On Sun, Nov 04, 2001 at 12:27:51AM -0900, Ethan Benson wrote: > On Sun, Nov 04, 2001 at 07:14:18PM +1000, Anthony Towns wrote: > > > > However this doesn't work quite frequently, since the umount fails > > if there's any file in /usr that's been deleted (ie, was in a package > > which has been updated), but is still being used (a /usr/lib/lib* that's > > referenced by a long running program, or a /usr/bin/* program that's still > > running, like apt-get or dselect). This is because the filesystem will > > be modified when the file is closed, because the inode will be freed, > > so it's obviously not read-only. > > > > That's the theory AIUI, anyway. > > thats presisly why. Tell it to Adam Heath... > simplest way to deal with it is run lsof +L1 after an upgrade and > kill/restart all offending processes. Like your X server? :) > in which case you will have an increasing supply of cruft as time goes > on? especially if reboots happen less then every 8 monthes. Or you can periodically take the box down to runlevel 1 and take care of it. Besides which, gaping security holes are found in the kernel with a greater frequency than once every 8 months, so you'll be rebooting to upgrade it anyway. > just IMO, but id rather just clean things up and kill offending > processes then hide the problem. Anthony's not saying everyone should run their system this way. He's helping those that want to do so effectively. -- G. Branden Robinson | One man's "magic" is another man's Debian GNU/Linux | engineering. "Supernatural" is a branden@debian.org | null word. http://people.debian.org/~branden/ | -- Robert Heinlein
Attachment:
pgpi5XvqcxChY.pgp
Description: PGP signature