use of org.apache.hyracks.algebricks.core.algebra.expressions.ScalarFunctionCallExpression in project asterixdb by apache.
the class NonTaggedDataFormat method partitioningEvaluatorFactory.
@SuppressWarnings("unchecked")
@Override
public Triple<IScalarEvaluatorFactory, ScalarFunctionCallExpression, IAType> partitioningEvaluatorFactory(ARecordType recType, List<String> fldName) throws AlgebricksException {
String[] names = recType.getFieldNames();
int n = names.length;
if (fldName.size() > 1) {
for (int i = 0; i < n; i++) {
if (names[i].equals(fldName.get(0))) {
IScalarEvaluatorFactory recordEvalFactory = new ColumnAccessEvalFactory(GlobalConfig.DEFAULT_INPUT_DATA_COLUMN);
ArrayBackedValueStorage abvs = new ArrayBackedValueStorage();
DataOutput dos = abvs.getDataOutput();
try {
AInt32 ai = new AInt32(i);
SerializerDeserializerProvider.INSTANCE.getSerializerDeserializer(ai.getType()).serialize(ai, dos);
} catch (HyracksDataException e) {
throw new AlgebricksException(e);
}
IScalarEvaluatorFactory fldIndexEvalFactory = new ConstantEvalFactory(Arrays.copyOf(abvs.getByteArray(), abvs.getLength()));
IScalarEvaluatorFactory evalFactory = new FieldAccessByIndexEvalFactory(recordEvalFactory, fldIndexEvalFactory, recType);
IFunctionInfo finfoAccess = BuiltinFunctions.getAsterixFunctionInfo(BuiltinFunctions.FIELD_ACCESS_BY_INDEX);
ScalarFunctionCallExpression partitionFun = new ScalarFunctionCallExpression(finfoAccess, new MutableObject<ILogicalExpression>(new VariableReferenceExpression(METADATA_DUMMY_VAR)), new MutableObject<ILogicalExpression>(new ConstantExpression(new AsterixConstantValue(new AInt32(i)))));
return new Triple<IScalarEvaluatorFactory, ScalarFunctionCallExpression, IAType>(evalFactory, partitionFun, recType.getFieldTypes()[i]);
}
}
} else {
IScalarEvaluatorFactory recordEvalFactory = new ColumnAccessEvalFactory(GlobalConfig.DEFAULT_INPUT_DATA_COLUMN);
ArrayBackedValueStorage abvs = new ArrayBackedValueStorage();
DataOutput dos = abvs.getDataOutput();
AOrderedList as = new AOrderedList(fldName);
try {
SerializerDeserializerProvider.INSTANCE.getSerializerDeserializer(as.getType()).serialize(as, dos);
} catch (HyracksDataException e) {
throw new AlgebricksException(e);
}
IScalarEvaluatorFactory evalFactory = new FieldAccessNestedEvalFactory(recordEvalFactory, recType, fldName);
IFunctionInfo finfoAccess = BuiltinFunctions.getAsterixFunctionInfo(BuiltinFunctions.FIELD_ACCESS_NESTED);
ScalarFunctionCallExpression partitionFun = new ScalarFunctionCallExpression(finfoAccess, new MutableObject<ILogicalExpression>(new VariableReferenceExpression(METADATA_DUMMY_VAR)), new MutableObject<ILogicalExpression>(new ConstantExpression(new AsterixConstantValue(as))));
return new Triple<IScalarEvaluatorFactory, ScalarFunctionCallExpression, IAType>(evalFactory, partitionFun, recType.getSubFieldType(fldName));
}
throw new AlgebricksException("Could not find field " + fldName + " in the schema.");
}
use of org.apache.hyracks.algebricks.core.algebra.expressions.ScalarFunctionCallExpression 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;
}
use of org.apache.hyracks.algebricks.core.algebra.expressions.ScalarFunctionCallExpression in project asterixdb by apache.
the class IntroduceSecondaryIndexInsertDeleteRule method injectFieldAccessesForIndexes.
