[Spacewalk-list] Spacewalk 1.7 Postgres out of connections

[Paul Robert Marino]

here is what to look at in the postgresql.conf

# - TCP Keepalives -
# see "man 7 tcp" for details

#tcp_keepalives_idle = 0                # TCP_KEEPIDLE, in seconds;
                                        # 0 selects the system default
#tcp_keepalives_interval = 0            # TCP_KEEPINTVL, in seconds;
                                        # 0 selects the system default
#tcp_keepalives_count = 0               # TCP_KEEPCNT;
                                        # 0 selects the system default


but I would start with the setting in sysctl.conf setting the
following will cause it to heartbeat if the client is idle for 10
minutes. This should kill off most of the stale connections but leave
the all of the connections where the client is really still there
intact. It's also not a bad thing for your firewall connection
tracking. I like 10 minutes as an interval for this because it doesn't
flood your network needlessly since it only triggers if an connection
is idle for the entire 10 minutes, however does significantly help
clean up dead client connections while keeping false positives down to
a minimum.
"
net.ipv4.tcp_keepalive_time = 600
"
You can also increase the maximum connections in the postgresql.conf
but that will require you to be a little more aware of how memory
consumption in PostgreSQL works. i have it set to 100 (the default) in
my install
"
max_connections = 100
"




On Sat, May 12, 2012 at 7:54 PM, Paul Robert Marino <prmarino1 at gmail.com> wrote:
> Ill send you the exact paramiters I tuned on Monday.
>
> On May 12, 2012 1:56 PM, "Scott Piazza" <scott.piazza at bespokess.com> wrote:
>>
>> Good to know it isn't just me having this issue.
>>
>> I agree that monitoring must play a role, since it wasn't happening in
>> 1.6, and started when I enabled monitoring in 1.7.  With osa-dispatcher
>> turned off on the server, the issue hasn't reappeared.  But I'd rather
>> keep osa on.
>>
>> How did you change the idle connection timeout in Postgresql?  I admit,
>> I don't use Postgres all that often, but I don't see any way of doing it
>> without using something like pgpool.  Am I missing something obvious?
>>
>> ----------
>>
>> From: spacewalk-list-bounces at redhat.com
>> [mailto:spacewalk-list-bounces at redhat.com] On Behalf Of Paul Robert
>> Marino
>> Sent: Friday, May 11, 2012 6:56 PM
>> To: spacewalk-list at redhat.com
>> Subject: Re: [Spacewalk-list] Spacewalk 1.7 Postgres out of connections
>>
>>
>> Well I've worked around it by reducing the tcp heartbeat to 10 minutes
>> in /etc/sysctl.conf and reducing the idle client timeout in postgresqls
>> configuration.
>> But yes there is an issue with this.
>> I think it the monitoring that does it but I'm not sure. I can tell you
>> the incidents of it do go up with monitoring turned on.
>> The problem is is seems like its a shared library that isn't logging if
>> the class goes out of scope out of the database correctly that make
>> tracking the root cause significantly harder.
>> I'm not familiar enough with python to figure explain how it should be
>> fixed but in perl the DBI driver had. DESTROY method that catches if the
>> programer forgot to logout of the database before the instance of the
>> class goes out of scope or the application exits. It seems to me this is
>> missing from the python postgresql driver and if that functionality
>> could be added it would resolve the issue.
>>
>> _______________________________________________
>> Spacewalk-list mailing list
>> Spacewalk-list at redhat.com
>> https://www.redhat.com/mailman/listinfo/spacewalk-list