Wrapping the query in a derived table is the obvious way to solve this:

SELECT *
FROM
    ( SELECT DISTINCT ON (item_id)
             item_id, data, timestamp, sort_order, ... 
        FROM [this table]
       WHERE group_id = [some value]
         AND timestamp < [some value]
    ORDER BY item_id, timestamp DESC
   ) AS t
ORDER BY sort_order ;

If your original query is efficient and doesn't return too many rows, the additional sort will not add much cost to the query.

Besides DISTINCT ON, there are a few other methods that are often more efficient in this type of "greatest-n-pre-group" queries - but efficiency depends on may parameters, like targeted indexes and distribution of values (how many distinct item_id in the table, how many rows per item_id, etc).

Most notable method is using LATERAL subqueries. See these excellent answers by Erwin:

• Optimize GROUP BY query to retrieve latest record per user

• Optimize groupwise maximum query

• Select first row in each GROUP BY group?

• Query last N related rows per row

• Optimise a LATERAL JOIN query on a big table

Wrapping the query in a derived table is the obvious way to solve this:

SELECT *
FROM
    ( SELECT DISTINCT ON (item_id)
             item_id, data, timestamp, sort_order, ... 
        FROM [this table]
       WHERE group_id = [some value]
         AND timestamp < [some value]
    ORDER BY item_id, timestamp DESC
   ) AS t
ORDER BY sort_order ;

If your original query is efficient and doesn't return too many rows, the additional sort will not add much cost to the query.

Besides DISTINCT ON, there are a few other methods that are often more efficient in this type of "greatest-n-pre-group" queries - but efficiency depends on many parameters, like targeted indexes and distribution of values (how many distinct item_id in the table, how many rows per item_id), Postgres version, etc.

Most notable method is using LATERAL subqueries. See these excellent answers by Erwin Brandstetter:

• Optimize GROUP BY query to retrieve latest record per user

• Optimize groupwise maximum query

• Select first row in each GROUP BY group?

• Query last N related rows per row

• Optimise a LATERAL JOIN query on a big table

Wrapping the query in a derived table is the obvious way to solve this:

SELECT *
FROM
    ( SELECT DISTINCT ON (item_id)
             item_id, data, timestamp, sort_order, ... 
        FROM [this table]
       WHERE group_id = [some value]
         AND timestamp < [some value]
    ORDER BY item_id, timestamp DESC
   ) AS t
ORDER BY sort_order ;

If your original query is efficient and doesn't return too many rows, the additional sort will not add much cost to the query.

Besides DISTINCT ON, there are a few other methods that are often more efficient in this type of "greatest-n-pre-group" queries. Most notable is using LATERAL subqueries. See this excellent answers by Erwin:

• Optimize GROUP BY query to retrieve latest record per user

Wrapping the query in a derived table is the obvious way to solve this:

SELECT *
FROM
    ( SELECT DISTINCT ON (item_id)
             item_id, data, timestamp, sort_order, ... 
        FROM [this table]
       WHERE group_id = [some value]
         AND timestamp < [some value]
    ORDER BY item_id, timestamp DESC
   ) AS t
ORDER BY sort_order ;

If your original query is efficient and doesn't return too many rows, the additional sort will not add much cost to the query.

Besides DISTINCT ON, there are a few other methods that are often more efficient in this type of "greatest-n-pre-group" queries - but efficiency depends on may parameters, like targeted indexes and distribution of values (how many distinct item_id in the table, how many rows per item_id, etc). 

Most notable method is using LATERAL subqueries. See these excellent answers by Erwin:

• Optimize GROUP BY query to retrieve latest record per user

• Optimize groupwise maximum query

• Select first row in each GROUP BY group?

• Query last N related rows per row

• Optimise a LATERAL JOIN query on a big table