Glenn@canit.se wrote:
[snip]
> As an AFS-admin i know that there are reasons when logging isn't good at a
ll,
> but the cases I know of are very few.

How does AFS interfer with UFS logging ?

----

Bye,
Roland

--
__ .  . __
(o.\ \/ /.o) roland.mainz@nrubsig.org
\__\/\/__/  MPEG specialist, C&&JAVA&&Sun&&Unix programmer
/O /==\ O\  TEL +49 641 7950090
(;O/ \/ \O;)

[ Post a follow-up to this message ]

> + On 19-Dec-04 04:01:54
+Roland Mainz <roland.mainz@nrubsig.org> wrote

>Glenn@canit.se wrote:
>[snip] 
[vbcol=seagreen]
>How does AFS interfer with UFS logging ?

If you use logging on your vicep partitions on your AFS fileserver you get
alot of problems, our "neighbour cell" at work have experiences of exactly t
hat 

[ Post a follow-up to this message ]

Glenn@canit.se wrote: 
> 
>
> If you use logging on your vicep partitions on your AFS fileserver you get
> alot of problems, our "neighbour cell" at work have experiences of exactly that 


Which kind of problems ? Did you report then to Sun yet ?

----

Bye,
Roland

--
__ .  . __
(o.\ \/ /.o) roland.mainz@nrubsig.org
\__\/\/__/  MPEG specialist, C&&JAVA&&Sun&&Unix programmer
/O /==\ O\  TEL +49 641 7950090
(;O/ \/ \O;)

[ Post a follow-up to this message ]

Roland Mainz <roland.mainz@nrubsig.org> writes:

> Which kind of problems ? Did you report then to Sun yet ?

Are you aware of how AFS manages the underlying file system?

I seem to remember that the AFS documentation explicitly states you
aren't supposed to use UFS logging on vicep partitions.  If you
deliberately do this anyway, is it, in your opinion,

a) IBM's problem (or OpenAFS' as it might be)
b) Sun's problem
c) the admin's own f*... problem?

Unless you mean that UFS+logging should be expected to work 100% like
regular UFS under any circumstances, including power interruptions and
other anomalies, I can't see how reporting this to Sun would serve any
meaningful purpose.

--
Atro Tossavainen (Mr.)               / The Institute of Biotechnology at
Systems Analyst, Techno-Amish &     / the university of Helsinki, Finland,
+358-9-19158939  UNIX Dinosaur     / employs me, but my opinions are my own.
< URL : http : / / www . helsinki . fi / %7E atossava / > NO FILE ATTACHMENT
S

[ Post a follow-up to this message ]

> + On 20-Dec-04 04:38:50
+Roland Mainz <roland.mainz@nrubsig.org> wrote

>Glenn@canit.se wrote: 
[vbcol=seagreen]
>Which kind of problems ? Did you report then to Sun yet ?

Well, the data get trashed.

But I dont think it's a bug, its just the way that AFS stores the data,
the same reason that you can't use Sun's fsck on the same partitions,
you have to use the transarc/OpenAFS fsck, otherwise it *will* destroy
the data.

..And personally I never tried it, but I know people that had to restore
backups both after using logging and sun's fsck..

[ Post a follow-up to this message ]

Glenn@canit.se writes:

>But I dont think it's a bug, its just the way that AFS stores the data,
>the same reason that you can't use Sun's fsck on the same partitions,
>you have to use the transarc/OpenAFS fsck, otherwise it *will* destroy
>the data.


Uhm, AFS actually modifies the UFS data structures while kernel UFS
modifies them too?

Double yuck.  (I know they also modify supposedly read-only kernel
data structures)

Casper
--
Expressed in this posting are my opinions.  They are in no way related
to opinions held by my employer, Sun Microsystems.
Statements on Sun products included here are not gospel and may
be fiction rather than truth.

[ Post a follow-up to this message ]

> + On 22-Dec-04 01:27:29
+Casper H.S. Dik <Casper.Dik@Sun.COM> wrote

>Glenn@canit.se writes:
 

[vbcol=seagreen]
>Uhm, AFS actually modifies the UFS data structures while kernel UFS
>modifies them too?

Dont know, I never looked into how it works technically, but I guess
the developers at openafs.org could tell you exactly how it works.

