On Tue, Nov 14, 2006 at 02:52:30PM -0500, Derrick J Brashear wrote: > On Tue, 14 Nov 2006, Jose Calhariz wrote: > > >>Can you get tcpdump from the client's point of view? Basically, at some > >>point the client is marking the server down, I assume. The question is on > >>the basis of what. > > > >I can imagine what is happening. Can I ask some questions? > > Asking that was itself a question :) > > >If they > >are explained somewhere, please point where. > > > >What the client see when the file server is: > > > > - Locking a volume to do a vos clone or vos backup? > > RW? RO? Locking is a vldb concept. Clients don't care. But taking a volume > offline causes VBUSY replies. But a clone can be done basically without a > period of inavailability, and a backup is just a special clone.
It's a RW volume and I think the locking takes long because it have many small files. > > > - When a volume is being salvaged? > > Here it's offline and you get VBUSY til it comes back, if it's one of > several ROs you can use another. > > > - When the fileserver is doing a weekly restart? > > VRESTARTING/VBUSY > > Look in src/viced/afsfileprocs.c, starting with > /* The volume we want may not be attached yet because > * the volume initialization is not yet complete. > > All of this should be dealt with by clients. That's the point. I starting to understand the problem you are looking. From reading to the comment, the client should had waited until the volume was ready. What I have seen is sometimes the client don't wait for the volume to be ready, so the compilation abort because of problems reading or writing to the volume. > > Derrick > _______________________________________________ > OpenAFS-info mailing list > [email protected] > https://lists.openafs.org/mailman/listinfo/openafs-info > José Calhariz -- A beleza e uma carta de recomendacao a curto prazo. -- Ninon de Lenclos
signature.asc
Description: Digital signature
