/**
* @brief Convert a Postgres HashState into a Peloton HashPlanNode
* @return Pointer to the constructed AbstractPlan
*/
std::unique_ptr<planner::AbstractPlan> PlanTransformer::TransformHashJoin(
const HashJoinPlanState *hj_plan_state) {
std::unique_ptr<planner::AbstractPlan> result;
PelotonJoinType join_type =
PlanTransformer::TransformJoinType(hj_plan_state->jointype);
if (join_type == JOIN_TYPE_INVALID) {
LOG_ERROR("unsupported join type: %d", hj_plan_state->jointype);
return nullptr;
}
LOG_INFO("Handle hash join with join type: %d", join_type);
/*
std::vector<planner::HashJoinPlan::JoinClause> join_clauses(
BuildHashJoinClauses(mj_plan_state->mj_Clauses,
mj_plan_state->mj_NumClauses);
*/
expression::AbstractExpression *join_filter = ExprTransformer::TransformExpr(
reinterpret_cast<ExprState *>(hj_plan_state->joinqual));
expression::AbstractExpression *plan_filter = ExprTransformer::TransformExpr(
reinterpret_cast<ExprState *>(hj_plan_state->qual));
std::unique_ptr<const expression::AbstractExpression> predicate(nullptr);
if (join_filter && plan_filter) {
predicate.reset(expression::ExpressionUtil::ConjunctionFactory(
EXPRESSION_TYPE_CONJUNCTION_AND, join_filter, plan_filter));
} else if (join_filter) {
predicate.reset(join_filter);
} else {
predicate.reset(plan_filter);
}
/* Transform project info */
std::unique_ptr<const planner::ProjectInfo> project_info(nullptr);
project_info.reset(BuildProjectInfoFromTLSkipJunk(hj_plan_state->targetlist));
if (project_info.get() != nullptr) {
LOG_INFO("%s", project_info.get()->Debug().c_str());
} else {
LOG_INFO("empty projection info");
}
std::shared_ptr<const catalog::Schema> project_schema(
SchemaTransformer::GetSchemaFromTupleDesc(
hj_plan_state->tts_tupleDescriptor));
std::vector<oid_t> outer_hashkeys =
BuildColumnListFromExpStateList(hj_plan_state->outer_hashkeys);
bool non_trivial = (project_info.get() != nullptr &&
project_info.get()->isNonTrivial());
if (non_trivial) {
// we have non-trivial projection
LOG_INFO("We have non-trivial projection");
result = std::unique_ptr<planner::AbstractPlan>(
new planner::ProjectionPlan(std::move(project_info), project_schema));
// set project_info to nullptr
project_info.reset();
} else {
LOG_INFO("We have direct mapping projection");
}
std::unique_ptr<planner::HashJoinPlan> plan_node(new planner::HashJoinPlan(
join_type, std::move(predicate), std::move(project_info), project_schema,
outer_hashkeys));
std::unique_ptr<planner::AbstractPlan> outer{std::move(
PlanTransformer::TransformPlan(outerAbstractPlanState(hj_plan_state)))};
std::unique_ptr<planner::AbstractPlan> inner{std::move(
PlanTransformer::TransformPlan(innerAbstractPlanState(hj_plan_state)))};
/* Add the children nodes */
plan_node->AddChild(std::move(outer));
plan_node->AddChild(std::move(inner));
if (non_trivial) {
result->AddChild(std::move(plan_node));
} else {
result.reset(plan_node.release());
}
LOG_INFO("Finishing mapping Hash join, JoinType: %d", join_type);
return result;
}
const planner::AbstractPlan *PlanTransformer::TransformAgg(
const AggPlanState *plan_state) {
// Alias all I need
const Agg *agg = plan_state->agg_plan;
auto numphases = plan_state->numphases;
auto numaggs = plan_state->numaggs;
auto targetlist = plan_state->ps_targetlist;
auto qual = plan_state->ps_qual;
auto peragg = plan_state->peragg;
auto tupleDesc = plan_state->result_tupleDescriptor;
auto aggstrategy = plan_state->agg_plan->aggstrategy;
LOG_INFO("Number of Agg phases: %d \n", numphases);
// When we'll have >1 phases?
