Fix the way we check for local/reference table joins in the executor

[pykello]

Fixes #3279

Previously we used plannedStmt->releationOids which is the set of relations query depends on to check for the relations "involved" in the query. But this is not correct. For example, citus plans query of the following form in the citus_drop_trigger() to have relationOids = [pg_class, reference_table], so we assumed it is a join between reference table and local table.

SELECT master_unmark_object_distributed(v_obj.classid, v_obj.objid, v_obj.objsubid)

This PR changes the behaviour to use the range table entries instead.

[marcocitus]

Does this contain RTEs for the entire query? (incl. CTEs, subqueries)

[pykello]

Yes, standard_planner stores the flattened rte lists in plannedStmt->rtable. set_plan_references concatenates all these lists in glob->finalrtable, and then standard_planner does plannedStmt->rtable = glob->finalrtable.

Added a comment.

[marcocitus]

why return false here? What if we a join between a local table, a reference table and a function?

[pykello]

To match distributed_planner.c's IsLocalReferenceTableJoin(). Joins between local, reference, and functions will error out in planner. We can remove that restriction in both planner and executor, but maybe in another PR.

Removing the check here, although I think it won't change any behaviour.

[marcocitus]

if I try without any works (just coordinator), I get:

postgres=# ALTER TABLE ref ADD COLUMN z int;
WARNING:  could not find any shard placements for shardId 102040
... stuck forever

[marcocitus]

if I try without any works (just coordinator), I get:

that actually seems to be a separate bug unrelated to having the coordinator in the metadata

[codecov]

Codecov Report

Merging #3280 into master will increase coverage by 0.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #3280      +/-   ##
==========================================
+ Coverage   92.04%   92.05%   +0.01%     
==========================================
  Files         163      163              
  Lines       32865    32873       +8     
==========================================
+ Hits        30250    30262      +12     
+ Misses       2615     2611       -4

[pykello]

if I try without any works (just coordinator), I get:

that actually seems to be a separate bug unrelated to having the coordinator in the metadata

I can look into this tomorrow.

[marcocitus]

Since views are RTE_RELATION they are also treated as local tables.

CREATE VIEW v AS SELECT * FROM ref WHERE a = 1;
BEGIN;
SELECT * FROM v JOIN ref ON (v.a = ref.a);
ERROR:  cannot join local tables and reference tables in a transaction block, udf block, or distributed CTE subquery

[pykello]

Fixed + tests. Also added tests and comments for the function join.