord.get(indexFieldId);
            secondaryKeyVars.add(skVar);
            secondaryExpressions.add(new MutableObject<ILogicalExpression>(new VariableReferenceExpression(skVar)));
            if (primaryIndexModificationOp.getOperation() == Kind.UPSERT) {
                beforeOpSecondaryExpressions.add(new MutableObject<ILogicalExpression>(new VariableReferenceExpression(fieldVarsForBeforeOperation.get(indexFieldId))));
            }
        }
        IndexInsertDeleteUpsertOperator indexUpdate;
        if (index.getIndexType() != IndexType.RTREE) {
            // Create an expression per key
            Mutable<ILogicalExpression> filterExpression = (primaryIndexModificationOp.getOperation() == Kind.UPSERT) ? null : createFilterExpression(secondaryKeyVars, context.getOutputTypeEnvironment(currentTop), index.isEnforcingKeyFileds());
            DataSourceIndex dataSourceIndex = new DataSourceIndex(index, dataverseName, datasetName, mp);
            // and index type is keyword or n-gram.
            if (index.getIndexType() != IndexType.BTREE && primaryIndexModificationOp.isBulkload()) {
                // Note: Bulk load case, we don't need to take care of it for upsert operation
                // Check whether the index is length-partitioned or not.
                // If partitioned, [input variables to TokenizeOperator,
                // token, number of token] pairs will be generated and
                // fed into the IndexInsertDeleteOperator.
                // If not, [input variables, token] pairs will be generated
                // and fed into the IndexInsertDeleteOperator.
                // Input variables are passed since TokenizeOperator is not an
                // filtering operator.
                boolean isPartitioned = index.getIndexType() == IndexType.LENGTH_PARTITIONED_WORD_INVIX || index.getIndexType() == IndexType.LENGTH_PARTITIONED_NGRAM_INVIX;
                // Create a new logical variable - token
                List<LogicalVariable> tokenizeKeyVars = new ArrayList<>();
                List<Mutable<ILogicalExpression>> tokenizeKeyExprs = new ArrayList<>();
                LogicalVariable tokenVar = context.newVar();
                tokenizeKeyVars.add(tokenVar);
                tokenizeKeyExprs.add(new MutableObject<ILogicalExpression>(new VariableReferenceExpression(tokenVar)));
                // Check the field type of the secondary key.
                IAType secondaryKeyType;
                Pair<IAType, Boolean> keyPairType = Index.getNonNullableOpenFieldType(index.getKeyFieldTypes().get(0), secondaryKeyFields.get(0), recType);
                secondaryKeyType = keyPairType.first;
                List<Object> varTypes = new ArrayList<>();
                varTypes.add(NonTaggedFormatUtil.getTokenType(secondaryKeyType));
                // The type is short, and this does not contain type info.
                if (isPartitioned) {
                    LogicalVariable lengthVar = context.newVar();
                    tokenizeKeyVars.add(lengthVar);
                    tokenizeKeyExprs.add(new MutableObject<ILogicalExpression>(new VariableReferenceExpression(lengthVar)));
                    varTypes.add(BuiltinType.SHORTWITHOUTTYPEINFO);
                }
                // TokenizeOperator to tokenize [SK, PK] pairs
                TokenizeOperator tokenUpdate = new TokenizeOperator(dataSourceIndex, primaryIndexModificationOp.getPrimaryKeyExpressions(), secondaryExpressions, tokenizeKeyVars, filterExpression, primaryIndexModificationOp.getOperation(), primaryIndexModificationOp.isBulkload(), isPartitioned, varTypes);
                tokenUpdate.getInputs().add(new MutableObject<ILogicalOperator>(currentTop));
                context.computeAndSetTypeEnvironmentForOperator(tokenUpdate);
                replicateOutput = tokenUpdate;
                indexUpdate = new IndexInsertDeleteUpsertOperator(dataSourceIndex, primaryIndexModificationOp.getPrimaryKeyExpressions(), tokenizeKeyExprs, filterExpression, primaryIndexModificationOp.getOperation(), primaryIndexModificationOp.isBulkload(), primaryIndexModificationOp.getAdditionalNonFilteringExpressions() == null ? 0 : primaryIndexModificationOp.getAdditionalNonFilteringExpressions().size());
                indexUpdate.setAdditionalFilteringExpressions(filteringExpressions);