if (numphases != 1) return nullptr;
/* Get project info */
std::unique_ptr<const planner::ProjectInfo> proj_info(
BuildProjectInfoFromTLSkipJunk(targetlist));
LOG_INFO("proj_info : \n%s", proj_info->Debug().c_str());
/* Get predicate */
std::unique_ptr<const expression::AbstractExpression> predicate(
BuildPredicateFromQual(qual));
/* Get Aggregate terms */
std::vector<planner::AggregatePlan::AggTerm> unique_agg_terms;
LOG_INFO("Number of (unique) Agg nodes: %d \n", numaggs);
for (int aggno = 0; aggno < numaggs; aggno++) {
auto transfn_oid = peragg[aggno].transfn_oid;
auto itr = peloton::bridge::kPgTransitFuncMap.find(transfn_oid);
if (kPgFuncMap.end() == itr) {
LOG_ERROR("Unmapped Transit function Id : %u\n", transfn_oid);
return nullptr;
}
// We don't check whether the mapped exprtype is a valid aggregate type
// here.
PltFuncMetaInfo fn_meta = itr->second;
// We only take the first argument as input to aggregator because
// we don't have multi-argument aggregator in Peloton at the moment.
// WARNING: there can be no arguments (e.g., COUNT(*))
auto arguments = peragg[aggno].aggrefstate->args;
expression::AbstractExpression *agg_expr = nullptr;
if (arguments) {
GenericExprState *gstate =
(GenericExprState *)lfirst(list_head(arguments));
LOG_INFO("Creating Agg Expr");
agg_expr = ExprTransformer::TransformExpr(gstate->arg);
LOG_INFO("Done creating Agg Expr");
}
/*
* AggStatePerAggData.sortColIdx along with other related attributes
* are used to handle ORDER BY and DISTINCT *within* aggregation.
* E.g.,
* SELECT count(DISTINCT x) ...
* SELECT str_agg(y ORDER BY x) ...
* Currently, we only handle the agg(DISTINCT x) case by
* checking whether numDistinctCols > 0.
* Note that numDistinctCols > 0 may be a necessary but not sufficient
* condition for agg(DISTINCT x).
*/
bool distinct = (peragg[aggno].numDistinctCols > 0);
unique_agg_terms.emplace_back(fn_meta.exprtype, agg_expr, distinct);
LOG_INFO(
"Unique Agg # : %d , transfn_oid : %u\n , aggtype = %s \n expr = %s, "
"numDistinctCols = %d",
aggno, transfn_oid, ExpressionTypeToString(fn_meta.exprtype).c_str(),
agg_expr ? agg_expr->Debug().c_str() : "<NULL>",
peragg[aggno].numDistinctCols);
for (int i = 0; i < peragg[aggno].numDistinctCols; i++) {
LOG_INFO("sortColIdx[%d] : %d \n", i, peragg[aggno].sortColIdx[i]);
}
} // end loop aggno
/* Get Group by columns */
std::vector<oid_t> groupby_col_ids;
LOG_INFO("agg.numCols = %d", agg->numCols);
for (int i = 0; i < agg->numCols; i++) {
LOG_INFO("agg.grpColIdx[%d] = %d \n", i, agg->grpColIdx[i]);
auto attrno = agg->grpColIdx[i];
if (AttributeNumberIsValid(attrno) &&
AttrNumberIsForUserDefinedAttr(attrno)) {
groupby_col_ids.emplace_back(AttrNumberGetAttrOffset(attrno));
}
}
/* Get output schema */
std::unique_ptr<catalog::Schema> output_schema(
SchemaTransformer::GetSchemaFromTupleDesc(tupleDesc));
/* Map agg stragegy */
LOG_INFO("aggstrategy : %s\n", (AGG_HASHED == aggstrategy)
? "HASH"
: (AGG_SORTED ? "SORT" : "PLAIN"));
PelotonAggType agg_type = AGGREGATE_TYPE_INVALID;
switch (aggstrategy) {
case AGG_SORTED:
agg_type = AGGREGATE_TYPE_SORTED;
break;
case AGG_HASHED:
agg_type = AGGREGATE_TYPE_HASH;
break;
case AGG_PLAIN:
agg_type = AGGREGATE_TYPE_PLAIN;
break;
}
