Postgresql: Query 10x Slower In A Different Client

Looking at the postgres server log, I see that the exact same query on the same postgres server takes much longer (about 10x longer) when invoked from a Linux client or from a Wind

Solution 1:

You may want to check if the slow client does SSL encryption or not. It happens by default when it's set up on the server and the client has been compiled with SSL support.

For queries that retrieve large amounts of data, the time difference is significant. Also some Linux distributions like Debian/Ubuntu have SSL on by default, even for TCP connections through localhost.

As an example, here's the time difference for a query retrieving 1,5M rows weighing a total of 64Mbytes, with a warm cache.

Without encryption:

$ psql "host=localhost dbname=mlists sslmode=disable"
Password: 
psql (9.1.7, server 9.1.9)
Type "help" for help.

mlists=> \timing
Timing is on.
mlists=> \o /dev/null
mlists=> select subject from mail;
Time: 1672.258 ms

With encryption:

$ psql "host=localhost dbname=mlists"
Password: 
psql (9.1.7, server 9.1.9)
SSL connection (cipher: DHE-RSA-AES256-SHA, bits: 256)
Type "help" for help.

mlists=> \o /dev/null
mlists=> \timing
Timing is on.
mlists=> select subject from mail;
Time: 7017.935 ms

To turn it off globally, one might set SSL=off in postgresql.conf.

To turn it off for specific ranges of client addresses, add entries in pg_hba.conf with hostnossl in the first field before the more generic host entries.

To turn if off client-side, it depends on how the driver exposes the sslmode connection parameter. If it doesn't, the PGSSLMODE environment variable may be used if the driver is implemented on top of libpq.

As for connections through Unix domain sockets (local), SSL is never used with them.