                indexUpdate.getInputs().add(new MutableObject<ILogicalOperator>(tokenUpdate));
            } else {
                // When TokenizeOperator is not needed
                indexUpdate = new IndexInsertDeleteUpsertOperator(dataSourceIndex, primaryIndexModificationOp.getPrimaryKeyExpressions(), secondaryExpressions, filterExpression, primaryIndexModificationOp.getOperation(), primaryIndexModificationOp.isBulkload(), primaryIndexModificationOp.getAdditionalNonFilteringExpressions() == null ? 0 : primaryIndexModificationOp.getAdditionalNonFilteringExpressions().size());
                indexUpdate.setAdditionalFilteringExpressions(filteringExpressions);
                replicateOutput = indexUpdate;
                // We add the necessary expressions for upsert
                if (primaryIndexModificationOp.getOperation() == Kind.UPSERT) {
                    indexUpdate.setBeforeOpSecondaryKeyExprs(beforeOpSecondaryExpressions);
                    if (filteringFields != null) {
                        indexUpdate.setBeforeOpAdditionalFilteringExpression(new MutableObject<ILogicalExpression>(new VariableReferenceExpression(primaryIndexModificationOp.getBeforeOpFilterVar())));
                    }
                }
                indexUpdate.getInputs().add(new MutableObject<ILogicalOperator>(currentTop));
            }
        } else {
            // Get type, dimensions and number of keys
            Pair<IAType, Boolean> keyPairType = Index.getNonNullableOpenFieldType(index.getKeyFieldTypes().get(0), secondaryKeyFields.get(0), recType);
            IAType spatialType = keyPairType.first;
            boolean isPointMBR = spatialType.getTypeTag() == ATypeTag.POINT || spatialType.getTypeTag() == ATypeTag.POINT3D;
            int dimension = NonTaggedFormatUtil.getNumDimensions(spatialType.getTypeTag());
            int numKeys = (isPointMBR && isBulkload) ? dimension : dimension * 2;
            // Get variables and expressions
            List<LogicalVariable> keyVarList = new ArrayList<>();
            List<Mutable<ILogicalExpression>> keyExprList = new ArrayList<>();
            for (int i = 0; i < numKeys; i++) {
                LogicalVariable keyVar = context.newVar();
                keyVarList.add(keyVar);
                AbstractFunctionCallExpression createMBR = new ScalarFunctionCallExpression(FunctionUtil.getFunctionInfo(BuiltinFunctions.CREATE_MBR));
                createMBR.getArguments().add(new MutableObject<ILogicalExpression>(new VariableReferenceExpression(secondaryKeyVars.get(0))));
                createMBR.getArguments().add(new MutableObject<ILogicalExpression>(new ConstantExpression(new AsterixConstantValue(new AInt32(dimension)))));
                createMBR.getArguments().add(new MutableObject<ILogicalExpression>(new ConstantExpression(new AsterixConstantValue(new AInt32(i)))));
                keyExprList.add(new MutableObject<ILogicalExpression>(createMBR));
            }
            secondaryExpressions.clear();
            for (LogicalVariable secondaryKeyVar : keyVarList) {
                secondaryExpressions.add(new MutableObject<ILogicalExpression>(new VariableReferenceExpression(secondaryKeyVar)));
            }
            if (isPointMBR && isBulkload) {
                //createFieldPermutationForBulkLoadOp(int) for more details.
                for (LogicalVariable secondaryKeyVar : keyVarList) {
                    secondaryExpressions.add(new MutableObject<ILogicalExpression>(new VariableReferenceExpression(secondaryKeyVar)));
                }
            }
            AssignOperator assignCoordinates = new AssignOperator(keyVarList, keyExprList);
            assignCoordinates.getInputs().add(new MutableObject<ILogicalOperator>(currentTop));
            context.computeAndSetTypeEnvironmentForOperator(assignCoordinates);
            replicateOutput = assignCoordinates;
            Mutable<ILogicalExpression> filterExpression = null;
            AssignOperator originalAssignCoordinates = null;
            // We do something similar for beforeOp key if the operation is an upsert
            if (primaryIndexModificationOp.getOperation() == Kind.UPSERT) {
                List<LogicalVariable> originalKeyVarList = new ArrayList<>();
                List<Mutable<ILogicalExpression>> originalKeyExprList = new ArrayList<>();
