@Override
public boolean rewritePost(Mutable<ILogicalOperator> opRef, IOptimizationContext context)
throws AlgebricksException {
AbstractLogicalOperator op = (AbstractLogicalOperator) opRef.getValue();
if (op.getOperatorTag() != LogicalOperatorTag.INNERJOIN) {
return false;
}
AbstractBinaryJoinOperator join = (AbstractBinaryJoinOperator) op;
ILogicalExpression expr = join.getCondition().getValue();
if (expr.getExpressionTag() != LogicalExpressionTag.FUNCTION_CALL) {
return false;
}
AbstractFunctionCallExpression fexp = (AbstractFunctionCallExpression) expr;
FunctionIdentifier fi = fexp.getFunctionIdentifier();
if (!fi.equals(AlgebricksBuiltinFunctions.AND)) {
return false;
}
List<Mutable<ILogicalExpression>> eqVarVarComps = new ArrayList<Mutable<ILogicalExpression>>();
List<Mutable<ILogicalExpression>> otherPredicates =
new ArrayList<Mutable<ILogicalExpression>>();
for (Mutable<ILogicalExpression> arg : fexp.getArguments()) {
if (isEqVarVar(arg.getValue())) {
eqVarVarComps.add(arg);
} else {
otherPredicates.add(arg);
}
}
if (eqVarVarComps.isEmpty() || otherPredicates.isEmpty()) {
return false;
}
// pull up
ILogicalExpression pulledCond = makeCondition(otherPredicates, context);
SelectOperator select =
new SelectOperator(new MutableObject<ILogicalExpression>(pulledCond), false, null);
ILogicalExpression newJoinCond = makeCondition(eqVarVarComps, context);
join.getCondition().setValue(newJoinCond);
select.getInputs().add(new MutableObject<ILogicalOperator>(join));
opRef.setValue(select);
context.computeAndSetTypeEnvironmentForOperator(select);
return true;
}
@SuppressWarnings("unchecked")
@Override
public void evaluate(IFrameTupleReference tuple) throws AlgebricksException {
array0.reset();
eval0.evaluate(tuple);
array1.reset();
eval1.evaluate(tuple);
array2.reset();
eval2.evaluate(tuple);
try {
if (array0.getByteArray()[0] == SER_NULL_TYPE_TAG
|| array1.getByteArray()[0] == SER_NULL_TYPE_TAG
|| array2.getByteArray()[0] == SER_NULL_TYPE_TAG) {
nullSerde.serialize(ANull.NULL, dout);
return;
} else if (array0.getByteArray()[0] != SER_STRING_TYPE_TAG
|| array1.getByteArray()[0] != SER_STRING_TYPE_TAG
|| array2.getByteArray()[0] != SER_STRING_TYPE_TAG) {
throw new AlgebricksException(
funcID.getName()
+ ": expects input type (STRING/NULL, STRING/NULL, STRING/NULL), but got ("
+ EnumDeserializer.ATYPETAGDESERIALIZER.deserialize(array0.getByteArray()[0])
+ ", "
+ EnumDeserializer.ATYPETAGDESERIALIZER.deserialize(array1.getByteArray()[0])
+ ", "
+ EnumDeserializer.ATYPETAGDESERIALIZER.deserialize(array2.getByteArray()[0])
+ ".");
}
} catch (HyracksDataException e) {
throw new AlgebricksException(e);
}
strPtr1st.set(array0.getByteArray(), array0.getStartOffset() + 1, array0.getLength());
strPtr2nd.set(array1.getByteArray(), array1.getStartOffset() + 1, array1.getLength());
strPtr3rd.set(array2.getByteArray(), array2.getStartOffset() + 1, array2.getLength());
String res = compute(strPtr1st, strPtr2nd, strPtr3rd);
resultBuffer.setValue(res);
try {
strSerde.serialize(resultBuffer, dout);
} catch (HyracksDataException e) {
throw new AlgebricksException(e);
}
}
@Override
public boolean rewritePost(Mutable<ILogicalOperator> opRef, IOptimizationContext context)
throws AlgebricksException {
AbstractLogicalOperator op = (AbstractLogicalOperator) opRef.getValue();
// current opperator is join
if (op.getOperatorTag() != LogicalOperatorTag.INNERJOIN
&& op.getOperatorTag() != LogicalOperatorTag.LEFTOUTERJOIN) {
return false;
}
// Find GET_ITEM function.
AbstractBinaryJoinOperator joinOp = (AbstractBinaryJoinOperator) op;
Mutable<ILogicalExpression> expRef = joinOp.getCondition();
Mutable<ILogicalExpression> getItemExprRef = getSimilarityExpression(expRef);
if (getItemExprRef == null) {
return false;
}
// Check if the GET_ITEM function is on one of the supported similarity-check functions.
