How to optimize this simple yet slow query - sql

Question

I have a relatively simple table with several columns, two of them are expires_at(Date) and museum_id(BIGINT, FOREIGN). Both indexed, also using a compound index. The table contains around 3 million of rows in it.
Running query as simple as this takes around 90 seconds to complete:
SELECT *
FROM external_users
WHERE museum_id = 356
AND ((expires_at > '2022-02-16 07:35:39.818117') OR expires_at IS NULL)
Here is the explain analyze output:
Bitmap Heap Scan on external_users (cost=2595.76..148500.40 rows=59259 width=1255) (actual time=4901.257..90786.702 rows=94272 loops=1)
Recheck Cond: (((museum_id = 356) AND (expires_at > '2022-02-16'::date)) OR ((museum_id = 356) AND (expires_at IS NULL)))
Rows Removed by Index Recheck: 391889
Heap Blocks: exact=34133 lossy=33698
-> BitmapOr (cost=2595.76..2595.76 rows=63728 width=0) (actual time=4671.804..4671.806 rows=0 loops=1)
-> Bitmap Index Scan on index_external_users_on_museum_id_and_expires_at (cost=0.00..2187.79 rows=54336 width=0) (actual time=1229.564..1229.564 rows=33671 loops=1)
Index Cond: ((museum_id = 356) AND (expires_at > '2022-02-16'::date))
-> Bitmap Index Scan on index_external_users_on_museum_id_and_expires_at (cost=0.00..378.34 rows=9391 width=0) (actual time=3442.238..3442.238 rows=64337 loops=1)
Index Cond: ((museum_id = 356) AND (expires_at IS NULL))
Planning Time: 266.470 ms
Execution Time: 90838.777 ms
I can't really see anything helpful in the explain/analyze output but that might be related to my lack of experience in such. My peer-reviewer also didn't saw anything interesting in there which makes me think - is there anything i can do in order to help postgres handle queries like that faster or is it just the way it is for tables with over 3M records?

Answer 1

I will explain to you some rules, ways to optimize this query.
1 - When you use OR command on the where conditions, then DB can not use indexes. Recommended using union all. Example:
select *
from external_users
where museum_id = 356
and expires_at > '2022-02-16 07:35:39.818117'
union all
select *
from external_users
where museum_id = 356
and expires_at is null
2 - Your expires_at field may be a timestamp type. But, date types are faster than timestamp types. Because in timestamp types stored hours, minutes, seconds. Also, indexing size timestamp types will be greater than date type indexing size. If you need to store full datetime, then you can use casting types. For the best performance you must create a function-based index (on PostgreSQL this is called expression index), but not a standard index.
select *
from external_users
where museum_id = 356
and expires_at::date > '2022-02-16'
union all
select *
from external_users
where museum_id = 356
and expires_at is NULL
/*
We must cast `expires_at` field type to date type during creating the indexing process. Because in our query we use casting this type, so we must create an index via casting data.
*/
create index external_users_expires_at_idx
ON external_users USING btree ((expires_at::date));
3 - In where conditions if you are using always two, three fields samely, recommended creating one index for these fields, but not separately. In your query maybe always use museum_id and expires_at fields. Create index sample code:
create index external_users_full_index on external_users using btree (museum_id, (expires_at ::date));
The most important of all these ways is the first rule, so not using OR command.