On 29.11.2011 23:38, Merlin Moncure wrote: > On Tue, Nov 29, 2011 at 7:49 AM, Heiko Wundram <modelnine@xxxxxxxxxxxxx> wrote: >> Hello! >> >> Sorry for that subscribe post I've just sent, that was bad reading on my >> part (for the subscribe info on the homepage). >> >> Anyway, the title says it all: is there any possibility to limit the number >> of connections that a client can have concurrently with a PostgreSQL-Server >> with "on-board" means (where I can't influence which user/database the >> clients use, rather, the clients mostly all use the same user/database, and >> I want to make sure that a single client which runs amok doesn't kill >> connectivity for other clients)? I could surely implement this with a proxy >> sitting in front of the server, but I'd rather implement this with >> PostgreSQL directly. >> >> I'm using (and need to stick with) PostgreSQL 8.3 because of the frontend >> software in question. >> >> Thanks for any hints! > > I think the (hypothetical) general solution for these types of > problems is to have logon triggers. It's one of the (very) few things > I envy from SQL Server -- see here: > http://msdn.microsoft.com/en-us/library/bb326598.aspx. I'd like to have logon triggers too, but I don't think that's the right solution for this problem. For example the logon triggers would be called after forking the backend, which means overhead. The connection limits should be checked when creating the connection (validation username/password etc.), before creating the backend. Anyway, I do have an idea how this could be done using a shared library (so it has the same disadvantages as logon triggers). Hopefully I'll have time to implement a PoC of this over the weekend. > Barring the above, if you can trust the client to call a function upon > connection I'd just do that and handle the error on the client with a > connection drop. Barring *that*, I'd be putting my clients in front of > pgbouncer with some patches to the same to get what I needed > (pgbouncer is single threaded making firewally type features quite > easy to implement in an ad hoc fashion). The connection pooler somehow easier and more complex at the same time. You can use connect_query to execute whatever you want after connecting to the database (not trusting the user to do that), but why would you do that? But the database will see the IP of the pgbouncer, not the IP of the original client. So executing the query is pointless. You can modify pgbouncer and it should be quite simple, but you can achieve different username/password (pgbouncer) to each customer, different database, set pool_size for each of the connections. It won't use IP to count connections, but the user's won't 'steal' connections from the other. Tomas -- Sent via pgsql-general mailing list (pgsql-general@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general
