[OpenAFS] Dummy fileserver putting clients 'on hold'

[Rainer Toebbicke]

Nathan Davis wrote:
> Why not migrate everything to another machine, perform your upgrade, 
> then migrate everything back when the machine is ready again?
> 


That's actually what we used to do - however:

1. a 300-odd GB RAID5 takes 1-2 days to transfer to somewhere else
2. you've got to have that somewhere else spare
3. we've got around twenty of those...

and all that for a 3-minute load firmware/power cycle operation.

Todd DeSantis proposed an easy solution, particularly elegant as it is already 
largely implemented for other purposes: make the fileserver allow '0' for the 
'-busyat' switch. A small code change in viced.c + rx.c.

First tests show that it works and does exactly what we needed. Perhaps I'll 
add some code to set the rx_BusyThreshold to 0 via a signal in order to avoid 
restarting the fileserver.

Derrick's suggestion (answer RX keepalive's but do nothing else - play sort of 
'half-dead' for those few minutes) looks like it would do something similar.

Thanks to all for the hints!

-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Rainer Toebbicke        http://cern.ch/~rtb         rtb@mail.cern.ch  O__
European Laboratory for Particle Physics(CERN) - Geneva, Switzerland   > |
Phone: +41 22 767 8985       Fax: +41 22 767 7155                     ( )\( )