                // we don't do any filtering since nulls are expected here and there
                for (int i = 0; i < numKeys; i++) {
                    LogicalVariable keyVar = context.newVar();
                    originalKeyVarList.add(keyVar);
                    AbstractFunctionCallExpression createMBR = new ScalarFunctionCallExpression(FunctionUtil.getFunctionInfo(BuiltinFunctions.CREATE_MBR));
                    createMBR.getArguments().add(beforeOpSecondaryExpressions.get(0));
                    createMBR.getArguments().add(new MutableObject<ILogicalExpression>(new ConstantExpression(new AsterixConstantValue(new AInt32(dimension)))));
                    createMBR.getArguments().add(new MutableObject<ILogicalExpression>(new ConstantExpression(new AsterixConstantValue(new AInt32(i)))));
                    originalKeyExprList.add(new MutableObject<ILogicalExpression>(createMBR));
                }
                beforeOpSecondaryExpressions.clear();
                for (LogicalVariable secondaryKeyVar : originalKeyVarList) {
                    beforeOpSecondaryExpressions.add(new MutableObject<ILogicalExpression>(new VariableReferenceExpression(secondaryKeyVar)));
                }
                originalAssignCoordinates = new AssignOperator(originalKeyVarList, originalKeyExprList);
                originalAssignCoordinates.getInputs().add(new MutableObject<ILogicalOperator>(assignCoordinates));
                context.computeAndSetTypeEnvironmentForOperator(originalAssignCoordinates);
            } else {
                // We must enforce the filter if the originating spatial type is
                // nullable.
                boolean forceFilter = keyPairType.second;
                filterExpression = createFilterExpression(keyVarList, context.getOutputTypeEnvironment(assignCoordinates), forceFilter);
            }
            DataSourceIndex dataSourceIndex = new DataSourceIndex(index, dataverseName, datasetName, mp);
            indexUpdate = new IndexInsertDeleteUpsertOperator(dataSourceIndex, primaryIndexModificationOp.getPrimaryKeyExpressions(), secondaryExpressions, filterExpression, primaryIndexModificationOp.getOperation(), primaryIndexModificationOp.isBulkload(), primaryIndexModificationOp.getAdditionalNonFilteringExpressions() == null ? 0 : primaryIndexModificationOp.getAdditionalNonFilteringExpressions().size());
            indexUpdate.setAdditionalFilteringExpressions(filteringExpressions);
            if (primaryIndexModificationOp.getOperation() == Kind.UPSERT) {
                // set before op secondary key expressions
                if (filteringFields != null) {
                    indexUpdate.setBeforeOpAdditionalFilteringExpression(new MutableObject<ILogicalExpression>(new VariableReferenceExpression(primaryIndexModificationOp.getBeforeOpFilterVar())));
                }
                // set filtering expressions
                indexUpdate.setBeforeOpSecondaryKeyExprs(beforeOpSecondaryExpressions);
                // assign --> assign beforeOp values --> secondary index upsert
                indexUpdate.getInputs().add(new MutableObject<ILogicalOperator>(originalAssignCoordinates));
            } else {
                indexUpdate.getInputs().add(new MutableObject<ILogicalOperator>(assignCoordinates));
            }
        }
        context.computeAndSetTypeEnvironmentForOperator(indexUpdate);
        if (!primaryIndexModificationOp.isBulkload() || secondaryIndexTotalCnt == 1) {
            currentTop = indexUpdate;
        } else {
            replicateOp.getOutputs().add(new MutableObject<>(replicateOutput));
        }
        if (primaryIndexModificationOp.isBulkload()) {
            // For bulk load, we connect all fanned out insert operator to a single SINK operator
            op0.getInputs().add(new MutableObject<ILogicalOperator>(indexUpdate));
        }
    }
    if (!hasSecondaryIndex) {
        return false;
    }
    if (!primaryIndexModificationOp.isBulkload()) {
        // If this is an upsert, we need to
        // Remove the current input to the SINK operator (It is actually already removed above)
        op0.getInputs().clear();
        // Connect the last index update to the SINK
        op0.getInputs().add(new MutableObject<ILogicalOperator>(currentTop));
    }
    return true;
}

