optimizing a join against a windowed function

[James Brown <james@xxxxxxxxxxxxxxx>]

Hello:

I'm attempting to figure out whether an optimizer behavior I'm seeing is a PostgreSQL bug or expected behavior. The scenario:

I have two tables: one named taxpayers which has a goodish number of columns an an integer PK id, and one named insights, which has a taxpayer_id foreign key to taxpayers, a year, and (again) a lot of other columns. There's an index on insights (taxpayer_id, year DESC). I'm executing the following SQL:

```
SELECT taxpayers.id, insight_id
   FROM taxpayers
   JOIN (
       WITH ordered_insights AS (
           SELECT taxpayer_id, id, RANK() OVER (PARTITION BY taxpayer_id ORDER BY year DESC) AS rank
           FROM insights
           WHERE year IS NOT NULL
       )
       SELECT taxpayer_id, id AS insight_id
       FROM ordered_insights
       WHERE rank = 1
   ) latest_insights ON latest_insights.taxpayer_id = taxpayers.id
WHERE taxpayers.id IN (?, ?)
```

(this is simplified example; the real code has the subselect in a view so that it can execute this kind of join from an ORM; it also joins quite a few tables downstream after this)

If there's only a single value in the IN clause, the EXPLAIN plan looks great:

Nested Loop  (cost=0.86..53.30 rows=1 width=16)
  ->  Index Only Scan using taxpayers_pkey on taxpayers  (cost=0.43..8.45 rows=1 width=8)
        Index Cond: (id = 650974)
  ->  Subquery Scan on ordered_insights  (cost=0.43..44.83 rows=1 width=16)
        Filter: (ordered_insights.rank = 1)
        ->  WindowAgg  (cost=0.43..44.71 rows=10 width=28)
              Run Condition: (rank() OVER (?) <= 1)
              ->  Index Scan using index_insights_on_taxpayer_id_year_desc on insights  (cost=0.43..44.53 rows=10 width=20)
                    Index Cond: ((taxpayer_id = 650974) AND (year IS NOT NULL))
(9 rows)

However, if there are multiple rows in the IN clause, the optimizer decides to execute the subselect against the entire giant table, and it is not great:

Hash Join  (cost=2611586.97..2800201.15 rows=1 width=16)
  Hash Cond: (ordered_insights.taxpayer_id = taxpayers.id)
  ->  Subquery Scan on ordered_insights  (cost=2611570.10..2799818.65 rows=28961 width=16)
        Filter: (ordered_insights.rank = 1)
        ->  WindowAgg  (cost=2611570.10..2727415.36 rows=5792263 width=28)
              Run Condition: (rank() OVER (?) <= 1)
              ->  Sort  (cost=2611570.10..2626050.76 rows=5792263 width=20)
                    Sort Key: insights.taxpayer_id, insights.year DESC
                    ->  Seq Scan on insights  (cost=0.00..1723354.01 rows=5792263 width=20)
                          Filter: (year IS NOT NULL)
  ->  Hash  (cost=16.85..16.85 rows=2 width=8)
        ->  Index Only Scan using taxpayers_pkey on taxpayers  (cost=0.43..16.85 rows=2 width=8)
              Index Cond: (id = ANY ('{650974,243848}'::bigint[]))

If I add in a second repetitive WHERE clause, it goes back to being happy, but that feels a bit like a hack:

# EXPLAIN SELECT taxpayers.id, insight_id
   FROM taxpayers
   JOIN (
       WITH ordered_insights AS (
           SELECT taxpayer_id, id, RANK() OVER (PARTITION BY taxpayer_id ORDER BY year DESC) AS rank
           FROM insights
           WHERE year IS NOT NULL
       )
       SELECT taxpayer_id, id AS insight_id
       FROM ordered_insights
       WHERE rank = 1
   ) latest_insights ON latest_insights.taxpayer_id = taxpayers.id
WHERE taxpayers.id IN (650974, 243848) AND latest_insights.taxpayer_id IN (650974, 243848);
                                                        QUERY PLAN                                                          
-----------------------------------------------------------------------------------------------------------------------------
Nested Loop  (cost=0.86..110.57 rows=1 width=16)
  Join Filter: (taxpayers.id = ordered_insights.taxpayer_id)
  ->  Subquery Scan on ordered_insights  (cost=0.43..93.69 rows=1 width=16)
        Filter: (ordered_insights.rank = 1)
        ->  WindowAgg  (cost=0.43..93.42 rows=21 width=28)
              Run Condition: (rank() OVER (?) <= 1)
              ->  Index Scan using index_insights_on_taxpayer_id_year_desc on insights  (cost=0.43..93.06 rows=21 width=20)
                    Index Cond: ((taxpayer_id = ANY ('{650974,243848}'::bigint[])) AND (year IS NOT NULL))
  ->  Index Only Scan using taxpayers_pkey on taxpayers  (cost=0.43..16.85 rows=2 width=8)
        Index Cond: (id = ANY ('{650974,243848}'::bigint[]))

This feels like a bug to me, but maybe I'm missing something obvious. I don't really get why the optimizer wouldn't be able to infer the second condition given that I'm doing a join on a non-nullable integer column (so there's no NaN nonsense to worry about), but maybe I'm missing something obvious.

I've reproduced this on PostgreSQL 15.7 and 17beta3.

Thanks for any insights y'all can provide!

-- 
James Brown