Github user katameru commented on a diff in the pull request:
https://github.com/apache/phoenix/pull/281#discussion_r151620890
--- Diff:
phoenix-core/src/main/java/org/apache/phoenix/optimize/QueryOptimizer.java ---
@@ -91,8 +91,23 @@ public QueryPlan optimize(PhoenixStatement statement,
SelectStatement select, Co
}
public QueryPlan optimize(QueryPlan dataPlan, PhoenixStatement
statement, List<? extends PDatum> targetColumns, ParallelIteratorFactory
parallelIteratorFactory) throws SQLException {
- List<QueryPlan>plans = getApplicablePlans(dataPlan, statement,
targetColumns, parallelIteratorFactory, true);
- return plans.get(0);
+ List<QueryPlan> plans = getApplicablePlans(dataPlan, statement,
targetColumns, parallelIteratorFactory, false);
+ if (plans.size() == 1) {
+ return plans.get(0);
+ }
+
+ // Get the best plan based on their costs. Costs will be ZERO if
stats are not
+ // available, thus the first plan will be returned.
+ Cost minCost = null;
+ QueryPlan bestPlan = null;
+ for (QueryPlan plan : plans) {
+ Cost cost = plan.getCost();
+ if (minCost == null || cost.compareTo(minCost) < 0) {
+ minCost = cost;
+ bestPlan = plan;
+ }
+ }
+ return bestPlan;
--- End diff --
I'd argue that the observable behavior should stay the same, so a hinted
query plan should be used above all others. Currently we could just sidestep
the cost calculations in favor of a hinted plan, but I think that having a way
to influence the optimizer via hints could be very useful.
---