Example 2 with DataSource

use of org.apache.asterix.metadata.declared.DataSource in project asterixdb by apache.

the class PushFieldAccessRule method propagateFieldAccessRec.

@SuppressWarnings("unchecked")
private boolean propagateFieldAccessRec(Mutable<ILogicalOperator> opRef, IOptimizationContext context, String finalAnnot) throws AlgebricksException {
    AssignOperator access = (AssignOperator) opRef.getValue();
    Mutable<ILogicalOperator> opRef2 = access.getInputs().get(0);
    AbstractLogicalOperator op2 = (AbstractLogicalOperator) opRef2.getValue();
    // rewritten into index search.
    if (op2.getOperatorTag() == LogicalOperatorTag.PROJECT || context.checkAndAddToAlreadyCompared(access, op2) && !(op2.getOperatorTag() == LogicalOperatorTag.SELECT && isAccessToIndexedField(access, context))) {
        return false;
    }
    Object annotation = op2.getAnnotations().get(IS_MOVABLE);
    if (annotation != null && !((Boolean) annotation)) {
        return false;
    }
    if (tryingToPushThroughSelectionWithSameDataSource(access, op2)) {
        return false;
    }
    if (testAndModifyRedundantOp(access, op2)) {
        propagateFieldAccessRec(opRef2, context, finalAnnot);
        return true;
    }
    List<LogicalVariable> usedInAccess = new LinkedList<>();
    VariableUtilities.getUsedVariables(access, usedInAccess);
    List<LogicalVariable> produced2 = new LinkedList<>();
    if (op2.getOperatorTag() == LogicalOperatorTag.GROUP) {
        VariableUtilities.getLiveVariables(op2, produced2);
    } else {
        VariableUtilities.getProducedVariables(op2, produced2);
    }
    boolean pushItDown = false;
    List<LogicalVariable> inter = new ArrayList<>(usedInAccess);
    if (inter.isEmpty()) {
        // ground value
        return false;
    }
    inter.retainAll(produced2);
    if (inter.isEmpty()) {
        pushItDown = true;
    } else if (op2.getOperatorTag() == LogicalOperatorTag.GROUP) {
        GroupByOperator g = (GroupByOperator) op2;
        List<Pair<LogicalVariable, LogicalVariable>> varMappings = new ArrayList<>();
        for (Pair<LogicalVariable, Mutable<ILogicalExpression>> p : g.getDecorList()) {
            ILogicalExpression e = p.second.getValue();
            if (e.getExpressionTag() == LogicalExpressionTag.VARIABLE) {
                LogicalVariable decorVar = GroupByOperator.getDecorVariable(p);
                if (inter.contains(decorVar)) {
                    inter.remove(decorVar);
                    LogicalVariable v1 = ((VariableReferenceExpression) e).getVariableReference();
                    varMappings.add(new Pair<>(decorVar, v1));
                }
            }
        }
        if (inter.isEmpty()) {
            boolean changed = false;
            for (Pair<LogicalVariable, LogicalVariable> m : varMappings) {
                LogicalVariable v2 = context.newVar();
                LogicalVariable oldVar = access.getVariables().get(0);
                g.getDecorList().add(new Pair<LogicalVariable, Mutable<ILogicalExpression>>(oldVar, new MutableObject<ILogicalExpression>(new VariableReferenceExpression(v2))));
                changed = true;
                access.getVariables().set(0, v2);
                VariableUtilities.substituteVariables(access, m.first, m.second, context);
            }
            if (changed) {
                context.computeAndSetTypeEnvironmentForOperator(g);
            }
            usedInAccess.clear();
            VariableUtilities.getUsedVariables(access, usedInAccess);
            pushItDown = true;
        }
    }
    if (pushItDown) {
        if (op2.getOperatorTag() == LogicalOperatorTag.NESTEDTUPLESOURCE) {
            Mutable<ILogicalOperator> childOfSubplan = ((NestedTupleSourceOperator) op2).getDataSourceReference().getValue().getInputs().get(0);
            pushAccessDown(opRef, op2, childOfSubplan, context, finalAnnot);
            return true;
        }
        if (op2.getInputs().size() == 1 && !op2.hasNestedPlans()) {
            pushAccessDown(opRef, op2, op2.getInputs().get(0), context, finalAnnot);
            return true;
        } else {
            for (Mutable<ILogicalOperator> inp : op2.getInputs()) {
                HashSet<LogicalVariable> v2 = new HashSet<>();
                VariableUtilities.getLiveVariables(inp.getValue(), v2);
                if (v2.containsAll(usedInAccess)) {
                    pushAccessDown(opRef, op2, inp, context, finalAnnot);
                    return true;
                }
            }
        }
        if (op2.hasNestedPlans()) {
