Github user zuyu commented on a diff in the pull request:

    https://github.com/apache/incubator-quickstep/pull/45#discussion_r68846898
  
    --- Diff: query_optimizer/ExecutionGenerator.cpp ---
    @@ -1575,21 +1579,12 @@ void ExecutionGenerator::convertSort(const 
P::SortPtr &physical_sort) {
       
merged_runs_destination_proto->set_relational_op_index(merge_run_operator_index);
       
sorted_output_destination_proto->set_relational_op_index(merge_run_operator_index);
     
    -  // Do not add merged_runs_relation into 'temporary_relation_info_vec_'
    -  // and create the DropTableOperator for it at the end. Instead, add the 
drop
    -  // operator right here, because the relation won't be used by any other 
operator.
    -  const QueryPlan::DAGNodeIndex drop_merged_runs_index =
    -      execution_plan_->addRelationalOperator(
    -          new DropTableOperator(
    -              query_handle_->query_id(),
    -              *merged_runs_relation,
    -              optimizer_context_->catalog_database(),
    -              false /* only_drop_blocks */));
    -  execution_plan_->addDirectDependency(
    -      drop_merged_runs_index,
    -      merge_run_operator_index,
    -      true /* is_pipeline_breaker */);
    -
    +  // REVIEW(Shixuan): There might be an issue with the original code: if a 
throw
    +  // happens before the plan is executed, merged_runs_relation will not be 
dropped
    +  // by dropAllTemporaryRelations() if not added to 
temporary_relation_info_vec,
    +  // which might cause "RelationNameCollision".
    +  temporary_relation_info_vec_.emplace_back(merge_run_operator_index,
    --- End diff --
    
    I think the actual executing order of `DropTable` depends on the scheduler 
based on a topological order. But in the optimizer generated DAG, a `sort` 
operator typically is placed as the last one, so in most case the new way is 
similar to the old way.
    
    I suggest to debug the query w/ `window aggregation` by printing out the 
DAG and the operator finish order (hack in `query_execution/QueryManager`).


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---

Reply via email to