On Sat, Nov 29, 2014 at 01:16:52AM +0100, Adam Mazur wrote:
> We are using almost newest version of OVS (master branch, pulled on
> the 14th of November) under Xen. There is no difference in datapath
> version - tested on both kernel (3.17) module and with OVS-dkms
> version. There is no difference if using a controller or not.
>
> The problem is: when there is more than 6-10k packets per second,
> from even one VM to varied destinations, then vswitch process eats
> more-and-more memory, until OOM. Memory usage starts growing a few
> seconds after such flood begins, and stops growing immediately when
> such flood is over. Valgrind is useless - it slows down too much to
> eat few thousands packets per second, therefore I'm unable to find
> out where the leak is.
>
> The simplest python code to generate such packet flood from VM:
>
> import random, socket, struct
> sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
> while True:
> ip_raw = struct.pack('>I', random.randint(1, 0xffffffff))
> ip = socket.inet_ntoa(ip_raw)
> try:
> sock.sendto("123", (ip, 12345))
> except:
> pass
Alex, you were working on an OOM issue a few weeks ago. Does this
sound like that one? Did you fix that issue?
Thanks,
Ben.
_______________________________________________
discuss mailing list
discuss@openvswitch.org
http://openvswitch.org/mailman/listinfo/discuss