(Or if you prefere, download the source and check how it's done ;)


>Double yuck.  (I know they also modify supposedly read-only kernel
>data structures)

Maybe, I just know that AFS is a beuty to administrate when you have alot
of untrusted users and computers, therefore I like it  ..how it's
done under the hood is SEP 

[ Post a follow-up to this message ]

Atro Tossavainen wrote:
>
> Roland Mainz <roland.mainz@nrubsig.org> writes:
> 
>
> Are you aware of how AFS manages the underlying file system?
>
> I seem to remember that the AFS documentation explicitly states you
> aren't supposed to use UFS logging on vicep partitions.  If you

that is correct, you have to disable logging an a filesystem
that serves as AFS backing store.

> deliberately do this anyway, is it, in your opinion,
>
> a) IBM's problem (or OpenAFS' as it might be)

A) is the correct answer, it comes from the way AFS hooks into UFS
and calls UFS internal functions directly and interacts with the
UFS locking protocol. what it can not do is to take the logging aspects
into consideration as well and so you can get all sorts of interesting deadl
ocks
while running with logging enabled. UFS is not supposed to be used
as a stackable filesystem in this way, in particular when logging comes
into play.

> b) Sun's problem
> c) the admin's own f*... problem?
>
> Unless you mean that UFS+logging should be expected to work 100% like
> regular UFS under any circumstances, including power interruptions and
> other anomalies, I can't see how reporting this to Sun would serve any
> meaningful purpose.

I already had contacts with the OpenAFS developers from time to time
on locking issues with UFS and UFS+logging so we could make at least the
non-logging part work better together with UFS, the problems that arise
when logging is enabled are not quickly resolvable the way AFS hooks
onto UFS as the code stands.

YOu can always use the namei server instead.

---
frankB

[ Post a follow-up to this message ]

Frank Batschulat wrote: 
>
> that is correct, you have to disable logging an a filesystem
> that serves as AFS backing store.

;-((
 
>
> A) is the correct answer, it comes from the way AFS hooks into UFS
> and calls UFS internal functions directly and interacts with the
> UFS locking protocol. what it can not do is to take the logging aspects
> into consideration as well and so you can get all sorts of interesting dea
dlocks
> while running with logging enabled. UFS is not supposed to be used
> as a stackable filesystem in this way, in particular when logging comes
> into play.

Will that be different with zfs (e.g. can it be safely be used as
stackable filesystem) ?
 
>
> I already had contacts with the OpenAFS developers from time to time
> on locking issues with UFS and UFS+logging so we could make at least the
> non-logging part work better together with UFS, the problems that arise
> when logging is enabled are not quickly resolvable the way AFS hooks
> onto UFS as the code stands.
>
> YOu can always use the namei server instead.

Uhm... what's that ?

----

Bye,
Roland

--
__ .  . __
(o.\ \/ /.o) roland.mainz@nrubsig.org
\__\/\/__/  MPEG specialist, C&&JAVA&&Sun&&Unix programmer
/O /==\ O\  TEL +49 641 7950090
(;O/ \/ \O;)

[ Post a follow-up to this message ]

In article <41D3469B.6E0F8A9F@nrubsig.org>,
Roland Mainz <roland.mainz@nrubsig.org> writes:
> Frank Batschulat wrote: 
>
> ;-((
> 
>
> Will that be different with zfs (e.g. can it be safely be used as
> stackable filesystem) ?
> 
>
> Uhm... what's that ?

I gather the default AFS server process (on those platforms that support
it) does underhanded and devious things (for performance, I suppose),
while the NAMEI version of the server (only the default on those platforms
where the more devious version isn't implemented; i.e. you have to do
some explicit configuration when you build AFS to get this) avoids doing
quite so devious things, which would presumably allow ufs+logging, vxfs,
etc. as the underlying filesystem.  Although whether you'd get better
performance with ufs+logging and the NAMEI server, or with the "inode"
server (the devious one) + plain ufs, is anybody's guess, unless of
course someone has tried both ways (and you'd probably need to know
just what their configuration, OS and AFS version, and patches were to be
able to use them to predict what you might see).

No, I haven't tried; AFS (client even) is just too darn unstable for me,
as with a mere dialup, I need to be able to stop it without trashing it
or crashing, no sure thing last I tried.

--
mailto:rlhamil@smart.net  http://www.smart.net/~rlhamil

Lasik/PRK theme music:
"In the Hall of the Mountain King", from "Peer Gynt"

[ Post a follow-up to this message ]