            AbstractOperatorWithNestedPlans nestedOp = (AbstractOperatorWithNestedPlans) op2;
            for (ILogicalPlan plan : nestedOp.getNestedPlans()) {
                for (Mutable<ILogicalOperator> root : plan.getRoots()) {
                    HashSet<LogicalVariable> v2 = new HashSet<>();
                    VariableUtilities.getLiveVariables(root.getValue(), v2);
                    if (v2.containsAll(usedInAccess)) {
                        pushAccessDown(opRef, op2, root, context, finalAnnot);
                        return true;
                    }
                }
            }
        }
        throw new AlgebricksException("Field access " + access.getExpressions().get(0).getValue() + " does not correspond to any input of operator " + op2);
    } else {
        // fields. If yes, we can equate the two variables.
        if (op2.getOperatorTag() == LogicalOperatorTag.DATASOURCESCAN) {
            DataSourceScanOperator scan = (DataSourceScanOperator) op2;
            int n = scan.getVariables().size();
            LogicalVariable scanRecordVar = scan.getVariables().get(n - 1);
            AbstractFunctionCallExpression accessFun = (AbstractFunctionCallExpression) access.getExpressions().get(0).getValue();
            ILogicalExpression e0 = accessFun.getArguments().get(0).getValue();
            LogicalExpressionTag tag = e0.getExpressionTag();
            if (tag == LogicalExpressionTag.VARIABLE) {
                VariableReferenceExpression varRef = (VariableReferenceExpression) e0;
                if (varRef.getVariableReference() == scanRecordVar) {
                    ILogicalExpression e1 = accessFun.getArguments().get(1).getValue();
                    if (e1.getExpressionTag() == LogicalExpressionTag.CONSTANT) {
                        IDataSource<DataSourceId> dataSource = (IDataSource<DataSourceId>) scan.getDataSource();
                        byte dsType = ((DataSource) dataSource).getDatasourceType();
                        if (dsType == DataSource.Type.FEED || dsType == DataSource.Type.LOADABLE) {
                            return false;
                        }
                        DataSourceId asid = dataSource.getId();
                        MetadataProvider mp = (MetadataProvider) context.getMetadataProvider();
                        Dataset dataset = mp.findDataset(asid.getDataverseName(), asid.getDatasourceName());
                        if (dataset == null) {
                            throw new AlgebricksException("Dataset " + asid.getDatasourceName() + " not found.");
                        }
                        if (dataset.getDatasetType() != DatasetType.INTERNAL) {
                            setAsFinal(access, context, finalAnnot);
                            return false;
                        }
                        ConstantExpression ce = (ConstantExpression) e1;
                        IAObject obj = ((AsterixConstantValue) ce.getValue()).getObject();
                        String fldName;
                        if (obj.getType().getTypeTag() == ATypeTag.STRING) {
                            fldName = ((AString) obj).getStringValue();
                        } else {
                            int pos = ((AInt32) obj).getIntegerValue();
                            String tName = dataset.getItemTypeName();
                            IAType t = mp.findType(dataset.getItemTypeDataverseName(), tName);
                            if (t.getTypeTag() != ATypeTag.OBJECT) {
                                return false;
                            }
                            ARecordType rt = (ARecordType) t;
                            if (pos >= rt.getFieldNames().length) {
                                setAsFinal(access, context, finalAnnot);
                                return false;
                            }
                            fldName = rt.getFieldNames()[pos];
                        }
                        int p = DatasetUtil.getPositionOfPartitioningKeyField(dataset, fldName);
                        if (p < 0) {
                            // not one of the partitioning fields
                            setAsFinal(access, context, finalAnnot);
                            return false;
                        }
                        LogicalVariable keyVar = scan.getVariables().get(p);
                        access.getExpressions().get(0).setValue(new VariableReferenceExpression(keyVar));
                        return true;
                    }
                }
            }
        }
        setAsFinal(access, context, finalAnnot);
        return false;
    }
}

