> If it shows as "zombie" in ps, then you can't kill it (because it's
> already dead). kill will have absolutely zero effect.
>
> This should not happen, though. The only way a Postgres child process
> could stay as "zombie" for more than a few milliseconds is if the
> postmaster is hung up and unable to notice the dead child process.
> If the postmaster were hung, you'd have a lot worse problems than zombie
> processes --- you'd not be able to make new connections at all. So I
> think you aren't giving us the straight scoop. Are you sure that these
> zombie processes are actually Postgres processes? Look to see what
> their parent process is.
Well, I'm saying 'zombie' not like *nix terminology. I'm talking about
server proces that are waiting for a client that die unexpectly and can't
close connection.
> > PostgreSQL 7.1.3 under freeBSD
>
> You could do worse than update to 7.2. It's *much* harder for a
> misbehaving client to hang the postmaster than it was pre-7.2.
> But I'm not convinced that you're describing a Postgres problem.
>
> regards, tom lane
No, no sorry, Postgres is HARDER a lot, we have more than 70 simultaneos
connection over ODBC, 107 tables, millions of rows and nothing has happened
since april, really good job ! Update the database is in our TODO LIST but
we try to wait for a version with native replication (We have heard that is
possible to be available in 7.4.X ) . We have still a lot of work and there
is a problem with time representation that has changed between 7.1.X to
7.2.X that we can't resolve (decimal fraction of time) and we have a lot of
server side code. Thanks!
Simeó Reig