Обсуждение: RE: [QUESTIONS] Identity Crisis

Forwarded to HACKERS.

> -----Original Message-----
> From:    Len Morgan [SMTP:len-morgan@xroadstx.com]
> Sent:    Saturday, May 02, 1998 8:21 AM
> To:    pgsql-questions@postgreSQL.org
> Subject:    [QUESTIONS] Identity Crisis
>
> I am running 6.3.2 at one of my client's sites that normally has 5
> users
> connected.  The problem I run into is that after a couple of days, I
> end
> up with as many as 12-15 postgres backends running on the server.  The
> users are encountering a bug in the code and then closing the window
> and
> restarting the program which never closes down the connection to the
> back end.  My question is this: Is there a way to identify on the line
> which starts up the backends, which host is starting it?  In other
> words, if I do ps -ax on the server, can I have an ip address or host
> name show up to let me know which host is connected to which backend?
> This will not entirely solve my problem, but it will let me know which
> user I need to educate about the proper way to work around the program
> bug.  Currently, I just kill the processes off one by one.  If
> somebody
> calls me in the next five minutes or so, it was a "live" one.
> Otherwise, nobody missed it.  Also, if I send -SIGTERM to the
> individual
> backend's pid, will this properly clean up any memory that is in use?
>
> Thanks,
>
> len morgan
>
> --
> Official WWW Site: http://www.postgresql.org
> Online Docs & FAQ: http://www.postgresql.org/docs
> Searchable Lists: http://www.postgresql.org/mhonarc

Len Morgan writes:
>> My question is this: Is there a way to identify on the line
>> which starts up the backends, which host is starting it?  In other
>> words, if I do ps -ax on the server, can I have an ip address or host
>> name show up to let me know which host is connected to which backend?

Hackers, if anyone does something with setting the backend process title,
this seems like a good idea to me.

However, a feature that may or may not show up in 6.4 is not going to
help Len with his immediate problem.  Len, I'd suggest a couple of
things you can do today:

1. netstat on your server will show open TCP connections.  Look for
   connections to port 5432 at your end.  If you have a lot of users
   it might be hard to spot the culprit --- but I suspect that looking
   for the machine that shows a number of open connections, not just
   one, will do it.

2. If that doesn't work, but you can identify a backend process that's
   been laying around for awhile, you can use "lsof" to find out
   which network connection leads to that process.  For example,
   I use ps to find that process 21309 is a backend, then:

$ lsof -p 21309
COMMAND    PID     USER   FD   TYPE     DEVICE SIZE/OFF  INODE NAME
postgres 21309 postgres    3u  inet 0x0d2d1b00      0t0    TCP *:5432 (LISTEN)
postgres 21309 postgres    5u  inet 0x0d749000    0t302    TCP localhost:5432->localhost:2325 (ESTABLISHED)
postgres 21309 postgres    6u  inet 0x0d749000    0t302    TCP localhost:5432->localhost:2325 (ESTABLISHED)
(lots of non-inet open files for this process snipped)

So I see the client connected to this server is at port 2325 on
localhost.

lsof (list open files) might already be installed on your machine,
if not see http://www-rcd.cc.purdue.edu/abe/.  It's an invaluable
tool for debugging all sorts of Unix problems, well worth having
in your sysadmin kit.

            regards, tom lane