On Wed, Mar 26, 2003 at 10:22:40PM +0200, Tzafrir Cohen wrote: > In other words: the guests still have to be shut down for every software > update in that method. Unlike what normally happens in an rpm -Uv .
Yes, although you certainly can apply several RPMS at once. But eventually you should reipl to release your local copy of /usr and get the shared copy with the service applied. There may be some way to coax the DASD driver into doing this on the fly, but I bet you have to shut down to single-user anyway. > rpms may have certain pre/post-[un]install script that deal with the data in > /etc or in /var . Frankly they may want to do some strange thing with > files in /usr , though I can't think of a specific example. Well, if they want machine-specific files in /usr, then you can't be using a shared-/usr setup in any event, can you? /etc and /var can be updated normally with the method I described. I don't understand what you're envisioning. I see there as being three cases of service application: 1) everything lives in /usr. Well, that's easy: you apply it to the template machine and, yes, force the guests (probably through an IPL) to get a copy of the new disk. 2) nothing lives in /usr. Also trivial: apply service once per guest. 3) some stuff lives in /usr and some doesn't; that's when you need to do a bind-mount. Adam