On Thu, Feb 27, 2014 at 03:27:37PM -0800, David Wolfskill wrote:
I am not at all clear how to perform a migration of machines that
run authoritative nameservers from FreeBSD 9.x - 10.x, given the
current setup of the dns/bind99 port. I'm hoping for some clues,
if not insight.
I seem to have managed to perform the above migration for one of
the two machines that I have here at home that meet the above
description. (I have not yet attempted it for the other; if things
seem OK after a week, it's next up.)
Unrelated to this issue, I acquired the use of a test machine (to
which I restored the backup images of the file systems of the machine
I just upgraded, then changed the hostname IP address). I was
thus able to experiment a bit.
On these machines, I have them set up to boot from either of 2
slices (each of which contains its own / and /usr; /var is the same
file system (on a 3rd slice) regardless of which slice is booted),
and flip from one slice to the other at each upgrade. I normally
track a stable/N branch, updating weekly. Salient parts of the
upgrade process for these intra-branch upgrades:
* Clone the running slice to the other one.
* Ensure that the root and usr file systems from the non-booted slice
are mounted at a suitable mountpoint.
* Mount /usr/src and /usr/obj read-only via NFS from the build machine.
* cd /usr/src \
make installkernel installworld DESTDIR=${other_slice_mountpoint};
mergemaster gets a -D flag for similar purposes.
* Reboot from the newly-populated other slice.
* Mount /usr/src and /usr/obj read-only via NFS from the build machine
(because some ports (e.g., sysutils/lsof) want access to that
information).
* Update installed ports. (I have been using portmaster -ad for some
time for this.)
* Perform the make delete-old-libs mentioned in src/UPDATING.
* Reboot to ensure that nothing is still using old ports.
For this exercise (9.x - 10.x on a machine running an authoritative
name server), here's what worked for me:
* Mount /usr/src and /usr/obj read-only via NFS from the build machine
(because some ports (e.g., sysutils/lsof) want access to that
information).
* Update installed ports.
* cp -pr /var/named/etc/namedb /usr/local/etc/
* cd /usr/local/etc/namedb
foreach f (`find . -type d -name RCS -prune -o -type f -print0 | \
xargs -0 grep -l '/etc/'`)
sed -i -e 's/\/etc/\/usr\/local\/etc/g' $f
end
* Clone the running slice to the other one.
* Ensure that the root and usr file systems from the non-booted slice
are mounted at a suitable mountpoint.
* Mount /usr/src and /usr/obj read-only via NFS from the build machine.
* cd /usr/src \
make installkernel installworld DESTDIR=${other_slice_mountpoint};
mergemaster gets a -D flag for similar purposes.
* Reboot from the newly-populated other slice.
* Install ports/misc/compat9x (e.g., portmaster misc/compat9x)/
* Re-install ports/dns/bind99 (e.g., portmaster dns/bind99)/
* service named restart (and verify that lookups are faster now that the
first nameserver listed in /etc/resolv.conf actually has named
running).
* Perform the make delete-old-libs mentioned in src/UPDATING.
* Reboot to ensure that nothing is still using old ports.
Note that while dns/bind99 and misc/compat9x were built/installed under
10.x, the rest of the ports on the system are still running after having
been built/installed under 9.x. This is intentional, so I have a
relatively easy fallback option in case of Something Bad happening
(reboot from the previous slice, which still has stable/9 installed).
After a week (or two), I expect to cut over fully, and perform the
process documented near the bottom of portmaster(8) to rebuild/iinstall
all installed ports under stable/10. And then I expect to do the same
for my laptop and the build machine... and then I'll stop tracking
stable/9.
YMMV, and all that.
Peace,
david
--
David H. Wolfskill da...@catwhisker.org
Taliban: Evil cowards with guns afraid of truth from a 14-year old girl.
See http://www.catwhisker.org/~david/publickey.gpg for my public key.
pgpB9XBjVnbhK.pgp
Description: PGP signature
