private StreamingAggregator createAggregatorInternal()
throws SchemaChangeException, ClassTransformationException, IOException {
ClassGenerator<StreamingAggregator> cg =
CodeGenerator.getRoot(
StreamingAggTemplate.TEMPLATE_DEFINITION, context.getFunctionRegistry());
container.clear();
LogicalExpression[] keyExprs = new LogicalExpression[popConfig.getKeys().length];
LogicalExpression[] valueExprs = new LogicalExpression[popConfig.getExprs().length];
TypedFieldId[] keyOutputIds = new TypedFieldId[popConfig.getKeys().length];
ErrorCollector collector = new ErrorCollectorImpl();
for (int i = 0; i < keyExprs.length; i++) {
final NamedExpression ne = popConfig.getKeys()[i];
final LogicalExpression expr =
ExpressionTreeMaterializer.materialize(
ne.getExpr(), incoming, collector, context.getFunctionRegistry());
if (expr == null) {
continue;
}
keyExprs[i] = expr;
final MaterializedField outputField =
MaterializedField.create(ne.getRef(), expr.getMajorType());
final ValueVector vector = TypeHelper.getNewVector(outputField, oContext.getAllocator());
keyOutputIds[i] = container.add(vector);
}
for (int i = 0; i < valueExprs.length; i++) {
final NamedExpression ne = popConfig.getExprs()[i];
final LogicalExpression expr =
ExpressionTreeMaterializer.materialize(
ne.getExpr(), incoming, collector, context.getFunctionRegistry());
if (expr instanceof IfExpression) {
throw UserException.unsupportedError(
new UnsupportedOperationException(
"Union type not supported in aggregate functions"))
.build(logger);
}
if (expr == null) {
continue;
}
final MaterializedField outputField =
MaterializedField.create(ne.getRef(), expr.getMajorType());
ValueVector vector = TypeHelper.getNewVector(outputField, oContext.getAllocator());
TypedFieldId id = container.add(vector);
valueExprs[i] = new ValueVectorWriteExpression(id, expr, true);
}
if (collector.hasErrors()) {
throw new SchemaChangeException(
"Failure while materializing expression. " + collector.toErrorString());
}
setupIsSame(cg, keyExprs);
setupIsSameApart(cg, keyExprs);
addRecordValues(cg, valueExprs);
outputRecordKeys(cg, keyOutputIds, keyExprs);
outputRecordKeysPrev(cg, keyOutputIds, keyExprs);
cg.getBlock("resetValues")._return(JExpr.TRUE);
getIndex(cg);
container.buildSchema(SelectionVectorMode.NONE);
StreamingAggregator agg = context.getImplementationClass(cg);
agg.setup(oContext, incoming, this);
return agg;
}
@Override
public IterOutcome innerNext() {
// if a special batch has been sent, we have no data in the incoming so exit early
if (specialBatchSent) {
return IterOutcome.NONE;
}
// this is only called on the first batch. Beyond this, the aggregator manages batches.
if (aggregator == null || first) {
IterOutcome outcome;
if (first && incoming.getRecordCount() > 0) {
first = false;
outcome = IterOutcome.OK_NEW_SCHEMA;
} else {
outcome = next(incoming);
}
logger.debug("Next outcome of {}", outcome);
switch (outcome) {
case NONE:
if (first && popConfig.getKeys().length == 0) {
// if we have a straight aggregate and empty input batch, we need to handle it in a
// different way
constructSpecialBatch();
first = false;
// set state to indicate the fact that we have sent a special batch and input is empty
specialBatchSent = true;
return IterOutcome.OK;
}
case OUT_OF_MEMORY:
case NOT_YET:
case STOP:
return outcome;
case OK_NEW_SCHEMA:
if (!createAggregator()) {
done = true;
return IterOutcome.STOP;
}
break;
case OK:
break;
default:
throw new IllegalStateException(String.format("unknown outcome %s", outcome));
}
}
AggOutcome out = aggregator.doWork();
recordCount = aggregator.getOutputCount();
logger.debug("Aggregator response {}, records {}", out, aggregator.getOutputCount());
switch (out) {
case CLEANUP_AND_RETURN:
if (!first) {
container.zeroVectors();
}
done = true;
// fall through
case RETURN_OUTCOME:
IterOutcome outcome = aggregator.getOutcome();
if (outcome == IterOutcome.NONE && first) {
first = false;
done = true;
return IterOutcome.OK_NEW_SCHEMA;
} else if (outcome == IterOutcome.OK && first) {
outcome = IterOutcome.OK_NEW_SCHEMA;
} else if (outcome != IterOutcome.OUT_OF_MEMORY) {
first = false;
}
return outcome;
case UPDATE_AGGREGATOR:
context.fail(
UserException.unsupportedError()
.message("Streaming aggregate does not support schema changes")
.build(logger));
close();
killIncoming(false);
return IterOutcome.STOP;
default:
throw new IllegalStateException(String.format("Unknown state %s.", out));
}
}
