On Mon, Feb 16, 2004 at 09:19:45AM +0100, Lars Marowsky-Bree wrote:
> On 2004-02-13T17:08:59,
>    Arjan van de Ven <arjanv@redhat.com> said:
> 
> > one thing you can do is provide a way for drivers to wake the userspace
> > tester early. Say by default it polls every minute, but if the fiber
> > channel driver gets a LIP UP event it (via a central API) makes the
> > userspace daemon *now*.
> 
> I may be missing something obvious, but a LIP UP should be accompanied
> with a round of 'device detections' on that link, which already should
> trigger a few hotplug events, no?
> 
> So this seems pretty much solved.

not normaly; there are several reasons the loop can bounce briefly and right
now the fiber drivers don't notify linux of that every time. Maybe that's
for the better .... if it's a frequent thing that is short-timed then it
would be obscene to yank the disks from under the user (and force-umount his
fs) every few hours..

while in multipath you do want to at least stop using the current path if
there is another path that is not in negotiation...