PostgreSQL equivalent for TOP n WITH TIES: LIMIT "with ties"?

Solution 1:

Postgres 13 finally adds WITH TIES . See:

• Greater than or equal to ALL() and equal to MAX() speed

There is no WITH TIES clause up to PostgreSQL 12, like there is in SQL Server.
In PostgreSQL I would substitute this for TOP n WITH TIES .. ORDER BY <something>:

WITH cte AS (
   SELECT *, rank() OVER (ORDER BY <something>) AS rnk
   FROM   tbl
   )
SELECT *
FROM   cte
WHERE  rnk <= n;

To be clear, rank() is right, dense_rank() would be wrong (return too many rows).
Consider this quote from the SQL Server docs (from the link above):

For example, if expression is set to 5 but 2 additional rows match the values of the ORDER BY columns in row 5, the result set will contain 7 rows.

The job of WITH TIES is to include all peers of the last row in the top n as defined by the ORDER BY clause. rank() gives the exact same result.

To make sure, I tested with SQL Server, here is a live demo.

db<>fiddle here

Faster alternatives for big tables in Postgres 12 or older:

• Equivalent for FETCH FIRST WITH TIES in PostgreSQL 11 with comparable performance

Solution 2:

Try this:

Output: 10, 9, 8, 8

with numbers (nums) as (
  values (10), (9), (8), (8), (2)
) 
SELECT nums FROM Numbers 
WHERE nums in (SELECT DISTINCT nums FROM Numbers ORDER BY nums DESC LIMIT 3)
ORDER BY nums DESC

Output: 10,10,9,8,8

with numbers (nums) as (
  values (10), (9), (8), (8), (2), (10)
) 
SELECT nums FROM Numbers 
WHERE nums in (SELECT DISTINCT nums FROM Numbers ORDER BY nums DESC LIMIT 3)
ORDER BY nums DESC