private ILogicalOperator injectFieldAccessesForIndexes(IOptimizationContext context, Dataset dataset, List<Index> indexes, Map<IndexFieldId, LogicalVariable> fieldAccessVars, ARecordType recType, ARecordType metaType, LogicalVariable recordVar, LogicalVariable metaVar, ILogicalOperator currentTop, boolean afterOp) throws AlgebricksException {
List<LogicalVariable> vars = new ArrayList<>();
List<Mutable<ILogicalExpression>> exprs = new ArrayList<>();
for (Index index : indexes) {
if (index.isPrimaryIndex()) {
continue;
}
List<IAType> skTypes = index.getKeyFieldTypes();
List<List<String>> skNames = index.getKeyFieldNames();
List<Integer> indicators = index.getKeyFieldSourceIndicators();
for (int i = 0; i < index.getKeyFieldNames().size(); i++) {
IndexFieldId indexFieldId = new IndexFieldId(indicators.get(i), skNames.get(i));
if (fieldAccessVars.containsKey(indexFieldId)) {
// already handled in a different index
continue;
}
ARecordType sourceType = dataset.hasMetaPart() ? indicators.get(i).intValue() == Index.RECORD_INDICATOR ? recType : metaType : recType;
LogicalVariable sourceVar = dataset.hasMetaPart() ? indicators.get(i).intValue() == Index.RECORD_INDICATOR ? recordVar : metaVar : recordVar;
LogicalVariable fieldVar = context.newVar();
// create record variable ref
Mutable<ILogicalExpression> varRef = new MutableObject<>(new VariableReferenceExpression(sourceVar));
IAType fieldType = sourceType.getSubFieldType(indexFieldId.fieldName);
AbstractFunctionCallExpression theFieldAccessFunc;
if (fieldType == null) {
// Open field. must prevent inlining to maintain the cast before the primaryOp and
// make handling of records with incorrect value type for this field easier and cleaner
context.addNotToBeInlinedVar(fieldVar);
// create field access
AbstractFunctionCallExpression fieldAccessFunc = getOpenOrNestedFieldAccessFunction(varRef, indexFieldId.fieldName);
// create cast
theFieldAccessFunc = new ScalarFunctionCallExpression(FunctionUtil.getFunctionInfo(BuiltinFunctions.CAST_TYPE));
// The first argument is the field
theFieldAccessFunc.getArguments().add(new MutableObject<ILogicalExpression>(fieldAccessFunc));
TypeCastUtils.setRequiredAndInputTypes(theFieldAccessFunc, skTypes.get(i), BuiltinType.ANY);
} else {
// Get the desired field position
int pos = indexFieldId.fieldName.size() > 1 ? -1 : sourceType.getFieldIndex(indexFieldId.fieldName.get(0));
// Field not found --> This is either an open field or a nested field. it can't be accessed by index
theFieldAccessFunc = (pos == -1) ? getOpenOrNestedFieldAccessFunction(varRef, indexFieldId.fieldName) : getClosedFieldAccessFunction(varRef, pos);
}
vars.add(fieldVar);
exprs.add(new MutableObject<ILogicalExpression>(theFieldAccessFunc));
fieldAccessVars.put(indexFieldId, fieldVar);
}
}
// AssignOperator assigns secondary keys to their vars
AssignOperator castedFieldAssignOperator = new AssignOperator(vars, exprs);
return introduceNewOp(context, currentTop, castedFieldAssignOperator, afterOp);
}
use of org.apache.hyracks.algebricks.core.algebra.expressions.ScalarFunctionCallExpression in project asterixdb by apache.
the class AccessMethodUtils method createPrimaryIndexUnnestMap.
public static AbstractUnnestMapOperator createPrimaryIndexUnnestMap(AbstractDataSourceOperator dataSourceOp, Dataset dataset, ARecordType recordType, ARecordType metaRecordType, ILogicalOperator inputOp, IOptimizationContext context, boolean sortPrimaryKeys, boolean retainInput, boolean retainNull, boolean requiresBroadcast) throws AlgebricksException {
List<LogicalVariable> primaryKeyVars = AccessMethodUtils.getPrimaryKeyVarsFromSecondaryUnnestMap(dataset, inputOp);
// Optionally add a sort on the primary-index keys before searching the primary index.
OrderOperator order = null;
if (sortPrimaryKeys) {
order = new OrderOperator();
for (LogicalVariable pkVar : primaryKeyVars) {
Mutable<ILogicalExpression> vRef = new MutableObject<>(new VariableReferenceExpression(pkVar));
order.getOrderExpressions().add(new Pair<>(OrderOperator.ASC_ORDER, vRef));
}
// The secondary-index search feeds into the sort.
order.getInputs().add(new MutableObject<>(inputOp));
order.setExecutionMode(ExecutionMode.LOCAL);
context.computeAndSetTypeEnvironmentForOperator(order);
}
// The job gen parameters are transferred to the actual job gen via the UnnestMapOperator's function arguments.
List<Mutable<ILogicalExpression>> primaryIndexFuncArgs = new ArrayList<>();
BTreeJobGenParams jobGenParams = new BTreeJobGenParams(dataset.getDatasetName(), IndexType.BTREE, dataset.getDataverseName(), dataset.getDatasetName(), retainInput, requiresBroadcast);
// Set low/high inclusive to true for a point lookup.
jobGenParams.setLowKeyInclusive(true);
jobGenParams.setHighKeyInclusive(true);
jobGenParams.setLowKeyVarList(primaryKeyVars, 0, primaryKeyVars.size());
jobGenParams.setHighKeyVarList(primaryKeyVars, 0, primaryKeyVars.size());
jobGenParams.setIsEqCondition(true);
jobGenParams.writeToFuncArgs(primaryIndexFuncArgs);
// Variables and types coming out of the primary-index search.
