[OpenAFS] Re: read-only to read-write fast?

[Brandon S. Allbery KF8NH]

On Saturday 16 August 2003 18:22, Hein Roehrig wrote:
> Thanks for your explanations. Is the vlserver change an atomic
> operation that can be performed while the original fileserver is down?

It's handled automatically:  when the original fileserver+volserver is shut 
down de-registers its volumes, when the new one comes up it re-registers 
them.  If the old fileserver crashed without deregistering, "vos syncvldb" 
can be used to force the issue.

> What happens to mmaps (e.g., running programs) and other open files?

That I couldn't tell you for certain, but for files open for read it shouldn't 
be an issue and files open for write are cached on the client and don't 
actually get written until closed.  (mmap()ing files in AFS is probably a bad 
idea in general; AFAIK the file mmap()ed will actually be in the client's 
cache, and the AFS copy will only be updated on close.)

AFS file servers and clients track file "generations"; as long as the 
backup/restore restores the same file generation (that is, you're using 
something based on "vos dump" instead of a file-by-file backup which loses 
AFS metadata; but if you do the latter you also lose ACLs) the right thing 
should happen in general.

-- 
brandon s. allbery   [linux,solaris,freebsd,perl]      allbery@kf8nh.com
system administrator      [WAY too many hats]        allbery@ece.cmu.edu
electrical and computer engineering, carnegie mellon univ.         KF8NH
URGENT!  E-xpedient nuked APK subdomains; kf8nh.apk.net is DEAD.  Sorry.