On 03/11/15 12:59, Toke Høiland-Jørgensen wrote:
> So we've seen a couple of issues with reconfiguration being weird (i.e.
> retaining previous values). Looking at the code in cake_change() and the
> iproute2 code, this seems to stem from the fact that tc only sets the
> parameters (in the netlink message) that are actually specified on the
> command line, and the kernel only changes the parameters specified in
> the netlink message.
>
> This leads to the different behaviour: On init, not specifying a
> parameter means "give me a sensible default", whereas on a reconfigure
> it means "don't change this parameter from its current value". This is
> obviously inconsistent. I'd argue that the right thing to do is the
> former in all cases; is there any reason why that would not be the case?
> If not, I'll go fix it :)

Interestingly I think the opposite.  On a change parameters should be
left alone unless they're the thing I'm changing.  It's a 'change' to an
instantiation, *not* an instantiation if you get my drift.

Kevin