[OpenAFS] failed fileserver ip address change
Christof Hanke
hanke@rzg.mpg.de
Fri, 27 Oct 2006 16:07:34 +0300
Juha J=C3=A4ykk=C3=A4 wrote:
>> Hmm, so somehow it cached the location of the volume.
>> I wonder if a "fs checkvolumes" would have helped.
>=20
> I'm not very experienced on afs, but I'm not *that* beginner either. =3D=
) I
> tried that, it simply complained that it could not contact the server a=
t
> the old address. I'm not sure if it contacted the server at the new
> address, though.=20
>=20
> It seemed to me as if openafs servers thought there are two distinct
> servers: the one with the old ip and the one with the new. It never
Yes, to remove the old server from the vldb entirely you should do a
"vos changeaddr -oldaddr <original IP address> -remove".
The former commands just removed the entries of the volumes on the old=20
server.
> seemed to occur to openafs that they were indeed the same server. (Woul=
d
> it even be possible? I mean, how could it tell?) The problem was that
> openafs thought the volumes were at the old ip and even though the fs
> instance at the new ip had attached the very same volumes, they were no=
t
> visible. That's not true: the RO replicas at the new ip *were* indeed
> visible, but the RW's were not. Only, after vos delentry, all the old
> ones were (naturally) removed from the vldb, but *still* those same
> volumes from the new ip stayed inaccessible *to this one client*, which=
,
> by the way, happened to be running at machine which had just got a new
> ip. Perhaps that had something to do with it?
>=20
Just a guess: the new location of the volume is told the client via a=20
callback. If that's the case and went wrong, then you should find=20
something about CallBack failed for the old client-address in the=20
server-logs. But I'm not sure how the client are notified exactly.
/Christof
> -Juha
>=20