Efficient latest record query Postgresql

Question

Alright I need to do a big query, but I only want the latest records.

For a single entry I would probably do something like

SELECT * FROM table WHERE id = ? ORDER BY date DESC LIMIT 1;

But I need to pull the latest records for a large (thousands of entries) number of records, but only the latest entry.

Here's what I have but It's not very efficient, I was wondering if there's a better way.

SELECT * FROM table a WHERE ID IN $LIST AND date = (SELECT max(date) FROM table b WHERE b.id = a.id);

Answer 1

this could be more eficient. Difference: query for table b is executed only 1 time, your correlated subquery is executed for every row:

SELECT * 
FROM table a 
JOIN (SELECT ID, max(date) maxDate
        FROM table
      GROUP BY ID) b
ON a.ID = b.ID AND a.date = b.maxDate
WHERE ID IN $LIST

Answer 2

If you have many rows per id's you definitely want a correlated subquery. It will make 1 index lookup per id, but this is faster than sorting the whole table.

Something like :

SELECT a.id, (SELECT max(t.date) FROM table t WHERE t.id = a.id) AS lastdate FROM table2;

The 'table2' you will use is not the table you mention in your query above, because here you need a list of distinct id's for good performance. Since your ids are probably FKs into another table, use this one.

Answer 3

On method - create a small derivative table containing the most recent update / insertion times on table a - call this table a_latest. Table a_latest will need sufficient granularity to meet your specific query requirements. In your case it should be sufficient to use

CREATE TABLE 
a_latest 
( id INTEGER NOT NULL, 
  date TSTAMP NOT NULL, 
  PRIMARY KEY (id, max_time) );

Then use a query similar to that suggested by najmeddine :

SELECT a.* 
FROM TABLE a, TABLE a_latest 
USING ( id, date );

The trick then is keeping a_latest up to date. Do this using a trigger on insertions and updates. A trigger written in plppgsql is fairly easy to write. I am happy to provide an example if you wish.

The point here is that computation of the latest update time is taken care of during the updates themselves. This shifts more of the load away from the query.

Answer 4

If you don't want to change your data model, you can use DISTINCT ON to fetch the newest record from table "b" for each entry in "a":

SELECT DISTINCT ON (a.id) *
FROM a
INNER JOIN b ON a.id=b.id
ORDER BY a.id, b.date DESC

If you want to avoid a "sort" in the query, adding an index like this might help you, but I am not sure:

CREATE INDEX b_id_date ON b (id, date DESC)

SELECT DISTINCT ON (b.id) *
FROM a
INNER JOIN b ON a.id=b.id
ORDER BY b.id, b.date DESC

Alternatively, if you want to sort records from table "a" some way:

SELECT DISTINCT ON (sort_column, a.id) *
FROM a
INNER JOIN b ON a.id=b.id
ORDER BY sort_column, a.id, b.date DESC

Alternative approaches

However, all of the above queries still need to read all referenced rows from table "b", so if you have lots of data, it might still just be too slow.

You could create a new table, which only holds the newest "b" record for each a.id -- or even move those columns into the "a" table itself.