std::vector<oid_t> column_ids;
for (auto agg_term : unique_agg_terms) {
if (agg_term.expression) {
LOG_INFO("AGG TERM :: %s", agg_term.expression->Debug().c_str());
}
BuildColumnListFromExpr(column_ids, agg_term.expression);
}
auto retval = new planner::AggregatePlan(
proj_info.release(), predicate.release(), std::move(unique_agg_terms),
std::move(groupby_col_ids), output_schema.release(), agg_type);
((planner::AggregatePlan *)retval)->SetColumnIds(column_ids);
// Find children
auto lchild = TransformPlan(outerAbstractPlanState(plan_state));
retval->AddChild(lchild);
return retval;
}
/**
* @brief Convert a Postgres HashState into a Peloton HashPlanNode
* @return Pointer to the constructed AbstractPlan
*/
const planner::AbstractPlan *PlanTransformer::TransformHashJoin(
const HashJoinPlanState *hj_plan_state) {
planner::AbstractPlan *result = nullptr;
planner::HashJoinPlan *plan_node = nullptr;
PelotonJoinType join_type =
PlanTransformer::TransformJoinType(hj_plan_state->jointype);
if (join_type == JOIN_TYPE_INVALID) {
LOG_ERROR("unsupported join type: %d", hj_plan_state->jointype);
return nullptr;
}
LOG_INFO("Handle hash join with join type: %d", join_type);
/*
std::vector<planner::HashJoinPlan::JoinClause> join_clauses(
BuildHashJoinClauses(mj_plan_state->mj_Clauses,
mj_plan_state->mj_NumClauses);
*/
expression::AbstractExpression *join_filter = ExprTransformer::TransformExpr(
reinterpret_cast<ExprState *>(hj_plan_state->joinqual));
expression::AbstractExpression *plan_filter = ExprTransformer::TransformExpr(
reinterpret_cast<ExprState *>(hj_plan_state->qual));
expression::AbstractExpression *predicate = nullptr;
if (join_filter && plan_filter) {
predicate = expression::ExpressionUtil::ConjunctionFactory(
EXPRESSION_TYPE_CONJUNCTION_AND, join_filter, plan_filter);
} else if (join_filter) {
predicate = join_filter;
} else {
predicate = plan_filter;
}
/* Transform project info */
std::unique_ptr<const planner::ProjectInfo> project_info(nullptr);
project_info.reset(BuildProjectInfoFromTLSkipJunk(hj_plan_state->targetlist));
LOG_INFO("%s", project_info.get()->Debug().c_str());
auto project_schema = SchemaTransformer::GetSchemaFromTupleDesc(
hj_plan_state->tts_tupleDescriptor);
std::vector<oid_t> outer_hashkeys =
BuildColumnListFromExpStateList(hj_plan_state->outer_hashkeys);
if (project_info.get()->isNonTrivial()) {
// we have non-trivial projection
LOG_INFO("We have non-trivial projection");
result =
new planner::ProjectionPlan(project_info.release(), project_schema);
plan_node = new planner::HashJoinPlan(join_type, predicate, nullptr,
project_schema, outer_hashkeys);
result->AddChild(plan_node);
} else {
LOG_INFO("We have direct mapping projection");
plan_node =
new planner::HashJoinPlan(join_type, predicate, project_info.release(),
project_schema, outer_hashkeys);
result = plan_node;
}
const planner::AbstractPlan *outer =
PlanTransformer::TransformPlan(outerAbstractPlanState(hj_plan_state));
const planner::AbstractPlan *inner =
PlanTransformer::TransformPlan(innerAbstractPlanState(hj_plan_state));
/* Add the children nodes */
plan_node->AddChild(outer);
plan_node->AddChild(inner);
LOG_INFO("Finishing mapping Hash join, JoinType: %d", join_type);
return result;
}
