For those who are interested, here is the solution I came up with, inspired from Craig Ringer's comment:
(...) use a cron job to look at when the connection was last active (see pg_stat_activity) and use pg_terminate_backend to kill old ones.(...)
The chosen solution comes down like this:
- First, we upgrade to Postgresql 9.2.
- Then, we schedule a thread to run every second.
- When the thread runs, it looks for any old inactive connections.
- A connection is considered inactive if its state is either
idle
, idle in transaction
, idle in transaction (aborted)
or disabled
.
- A connection is considered old if its state stayed the same during more than 5 minutes.
- There are additional threads that do the same as above. However, those threads connect to the database with different user.
- We leave at least one connection open for any application connected to our database. (
rank()
function)
This is the SQL query run by the thread:
WITH inactive_connections AS (
SELECT
pid,
rank() over (partition by client_addr order by backend_start ASC) as rank
FROM
pg_stat_activity
WHERE
-- Exclude the thread owned connection (ie no auto-kill)
pid <> pg_backend_pid( )
AND
-- Exclude known applications connections
application_name !~ '(?:psql)|(?:pgAdmin.+)'
AND
-- Include connections to the same database the thread is connected to
datname = current_database()
AND
-- Include connections using the same thread username connection
usename = current_user
AND
-- Include inactive connections only
state in ('idle', 'idle in transaction', 'idle in transaction (aborted)', 'disabled')
AND
-- Include old connections (found with the state_change field)
current_timestamp - state_change > interval '5 minutes'
)
SELECT
pg_terminate_backend(pid)
FROM
inactive_connections
WHERE
rank > 1 -- Leave one connection for each application connected to the database
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…