11.1. Determining Disk Usage

Each table has a primary heap disk file where most of the data is stored. To store long column values, there is also a TOAST file associated with the table, named based on the table's OID (actually pg_class.relfilenode), and an index on the TOAST table. There also may be indexes associated with the base table.

You can monitor disk space from three places: from psql using VACUUM information, from psql using contrib/dbsize, and from the command line using contrib/oid2name. Using psql on a recently vacuumed (or analyzed) database, you can issue queries to see the disk usage of any table:

play=# SELECT relfilenode, relpages
play-# FROM pg_class
play-# WHERE relname = 'customer';
 relfilenode | relpages 
-------------+----------
       16806 |       60
(1 row)

Each page is typically 8 kilobytes. (Remember, relpages is only updated by VACUUM and ANALYZE.) To show the space used by TOAST tables, use a query based on the heap relfilenode shown above:

play=# SELECT relname, relpages
play-# FROM pg_class
play-# WHERE relname = 'pg_toast_16806' OR
play-#       relname = 'pg_toast_16806_index'
play-# ORDER BY relname;
       relname        | relpages 
----------------------+----------
 pg_toast_16806       |        0
 pg_toast_16806_index |        1

You can easily display index usage too:

play=# SELECT c2.relname, c2.relpages
play-# FROM pg_class c, pg_class c2, pg_index i
play-# WHERE c.relname = 'customer' AND
play-#       c.oid = i.indrelid AND
play-#       c2.oid = i.indexrelid
play-# ORDER BY c2.relname;
       relname        | relpages 
----------------------+----------
 customer_id_indexdex |       26

It is easy to find your largest files using psql:

play=# SELECT relname, relpages
play-# FROM pg_class
play-# ORDER BY relpages DESC;
       relname        | relpages 
----------------------+----------
 bigtable             |     3290
 customer             |     3144

contrib/dbsize loads functions into your database that allow you to find the size of a table or database from inside psql without the need for VACUUM/ANALYZE.

You can also use contrib/oid2name to show disk usage. See README.oid2name for examples. It includes a script that shows disk usage for each database.