Example 3 with DataSource

use of org.apache.asterix.metadata.declared.DataSource in project asterixdb by apache.

the class UnnestToDataScanRule method handleFunction.

protected boolean handleFunction(Mutable<ILogicalOperator> opRef, IOptimizationContext context, UnnestOperator unnest, AbstractFunctionCallExpression f) throws AlgebricksException {
    FunctionIdentifier fid = f.getFunctionIdentifier();
    if (fid.equals(BuiltinFunctions.DATASET)) {
        if (unnest.getPositionalVariable() != null) {
            // TODO remove this after enabling the support of positional variables in data scan
            throw new AlgebricksException("No positional variables are allowed over datasets.");
        }
        ILogicalExpression expr = f.getArguments().get(0).getValue();
        if (expr.getExpressionTag() != LogicalExpressionTag.CONSTANT) {
            return false;
        }
        ConstantExpression ce = (ConstantExpression) expr;
        IAlgebricksConstantValue acv = ce.getValue();
        if (!(acv instanceof AsterixConstantValue)) {
            return false;
        }
        AsterixConstantValue acv2 = (AsterixConstantValue) acv;
        if (acv2.getObject().getType().getTypeTag() != ATypeTag.STRING) {
            return false;
        }
        String datasetArg = ((AString) acv2.getObject()).getStringValue();
        MetadataProvider metadataProvider = (MetadataProvider) context.getMetadataProvider();
        Pair<String, String> datasetReference = parseDatasetReference(metadataProvider, datasetArg);
        String dataverseName = datasetReference.first;
        String datasetName = datasetReference.second;
        Dataset dataset = metadataProvider.findDataset(dataverseName, datasetName);
        if (dataset == null) {
            throw new AlgebricksException("Could not find dataset " + datasetName + " in dataverse " + dataverseName);
        }
        DataSourceId asid = new DataSourceId(dataverseName, datasetName);
        List<LogicalVariable> variables = new ArrayList<>();
        if (dataset.getDatasetType() == DatasetType.INTERNAL) {
            int numPrimaryKeys = dataset.getPrimaryKeys().size();
            for (int i = 0; i < numPrimaryKeys; i++) {
                variables.add(context.newVar());
            }
        }
        variables.add(unnest.getVariable());
        DataSource dataSource = metadataProvider.findDataSource(asid);
        boolean hasMeta = dataSource.hasMeta();
        if (hasMeta) {
            variables.add(context.newVar());
        }
        DataSourceScanOperator scan = new DataSourceScanOperator(variables, dataSource);
        List<Mutable<ILogicalOperator>> scanInpList = scan.getInputs();
        scanInpList.addAll(unnest.getInputs());
        opRef.setValue(scan);
        addPrimaryKey(variables, dataSource, context);
        context.computeAndSetTypeEnvironmentForOperator(scan);
        // Adds equivalence classes --- one equivalent class between a primary key
