postgresql is crashing when selecting from citus columnar table

[vnoreika]

Hello,

we have a problem that postgresql is crashing when we left join citus table on heap tables.
OS Redhat 8.9 , Postgresql 16.3 , Citus 12.1.4

How to reproduce:

create extension citus;
create table test1 (a integer);
create table test2 (a integer);
create table test3 (b integer) using columnar;

SELECT *
   FROM test1 a1
     LEFT JOIN test2 a2 ON a1.a = a2.a
     LEFT JOIN test3 b  ON a2.a = b.b
;

server closed the connection unexpectedly

From logs:
<2024-07-09 10:00:21.484 EEST%2048391%668cdfda.1f4187%%%> DETAIL: Failed process was running: select *
FROM test1 a1
LEFT JOIN test2 a2 ON a1.a = a2.a
left JOIN test3 b ON a2.a = b.b
;
<2024-07-09 10:00:21.484 EEST%2048391%668cdfda.1f4187%%%> LOG: terminating any other active server processes
<2024-07-09 10:00:21.489 EEST%2048570%668ce005.1f423a%[unknown]%[local]%[unknown]> LOG: connection received: host=[local]
<2024-07-09 10:00:21.489 EEST%2048570%668ce005.1f423a%postgres%[local]%test> FATAL: the database system is in recovery mode
<2024-07-09 10:00:21.492 EEST%2048391%668cdfda.1f4187%%%> LOG: all server processes terminated; reinitializing

[copetol]

Hello,

as a workaround for this you can set columnar.enable_custom_scan to false

[Fenoman]

Absolutely the same problem! It would be great to fix this bug as soon as possible...

[OlgaSergeyevaB]

Hello,
It is enough to exclude outer_join_rels from candidateRelids in columnar_customscan.c FindCandidateRelids and it will work:

#if PG_VERSION_NUM >= PG_VERSION_16
	candidateRelids = bms_del_members(candidateRelids, root->outer_join_rels);	
#endif