AbstractFunctionCallExpression getItemFuncExpr =
(AbstractFunctionCallExpression) getItemExprRef.getValue();
Mutable<ILogicalExpression> argRef = getItemFuncExpr.getArguments().get(0);
AbstractFunctionCallExpression simFuncExpr = (AbstractFunctionCallExpression) argRef.getValue();
if (!simFuncs.contains(simFuncExpr.getFunctionIdentifier())) {
return false;
}
// Skip this rule based on annotations.
if (simFuncExpr.getAnnotations().containsKey(IndexedNLJoinExpressionAnnotation.INSTANCE)) {
return false;
}
List<Mutable<ILogicalOperator>> inputOps = joinOp.getInputs();
ILogicalOperator leftInputOp = inputOps.get(0).getValue();
ILogicalOperator rightInputOp = inputOps.get(1).getValue();
List<Mutable<ILogicalExpression>> inputExps = simFuncExpr.getArguments();
ILogicalExpression inputExp0 = inputExps.get(0).getValue();
ILogicalExpression inputExp1 = inputExps.get(1).getValue();
// left and right expressions are variables
if (inputExp0.getExpressionTag() != LogicalExpressionTag.VARIABLE
|| inputExp1.getExpressionTag() != LogicalExpressionTag.VARIABLE) {
return false;
}
LogicalVariable inputVar0 = ((VariableReferenceExpression) inputExp0).getVariableReference();
LogicalVariable inputVar1 = ((VariableReferenceExpression) inputExp1).getVariableReference();
LogicalVariable leftInputVar;
LogicalVariable rightInputVar;
liveVars.clear();
VariableUtilities.getLiveVariables(leftInputOp, liveVars);
if (liveVars.contains(inputVar0)) {
leftInputVar = inputVar0;
rightInputVar = inputVar1;
} else {
leftInputVar = inputVar1;
rightInputVar = inputVar0;
}
List<LogicalVariable> leftInputPKs = context.findPrimaryKey(leftInputVar);
List<LogicalVariable> rightInputPKs = context.findPrimaryKey(rightInputVar);
// Bail if primary keys could not be inferred.
if (leftInputPKs == null || rightInputPKs == null) {
return false;
}
// primary key has only one variable
if (leftInputPKs.size() != 1 || rightInputPKs.size() != 1) {
return false;
}
IAType leftType =
(IAType) context.getOutputTypeEnvironment(leftInputOp).getVarType(leftInputVar);
IAType rightType =
(IAType) context.getOutputTypeEnvironment(rightInputOp).getVarType(rightInputVar);
// left-hand side and right-hand side of "~=" has the same type
IAType left2 = TypeHelper.getNonOptionalType(leftType);
IAType right2 = TypeHelper.getNonOptionalType(rightType);
if (!left2.deepEqual(right2)) {
return false;
}
//
// -- - FIRE - --
//
AqlMetadataProvider metadataProvider = ((AqlMetadataProvider) context.getMetadataProvider());
FunctionIdentifier funcId = FuzzyUtils.getTokenizer(leftType.getTypeTag());
String tokenizer;
if (funcId == null) {
tokenizer = "";
} else {
tokenizer = funcId.getName();
}
float simThreshold = FuzzyUtils.getSimThreshold(metadataProvider);
String simFunction = FuzzyUtils.getSimFunction(metadataProvider);
// finalize AQL+ query
String prepareJoin;
switch (joinOp.getJoinKind()) {
case INNER:
{
prepareJoin = "join" + AQLPLUS;
break;
}
case LEFT_OUTER:
{
// TODO To make it work for Left Outer Joins, we should permute
// the #LEFT and #RIGHT at the top of the AQL+ query. But, when
// doing this, the
// fuzzyjoin/user-vis-int-vis-user-lot-aqlplus_1.aql (the one
// doing 3-way fuzzy joins) gives a different result. But even
// if we don't change the FuzzyJoinRule, permuting the for
// clauses in fuzzyjoin/user-vis-int-vis-user-lot-aqlplus_1.aql
// leads to different results, which suggests there is some
// other sort of bug.
return false;
// prepareJoin = "loj" + AQLPLUS;
// break;
}
default:
{
throw new IllegalStateException();
}
}
String aqlPlus =
String.format(
Locale.US,
prepareJoin,
tokenizer,
tokenizer,
simFunction,
simThreshold,
tokenizer,
tokenizer,
simFunction,
simThreshold,
simFunction,
simThreshold,
simThreshold);
LogicalVariable leftPKVar = leftInputPKs.get(0);
LogicalVariable rightPKVar = rightInputPKs.get(0);
Counter counter = new Counter(context.getVarCounter());
AQLPlusParser parser = new AQLPlusParser(new StringReader(aqlPlus));
parser.initScope();
parser.setVarCounter(counter);
List<Clause> clauses;
try {
clauses = parser.Clauses();
} catch (ParseException e) {
throw new AlgebricksException(e);
}
// The translator will compile metadata internally. Run this compilation
// under the same transaction id as the "outer" compilation.