        // variable and a record field-access expression.
        IAType[] schemaTypes = dataSource.getSchemaTypes();
        ARecordType recordType = (ARecordType) (hasMeta ? schemaTypes[schemaTypes.length - 2] : schemaTypes[schemaTypes.length - 1]);
        ARecordType metaRecordType = (ARecordType) (hasMeta ? schemaTypes[schemaTypes.length - 1] : null);
        EquivalenceClassUtils.addEquivalenceClassesForPrimaryIndexAccess(scan, variables, recordType, metaRecordType, dataset, context);
        return true;
    } else if (fid.equals(BuiltinFunctions.FEED_COLLECT)) {
        if (unnest.getPositionalVariable() != null) {
            throw new AlgebricksException("No positional variables are allowed over feeds.");
        }
        String dataverse = ConstantExpressionUtil.getStringArgument(f, 0);
        String sourceFeedName = ConstantExpressionUtil.getStringArgument(f, 1);
        String getTargetFeed = ConstantExpressionUtil.getStringArgument(f, 2);
        String subscriptionLocation = ConstantExpressionUtil.getStringArgument(f, 3);
        String targetDataset = ConstantExpressionUtil.getStringArgument(f, 4);
        String outputType = ConstantExpressionUtil.getStringArgument(f, 5);
        MetadataProvider metadataProvider = (MetadataProvider) context.getMetadataProvider();
        DataSourceId asid = new DataSourceId(dataverse, getTargetFeed);
        String policyName = metadataProvider.getConfig().get(FeedActivityDetails.FEED_POLICY_NAME);
        FeedPolicyEntity policy = metadataProvider.findFeedPolicy(dataverse, policyName);
        if (policy == null) {
            policy = BuiltinFeedPolicies.getFeedPolicy(policyName);
            if (policy == null) {
                throw new AlgebricksException("Unknown feed policy:" + policyName);
            }
        }
        ArrayList<LogicalVariable> feedDataScanOutputVariables = new ArrayList<>();
        String csLocations = metadataProvider.getConfig().get(FeedActivityDetails.COLLECT_LOCATIONS);
        List<LogicalVariable> pkVars = new ArrayList<>();
        FeedDataSource ds = createFeedDataSource(asid, targetDataset, sourceFeedName, subscriptionLocation, metadataProvider, policy, outputType, csLocations, unnest.getVariable(), context, pkVars);
        // The order for feeds is <Record-Meta-PK>
        feedDataScanOutputVariables.add(unnest.getVariable());
        // Does it produce meta?
        if (ds.hasMeta()) {
            feedDataScanOutputVariables.add(context.newVar());
        }
        // Does it produce pk?
        if (ds.isChange()) {
            feedDataScanOutputVariables.addAll(pkVars);
        }
        DataSourceScanOperator scan = new DataSourceScanOperator(feedDataScanOutputVariables, ds);
        List<Mutable<ILogicalOperator>> scanInpList = scan.getInputs();
        scanInpList.addAll(unnest.getInputs());
        opRef.setValue(scan);
        context.computeAndSetTypeEnvironmentForOperator(scan);
        return true;
    }
    return false;
}

