Hi, Daniel
I run a similar environment and I've seen this problem a lot.
It's a java code problem -- your developers need to run a profiler to see where the resource leaks are -- they need to make sure that everytime they open a connection, they also close it as well -- and don't forget to to close the statements. We use borland's optimizeit which is not free (by a long shot) but it's probably worth the investment -- you can't promote code to production that is leaking resources like this. (Are you using connection pooling on tomcat?)
If they all shutdown their tomcat instances, it should release the connection, don't know how popular that suggestion would be though.
If you have the berkely version of ps (on solaris I use /usr/ucb/ps -auxww | grep post) you can see where each of these connections is coming from and some indication of what it's doing, and you can do a kill -15 on each one of them -- which is a more graceful recovery.
For a quick and dirty fix -- do pg_ctl stop -m i That's an immediate shutdown -- though it's still a "dirty" shutdown and the db must recover on startup, so make sure your backups and txlogs are good -- *don't* run out of room for the txlogs.
Daniel Rubio wrote:
Hi all
Our developers are making some kind of DoS on my postgres installation.
They use Tomcat to develop in java and they have an instance on every computer (only 3).
It seems that tomcat makes permanent connections to the configured database, but they have this number limited to 5, and max_connections in postgresql.conf is 40, and no more connections are made to the database because is a development environment.
The number of connections starts increasing and when arrives to 40 ( ps -ef | grep postm | wc -l ) the server doesn't accept more connections but it seems that neither finish these that are open, and no one can acces to the database (here is the DoS I say).
Here is what the log says
2005-02-07 10:13:26 LOG: connection received: host=127.0.0.1 port=55834 2005-02-07 10:13:26 LOG: connection authorized: user=test database=test 2005-02-07 10:13:26 FATAL: connection limit exceeded for non-superusers
I try to stop the postmaster with pg_ctl but it says that postmaster doesn't stops and I have to make kill -9 for the processes to stop :(
What could I make to solve this?
I've tried with:
statement_timeout = 100000
but nothing happens ... :(
---------------------------(end of broadcast)--------------------------- TIP 7: don't forget to increase your free space map settings
