On 10/12/07, Richard L. Hamilton <rlhamil at smart.net> wrote:
> >   3) Install in single user mode and/or reboot after
> >     installation (very commonly over-cautious advice)
> about dependencies of the earlier incarnation anymore, do I?).  As
> to the third point in particular, I would suppose it depends on one's
> imagination as to what's overcautious and what isn't, but if and when
> everyone has zfs root and promotable clones become part of patching,
> wouldn't it be a matter of dividing up patches that didn't need a reboot
> and whose benefits were needed immediately and which were very
> low-risk from the rest, and doing the first bunch conventionally and the
> second bunch via a clone, and scheduling the reboot that switched clones
> at one's convenience?

I am thinking of the situation where you need a fix to a particular
problem that really doesn't require reboot or single user mode and
this would be fairly widely agreed upon by competent developers.  A
key recent example is telnetd, but I've also had the same thing for
patches that replaced single binaries like iostat and vmstat.  The
difference for such a patch that has huge impact is:

"I need to apply this patch.  It will take 5 minutes and there will be
no interruption to service.  There is extremely low risk that this
will affect anything on the system."

vs.

"I need to apply a patch that requires a reboot.  This is the database
server for X that affects several app tiers.  While it is down, we
will be unable to perform many functions required to keep factories
running.  The outage will last about 1 hour."

That is: 5 minutes of a sysadmins time during the middle of the day
(plus light change control activity) vs. hundreds to thousands of man
hours of lost productivity because a critical system is down during
work hours or fewer lost man hours and a sysadmin gets to put in some
more OT.

-- 
Mike Gerdts
http://mgerdts.blogspot.com/

Reply via email to