AqlPlusExpressionToPlanTranslator translator =
new AqlPlusExpressionToPlanTranslator(
metadataProvider.getJobId(), metadataProvider, counter, null, null);
context.setVarCounter(counter.get());
LogicalOperatorDeepCopyWithNewVariablesVisitor deepCopyVisitor =
new LogicalOperatorDeepCopyWithNewVariablesVisitor(context);
translator.addOperatorToMetaScope(new Identifier("#LEFT"), leftInputOp);
translator.addVariableToMetaScope(new Identifier("$$LEFT"), leftInputVar);
translator.addVariableToMetaScope(new Identifier("$$LEFTPK"), leftPKVar);
translator.addOperatorToMetaScope(new Identifier("#RIGHT"), rightInputOp);
translator.addVariableToMetaScope(new Identifier("$$RIGHT"), rightInputVar);
translator.addVariableToMetaScope(new Identifier("$$RIGHTPK"), rightPKVar);
translator.addOperatorToMetaScope(
new Identifier("#LEFT_1"), deepCopyVisitor.deepCopy(leftInputOp, null));
translator.addVariableToMetaScope(
new Identifier("$$LEFT_1"), deepCopyVisitor.varCopy(leftInputVar));
translator.addVariableToMetaScope(
new Identifier("$$LEFTPK_1"), deepCopyVisitor.varCopy(leftPKVar));
deepCopyVisitor.updatePrimaryKeys(context);
deepCopyVisitor.reset();
// translator.addOperatorToMetaScope(new Identifier("#LEFT_2"),
// deepCopyVisitor.deepCopy(leftInputOp, null));
// translator.addVariableToMetaScope(new Identifier("$$LEFT_2"),
// deepCopyVisitor.varCopy(leftInputVar));
// translator.addVariableToMetaScope(new Identifier("$$LEFTPK_2"),
// deepCopyVisitor.varCopy(leftPKVar));
// deepCopyVisitor.updatePrimaryKeys(context);
// deepCopyVisitor.reset();
//
// translator.addOperatorToMetaScope(new Identifier("#LEFT_3"),
// deepCopyVisitor.deepCopy(leftInputOp, null));
// translator.addVariableToMetaScope(new Identifier("$$LEFT_3"),
// deepCopyVisitor.varCopy(leftInputVar));
// translator.addVariableToMetaScope(new Identifier("$$LEFTPK_3"),
// deepCopyVisitor.varCopy(leftPKVar));
// deepCopyVisitor.updatePrimaryKeys(context);
// deepCopyVisitor.reset();
translator.addOperatorToMetaScope(
new Identifier("#RIGHT_1"), deepCopyVisitor.deepCopy(rightInputOp, null));
translator.addVariableToMetaScope(
new Identifier("$$RIGHT_1"), deepCopyVisitor.varCopy(rightInputVar));
translator.addVariableToMetaScope(
new Identifier("$$RIGHTPK_1"), deepCopyVisitor.varCopy(rightPKVar));
deepCopyVisitor.updatePrimaryKeys(context);
deepCopyVisitor.reset();
// TODO pick side to run Stage 1, currently always picks RIGHT side
translator.addOperatorToMetaScope(
new Identifier("#RIGHT_2"), deepCopyVisitor.deepCopy(rightInputOp, null));
translator.addVariableToMetaScope(
new Identifier("$$RIGHT_2"), deepCopyVisitor.varCopy(rightInputVar));
translator.addVariableToMetaScope(
new Identifier("$$RIGHTPK_2"), deepCopyVisitor.varCopy(rightPKVar));
deepCopyVisitor.updatePrimaryKeys(context);
deepCopyVisitor.reset();
translator.addOperatorToMetaScope(
new Identifier("#RIGHT_3"), deepCopyVisitor.deepCopy(rightInputOp, null));
translator.addVariableToMetaScope(
new Identifier("$$RIGHT_3"), deepCopyVisitor.varCopy(rightInputVar));
translator.addVariableToMetaScope(
new Identifier("$$RIGHTPK_3"), deepCopyVisitor.varCopy(rightPKVar));
deepCopyVisitor.updatePrimaryKeys(context);
deepCopyVisitor.reset();
ILogicalPlan plan;
try {
plan = translator.translate(clauses);
} catch (AsterixException e) {
throw new AlgebricksException(e);
}
context.setVarCounter(counter.get());
ILogicalOperator outputOp = plan.getRoots().get(0).getValue();
SelectOperator extraSelect = null;
if (getItemExprRef != expRef) {
// more than one join condition
getItemExprRef.setValue(ConstantExpression.TRUE);
switch (joinOp.getJoinKind()) {
case INNER:
{
extraSelect = new SelectOperator(expRef, false, null);
extraSelect.getInputs().add(new MutableObject<ILogicalOperator>(outputOp));
outputOp = extraSelect;
break;
}
case LEFT_OUTER:
{
if (((AbstractLogicalOperator) outputOp).getOperatorTag()
!= LogicalOperatorTag.LEFTOUTERJOIN) {
throw new IllegalStateException();
}
LeftOuterJoinOperator topJoin = (LeftOuterJoinOperator) outputOp;
topJoin.getCondition().setValue(expRef.getValue());
break;
}
default:
{
throw new IllegalStateException();
}
}
}
opRef.setValue(outputOp);
OperatorPropertiesUtil.typeOpRec(opRef, context);
return true;
}