> That is fine for installation of new OS. I do not see
> how that would beat thousands of systems grabbing the
> latest package, updating and then continue to run as
> normal whereas flar would require those systems to
> reboot for a mere update.

While I think a lot of current Solaris patch READMEs simply call for
a reboot rather than bothering to explain under what conditions
it might be necessary, I also suspect that anyone who believes that
almost arbitrary updates can be done without any need for a reboot
is sadly misguided.  Even if a reboot wouldn't be needed to prevent
a crash, I think certain updates are best done in something like
single-user mode, so that it's assured that an application is using either
entirely the pre-update environment, or the post-update environment,
but not some possibly unstable mix of the two.  Yes, I know that in most
isolated cases of a lib or something being in use, it's not a problem as long
as files being updated are not overwritten, but are replaced atomically
(load new version under derived name, and then rename it into place).  But
AFAIK, there's no way to simultaneously replace multiple files atomically,
which means there's no way to have a totally clean cutover in case for
example libs have private interfaces between one another, or a lib
has a set of run-time-loaded plugins, or the like).  And that's just a
simple example.

There are probably cases where one can say an update is definitely safe
without a reboot; there are probably relatively few cases where one can
say that a reboot will almost always be a very good idea.  IMO, there are
probably far more cases where an update without a reboot will _usually_
work, but occasionally make an incredible mess; and identifying those,
let alone the circumstances under which avoiding the reboot would be
safe, is obviously difficult, and not entirely amenable to being dealt with
automatically, except by erring on the side of caution.  (this from someone
that used to use the debugger in write mode on the kernel to work around
problems that would have otherwise required a reboot at an inconvenient
time, decades ago; but note that even then, I'd reboot the box as soon as
rush hour was over, because I couldn't _prove_ that I didn't put it on the
road to future insanity)
 
 
This message posted from opensolaris.org
_______________________________________________
opensolaris-discuss mailing list
[email protected]

Reply via email to