List<LogicalVariable> primaryIndexUnnestVars = new ArrayList<>();
List<Object> primaryIndexOutputTypes = new ArrayList<>();
// Append output variables/types generated by the primary-index search (not forwarded from input).
primaryIndexUnnestVars.addAll(dataSourceOp.getVariables());
appendPrimaryIndexTypes(dataset, recordType, metaRecordType, primaryIndexOutputTypes);
// An index search is expressed as an unnest over an index-search function.
IFunctionInfo primaryIndexSearch = FunctionUtil.getFunctionInfo(BuiltinFunctions.INDEX_SEARCH);
AbstractFunctionCallExpression primaryIndexSearchFunc = new ScalarFunctionCallExpression(primaryIndexSearch, primaryIndexFuncArgs);
// This is the operator that jobgen will be looking for. It contains an unnest function that has all necessary arguments to determine
// which index to use, which variables contain the index-search keys, what is the original dataset, etc.
AbstractUnnestMapOperator primaryIndexUnnestOp = null;
if (retainNull) {
if (retainInput) {
primaryIndexUnnestOp = new LeftOuterUnnestMapOperator(primaryIndexUnnestVars, new MutableObject<ILogicalExpression>(primaryIndexSearchFunc), primaryIndexOutputTypes, retainInput);
} else {
// Left-outer-join without retainNull and retainInput doesn't make sense.
throw new AlgebricksException("Left-outer-join should propagate all inputs from the outer branch.");
}
} else {
primaryIndexUnnestOp = new UnnestMapOperator(primaryIndexUnnestVars, new MutableObject<ILogicalExpression>(primaryIndexSearchFunc), primaryIndexOutputTypes, retainInput);
}
// Fed by the order operator or the secondaryIndexUnnestOp.
if (sortPrimaryKeys) {
primaryIndexUnnestOp.getInputs().add(new MutableObject<ILogicalOperator>(order));
} else {
primaryIndexUnnestOp.getInputs().add(new MutableObject<>(inputOp));
}
context.computeAndSetTypeEnvironmentForOperator(primaryIndexUnnestOp);
primaryIndexUnnestOp.setExecutionMode(ExecutionMode.PARTITIONED);
return primaryIndexUnnestOp;
}
use of org.apache.hyracks.algebricks.core.algebra.expressions.ScalarFunctionCallExpression in project asterixdb by apache.
the class AccessMethodUtils method findLOJIsMissingFuncInGroupBy.
public static ScalarFunctionCallExpression findLOJIsMissingFuncInGroupBy(GroupByOperator lojGroupbyOp) throws AlgebricksException {
//find IS_NULL function of which argument has the nullPlaceholder variable in the nested plan of groupby.
ALogicalPlanImpl subPlan = (ALogicalPlanImpl) lojGroupbyOp.getNestedPlans().get(0);
Mutable<ILogicalOperator> subPlanRootOpRef = subPlan.getRoots().get(0);
AbstractLogicalOperator subPlanRootOp = (AbstractLogicalOperator) subPlanRootOpRef.getValue();
boolean foundSelectNonNull = false;
ScalarFunctionCallExpression isNullFuncExpr = null;
AbstractLogicalOperator inputOp = subPlanRootOp;
while (inputOp != null) {
if (inputOp.getOperatorTag() == LogicalOperatorTag.SELECT) {
SelectOperator selectOp = (SelectOperator) inputOp;
if (selectOp.getCondition().getValue().getExpressionTag() == LogicalExpressionTag.FUNCTION_CALL) {
if (((AbstractFunctionCallExpression) selectOp.getCondition().getValue()).getFunctionIdentifier().equals(AlgebricksBuiltinFunctions.NOT)) {
ScalarFunctionCallExpression notFuncExpr = (ScalarFunctionCallExpression) selectOp.getCondition().getValue();
if (notFuncExpr.getArguments().get(0).getValue().getExpressionTag() == LogicalExpressionTag.FUNCTION_CALL) {
if (((AbstractFunctionCallExpression) notFuncExpr.getArguments().get(0).getValue()).getFunctionIdentifier().equals(AlgebricksBuiltinFunctions.IS_MISSING)) {
isNullFuncExpr = (ScalarFunctionCallExpression) notFuncExpr.getArguments().get(0).getValue();
if (isNullFuncExpr.getArguments().get(0).getValue().getExpressionTag() == LogicalExpressionTag.VARIABLE) {
foundSelectNonNull = true;
break;
}
}
}
}
}
}
inputOp = inputOp.getInputs().size() > 0 ? (AbstractLogicalOperator) inputOp.getInputs().get(0).getValue() : null;
}
if (!foundSelectNonNull) {
throw new AlgebricksException("Could not find the non-null select operator in GroupByOperator for LEFTOUTERJOIN plan optimization.");
}
return isNullFuncExpr;
}
Aggregations