> Basically, I agree in principle with all the issues raised by John and
> Owen, save the complaint about -localauth.
I can see why -localauth confers authority on a machine,
but I wonder about it confering power on other machines.
Of course, this is useful, but supposed we had "Dept. A"
which would like to have their own fileserver (politics,
of course). If -localauth's power was confined to that
machine, I think we'd be happy (if they zorch their file
server, cie la vie). But since they could zorch more we
are fearful of this.
> As to the cache consistency problem John is seeing, that looks like a
> serious bug that should always be taken to our product support staff.
> John, I've suggested to your PSR (Todd) that you may have a replica
> which is corrupted: a volser bug, not a cache manager bug.
I have brought it to his attention and he is working on it.
> FWIW:
> The cmdebug output doesn't look terribly fishy, that's what an
> unintialized/recycled vcache structure usually looks like. The
> "states" field is a bitmap whose possible values can be found in
> afs.h. 0x4 means it's RO and not CStatd, ie, it doesn't have a
> callback and the status information is potentially bogus. (In fact, it
> is false, as you note. That's ok.).
> ... or else you need to run cmdebug _after_ "ls -l".
> The size reported by "ls -l" comes straight out of the bytes field in
> the vcache struct. cmdebug and/or kdump are definitely the tools that
> I would use to try to solve this.
Of course, ls -l on it doesn't work, that's the problem...
> I'm curious about the suggestion that:
> "I am actually wondering if this isn't the root of the problem." I
> think you may know something you don't know you know.
As I understand it, RO volumes now have (new? different?) callbacks.
What I wonder is what the 3.2 afsd's do with them? Do they understand
them properly. Since the cache can get confused right after a reboot,
before anyone logs in, how is thay bogus data getting there -- seems
like it must be a callback or something... On the other hand, what do
I know, maybe we should just hurry up and fork over the dough for a
source license...
John
