[lxc-users] LXC leaking ptys?

[Serge Hallyn]

Quoting Harald Dunkel (harald.dunkel at aixigo.de):
> On 07/15/14 16:24, Serge Hallyn wrote:
> > Quoting Harald Dunkel (harald.dunkel at aixigo.de):
> >>
> >> Nothing unusual. The log files show a lot of ssh sessions run by
> >> our monitoring software (active just for a few milliseconds), plus
> >> my own ssh sessions for maintenance. Thats all.
> > 
> > Ah, interesting.  That could be fun to try to reproduce.  But I
> > would expect that to be transient, so that the monitoring software
> > could just retry, by which point the other very-short connections
> > should be closed and new connections should succeed...
> > 
> 
> To give you more information about these high-frequency ssh sessions:
> Every 5 minutes I check the status of the LXC clients, using something
> like
> 	ssh $server test -d /cgroup/lxc/$client && echo "${server}:${client} OK"
> 
> There are 4 ssh jobs per client, i.e. I get about 120 ssh sessions
> within a few seconds. Then its silent for the next 5 minutes. Important
> is that these ssh sessions don't need a pty. They succeed, even when
> an interactive ssh session would report "stdin: not a tty". Surely
> they do not overlap each other.
> 
> I have reduced this to run a test every 30 minutes now, but actually I
> do not see a problem with this. I monitor all my LXC and KVM servers
> using this script. Only interactive sessions on the server are affected.

Could you be simply running out of memory?  Out of fds?