[OpenAFS] another problem with vos release
Thu, 16 Jun 2005 17:09:16 -0400
On Thursday, June 16, 2005 12:30:27 PM +0200 Hans-Gunther Borrmann
> On Wednesday 15 June 2005 16:48, Ken Hornstein wrote:
>> > RWrite: 536871953
>> > number of sites -> 2
>> > server leda partition /vicepa RW Site -- New release -- Old
>> > release server atlas.cg.cs.tu-bs.de partition /vicepb RO Site -- Old
>> > release -- Old release
>> I couldn't help noticing that you don't have a RO volume on the same
>> machine as the RW copy.
>> This once happened to us here on one volume (don't ask; someone here
>> INSISTED that it didn't cause any problems). I noticed "weird"
>> behavior during vos releases. I can't quantify "weird" anymore than
>> that; it was strange and it was a while ago, and I pestered people
>> until it had a RO on the same machine as the RW copy, and the weirdness
>> went away. I could easily believe that not having a RO copy on the
>> same machine as the RW copy would be a corner case that isn't tested
>> that much, and maybe it would be worth putting one there just to see if
>> it solves some of your problems.
> What I noticed in this situation was , that "vos release" of a "large"
> volume took an incredible long time. "large" in this case meant ~100000
> Files. Having an ro on the same server reduced the time to a normal
> value. --
Yes, that's going to be true. If you don't have an RO site in the same
place as the RW, then it needs to create a temporary release clone for each
release. Doing that takes as long as creating the local RO site for a full
release, which is basically linear in the total number of vnodes in the
volume. If there is a permanent local RO site, then it requires only an
incremental update, which is still linear in the number of vnodes, but with
a much smaller constant.