With yesterday’s release of PostgreSQL 11.15, 12.10, and 13.6
(presumably 10.20 and 14.2 as well), Zulip’s test suite started failing
with “variable not found in subplan target list” errors from PostgreSQL
on a table that has a PGroonga index. I found the following
reproduction recipe from a fresh database:
psql (11.15 (Debian 11.15-1.pgdg100+1))
Type "help" for help.
test=# CREATE EXTENSION pgroonga;
CREATE EXTENSION
test=# CREATE TABLE t AS SELECT CAST(c AS text) FROM generate_series(1,
10000) AS c;
SELECT 10000
test=# CREATE INDEX t_c ON t USING pgroonga (c);
CREATE INDEX
test=# VACUUM t;
VACUUM
test=# SELECT COUNT(*) FROM t;
ERROR: variable not found in subplan target list
I filed https://github.com/pgroonga/pgroonga/issues/203, and confirmed
with a Git bisection of PostgreSQL that this error first appears with
ec36745217 (REL_11_15~42) “Fix index-only scan plans, take 2.” I’m
aware that this likely just exposed a previously hidden PGroonga bug,
but I figure PostgreSQL developers might want to know about this anyway
and help come up with the right fix. The PGroonga author suggested I
start a thread here:
https://github.com/pgroonga/pgroonga/issues/203#issuecomment-1036708841
Thanks for investigating this!
Our CI is also broken with new PostgreSQL:
https://github.com/pgroonga/pgroonga/runs/5149762901?check_suite_focus=true
https://www.postgresql.org/message-id/602391641208390%40iva4-92c901fae84c.qloud-c.yandex.net
says partial-retrieval index-only scan but our case is
non-retrievable index-only scan. In non-retrievable index-only scan,
the error is occurred.
We asked about non-retrievable index-only scan on the PostgreSQL
mailing list in the past:
https://www.postgresql.org/message-id/5148.1584372043%40sss.pgh.pa.us
We thought non-retrievable index-only scan should not be used but
PostgreSQL may use it as a valid plan. So I think that our case
should be supported with postgres/postgres@ec36745 “Fix index-only > scan
plans, take 2.”
Could you ask about this case on the PostgreSQL mailing list
https://www.postgresql.org/list/pgsql-hackers/ ?
The following patch fixes our case and PostgreSQL's test cases are
still passed but a review by the original author is needed:
--- postgresql-11.15.orig/src/backend/optimizer/plan/setrefs.c 2022-02-08
06:20:23.000000000 +0900
+++ postgresql-11.15/src/backend/optimizer/plan/setrefs.c 2022-02-12
07:32:20.355567317 +0900
@@ -1034,7 +1034,7 @@
{
TargetEntry *indextle = (TargetEntry *) lfirst(lc);
- if (!indextle->resjunk)
+ if (!indextle->resjunk || indextle->expr->type == T_Var)
stripped_indextlist = lappend(stripped_indextlist,
indextle);
}
```
Anders