Example 4 with DataSource

use of org.apache.asterix.metadata.declared.DataSource in project asterixdb by apache.

the class IntroduceDynamicTypeCastRule method rewritePost.

@Override
public boolean rewritePost(Mutable<ILogicalOperator> opRef, IOptimizationContext context) throws AlgebricksException {
    // Depending on the operator type, we need to extract the following pieces of information.
    AbstractLogicalOperator op;
    ARecordType requiredRecordType;
    LogicalVariable recordVar;
    // We identify INSERT and DISTRIBUTE_RESULT operators.
    AbstractLogicalOperator op1 = (AbstractLogicalOperator) opRef.getValue();
    switch(op1.getOperatorTag()) {
        case SINK:
        case DELEGATE_OPERATOR:
            {
                /**
                 * pattern match: commit insert assign
                 * resulting plan: commit-insert-project-assign
                 */
                if (op1.getOperatorTag() == LogicalOperatorTag.DELEGATE_OPERATOR) {
                    DelegateOperator eOp = (DelegateOperator) op1;
                    if (!(eOp.getDelegate() instanceof CommitOperator)) {
                        return false;
                    }
                }
                AbstractLogicalOperator op2 = (AbstractLogicalOperator) op1.getInputs().get(0).getValue();
                if (op2.getOperatorTag() == LogicalOperatorTag.INSERT_DELETE_UPSERT) {
                    InsertDeleteUpsertOperator insertDeleteOp = (InsertDeleteUpsertOperator) op2;
                    if (insertDeleteOp.getOperation() == InsertDeleteUpsertOperator.Kind.DELETE) {
                        return false;
                    }
                    // Remember this is the operator we need to modify
                    op = insertDeleteOp;
                    // Derive the required ARecordType based on the schema of the DataSource
                    InsertDeleteUpsertOperator insertDeleteOperator = (InsertDeleteUpsertOperator) op2;
                    DataSource dataSource = (DataSource) insertDeleteOperator.getDataSource();
                    requiredRecordType = (ARecordType) dataSource.getItemType();
                    // Derive the Variable which we will potentially wrap with cast/null functions
                    ILogicalExpression expr = insertDeleteOperator.getPayloadExpression().getValue();
                    List<LogicalVariable> payloadVars = new ArrayList<>();
                    expr.getUsedVariables(payloadVars);
                    recordVar = payloadVars.get(0);
                } else {
                    return false;
                }
                break;
            }
        case DISTRIBUTE_RESULT:
            {
                // First, see if there was an output-record-type specified
                requiredRecordType = (ARecordType) op1.getAnnotations().get("output-record-type");
                if (requiredRecordType == null) {
                    return false;
                }
                // Remember this is the operator we need to modify
                op = op1;
                recordVar = ((VariableReferenceExpression) ((DistributeResultOperator) op).getExpressions().get(0).getValue()).getVariableReference();
                break;
            }
        default:
            {
                return false;
            }
    }
    // Derive the statically-computed type of the record
    IVariableTypeEnvironment env = op.computeOutputTypeEnvironment(context);
    IAType inputRecordType = (IAType) env.getVarType(recordVar);
    /** the input record type can be an union type -- for the case when it comes from a subplan or left-outer join */
    boolean checkUnknown = false;
    while (NonTaggedFormatUtil.isOptional(inputRecordType)) {
        /** while-loop for the case there is a nested multi-level union */
        inputRecordType = ((AUnionType) inputRecordType).getActualType();
        checkUnknown = true;
    }
    /** see whether the input record type needs to be casted */
    boolean cast = !compatible(requiredRecordType, inputRecordType);
    if (checkUnknown) {
        recordVar = addWrapperFunction(requiredRecordType, recordVar, op, context, BuiltinFunctions.CHECK_UNKNOWN);
    }
    if (cast) {
        addWrapperFunction(requiredRecordType, recordVar, op, context, BuiltinFunctions.CAST_TYPE);
    }
    return cast || checkUnknown;
}

Example 5 with DataSource

use of org.apache.asterix.metadata.declared.DataSource in project asterixdb by apache.

the class IntroduceMaterializationForInsertWithSelfScanRule method checkIfInsertAndScanDatasetsSame.

private boolean checkIfInsertAndScanDatasetsSame(AbstractLogicalOperator op, String insertDatasetName) {
    boolean sameDataset = false;
    for (int i = 0; i < op.getInputs().size(); ++i) {
        AbstractLogicalOperator descendantOp = (AbstractLogicalOperator) op.getInputs().get(i).getValue();
        if (descendantOp.getOperatorTag() == LogicalOperatorTag.UNNEST_MAP) {
            UnnestMapOperator unnestMapOp = (UnnestMapOperator) descendantOp;
            ILogicalExpression unnestExpr = unnestMapOp.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());
                boolean isPrimaryIndex = jobGenParams.isPrimaryIndex();
                String indexName = jobGenParams.getIndexName();
                if (isPrimaryIndex && indexName.compareTo(insertDatasetName) == 0) {
                    return true;
                }
            }
        } else if (descendantOp.getOperatorTag() == LogicalOperatorTag.DATASOURCESCAN) {
            DataSourceScanOperator dataSourceScanOp = (DataSourceScanOperator) descendantOp;
            DataSource ds = (DataSource) dataSourceScanOp.getDataSource();
            if ((ds.getDatasourceType() == Type.INTERNAL_DATASET || ds.getDatasourceType() == Type.EXTERNAL_DATASET) && ((DatasetDataSource) ds).getDataset().getDatasetName().compareTo(insertDatasetName) == 0) {
                return true;
            }
        }
        sameDataset = checkIfInsertAndScanDatasetsSame(descendantOp, insertDatasetName);
        if (sameDataset) {
            break;
        }
    }
    return sameDataset;
}