[quagga-dev 7641] Re: [PATCH] bgp: add snapshot to workaround slow commands
ccaputo at alt.net
Thu Jan 14 19:27:29 GMT 2010
On Thu, 14 Jan 2010, Stephen Hemminger wrote:
> On Thu, 14 Jan 2010 17:34:53 +0000 (GMT) paul at jakma.org wrote:
> > On Wed, 13 Jan 2010, Stephen Hemminger wrote:
> > > On a fully loaded BGP instance, doing 'show ip bgp' can effectively
> > > kill the BGP process because the single BGP daemon thread is stuck waiting
> > > to feed information back to the vtysh.
> > > This patch is a simplistic solution to the problem. It adds:
> > > show bgp snapshot
> > > which cause the original BGP daemon to fork. The vty client is then
> > > talking to the child which can then be used to do slow commands, and the
> > > parent BGP process is available to do normal duties.
> > That's simpler than all the infrastructure to wait on children in the
> > parent and take back responsibility for the vty alright.
> > Thing is, how useful is this mode-change version? Usually the point
> > at which I realise a command may take so long that it might hurt BGP
> > is when waiting on a command to finish - after having executed it by
> > definition.
> > I.e. this approach implies we should give up on the end-user
> > transparent way. Can it not be gotten to work?
> Yes, transparent is possible, just more difficult.
fork() on any display command that can take too much time? (ie., "show ip
bgp", "show ip bgp regexp", etc.)
More information about the Quagga-dev