Function index with multi-tenant infrastructure not working - postgresql

Question

I am currently working on a Postgresql DB where we implemented a multi-tenant infrastructure.
Here's how it works :
We have several tables (table1, tables2, ...) where we added a tenant column. This column is used to filter rows based on different DB users. We have several users (tenant1, tenant2, ...) and a superuser (no tenant applied to it).
I want to optimize the simple following query :
SELECT id
FROM table
WHERE UPPER("table"."column"::text) = UPPER('blablabla')
Thus, I created a function index :
CREATE INDEX "upper_idx" ON "table" (upper("column") );
If I connect to the DB as superuser and execute my SELECT query, it runs smoothly and fast.
Bitmap Heap Scan on table (cost=71.66..9225.47 rows=2998 width=4)
Recheck Cond: (upper((column)::text) = 'blablabla'::text)
-> Bitmap Index Scan on upper_idx (cost=0.00..70.91 rows=2998 width=0)
Index Cond: (upper((column)::text) = 'blablabla'::text)
However, when I connect as tenant1 the index is not picked up and the db runs a sequential scan instead :
Gather (cost=1000.00..44767.19 rows=15 width=4)
Workers Planned: 2
-> Parallel Seq Scan on table (cost=0.00..43765.69 rows=6 width=4)
Filter: (((tenant)::text = (CURRENT_USER)::text) AND (upper((column)::text) = 'blablabla'::text))
Do you know how to make it work in this case ?
EDIT - added EXPLAIN(ANALYZE, BUFFER)
Gather (cost=1000.00..44767.19 rows=15 width=4) (actual time=502.601..503.466 rows=0 loops=1)
Workers Planned: 2
Workers Launched: 2
Buffers: shared hit=781 read=36160
-> Parallel Seq Scan on table (cost=0.00..43765.69 rows=6 width=4) (actual time=498.978..498.978 rows=0 loops=3)
Filter: (((tenant)::text = (CURRENT_USER)::text) AND (upper((column)::text) = 'blablabla'::text))
Rows Removed by Filter: 199846
Buffers: shared hit=781 read=36160
Planning Time: 1.650 ms
Execution Time: 503.510 ms
EDIT 2 - added the (truncated) CREATE TABLE statement
-- public.table definition
-- Drop table
-- DROP TABLE public.table;
CREATE TABLE public.table (
id serial NOT NULL,
created timestamptz NOT NULL,
modified timestamptz NOT NULL,
...
column varchar(100) NULL,
"tenant" tenant NOT NULL,
...
);
...
CREATE INDEX upper_idx ON public.table USING btree (upper((column)::text), tenant);
CREATE INDEX table_column_91bdd18f ON public.table USING btree (column);
CREATE INDEX table_column_91bdd18f_like ON public.table USING btree (column varchar_pattern_ops);
...
-- Table Triggers
create trigger archive_deleted_rows after
delete
on
public.table for each row execute procedure archive.archive('{id}');
create trigger set_created_modified before
insert
or
update
on
public.table for each row execute procedure set_created_modified();
create trigger set_tenant before
insert
or
update
on
public.table for each row execute procedure set_tenant();
-- public.table foreign keys
...
EDIT 3 - dump of \d table
Table "public.table"
Column | Type | Collation | Nullable | Default
------------------------------------------+--------------------------+-----------+----------+--------------------------------------------
id | integer | | not null | nextval('table_id_seq'::regclass)
........
column | character varying(100) | | |
tenant | tenant | | not null |
........
Indexes:
"table_pkey" PRIMARY KEY, btree (id)
..........
"table__column_upper_idx" btree (upper(column::text), tenant)
"table_column_91bdd18f" btree (column)
"table_column_91bdd18f_like" btree (column varchar_pattern_ops)
.........
Check constraints:
.........
Foreign-key constraints:
.........
Referenced by:
.........
Policies:
POLICY "tenant_policy"
TO tenant1,tenant2
USING (((tenant)::text = (CURRENT_USER)::text))
Triggers:
........
set_tenant BEFORE INSERT OR UPDATE ON table FOR EACH ROW EXECUTE PROCEDURE set_tenant()
EDIT 4 - Added tenant data type
CREATE TYPE tenant AS ENUM (
'tenant1',
'tenant2');

Answer 1

You should add a multi-column index like whis:
CREATE INDEX "upper_column_for_tenant_idx" ON "table" (upper("column") , tenant);
In order to have only one index you should place upper("column") first and then tenant.
PostgreSQL docs state:
A multicolumn B-tree index can be used with query conditions that involve any subset of the index's columns, but the index is most efficient when there are constraints on the leading (leftmost) columns.
I've tried to recreate your setup in db<>fiddle. you can see the plan for query
EXPLAIN ANALYZE
SELECT * FROM public.table WHERE upper(("column")::text) = '50' and tenant='5'
is:
QUERY PLAN
Bitmap Heap Scan on "table" (cost=14.20..2333.45 rows=954 width=24) (actual time=0.093..0.831 rows=1000 loops=1)
Recheck Cond: ((upper(("column")::text) = '50'::text) AND ((tenant)::text = '5'::text))
Heap Blocks: exact=74
-> Bitmap Index Scan on upper_idx (cost=0.00..13.97 rows=954 width=0) (actual time=0.065..0.065 rows=1000 loops=1)
Index Cond: ((upper(("column")::text) = '50'::text) AND ((tenant)::text = '5'::text))
Planning Time: 0.361 ms
Execution Time: 0.997 ms

Answer 2

You should create an index on both columns. Since tenant is an enum data type and you compare it to a function result of type name, both sides are cast to the "common denominator" text. So use this index:
CREATE INDEX ON "table" (upper("column"), tenant::text);
Then calculate new statistics for the table:
ANALYZE "table";