On 11/06/2011 03:17 PM, Nico Kadel-Garcia wrote:
On Sun, Nov 6, 2011 at 3:56 PM, Yasha Karant<[email protected]>  wrote:
Rather than hijacking the thread on vlc, I am posing this as a new set of
questions.

My understanding of the EL upgrade process, when moving from EL N to EL N+1
(e.g. EL 5.7 to EL 6.1), there are only two ways to protect the information
on the hard drives that have "systems" directories (e.g., / , /boot , /usr ,
... ).

Definen "protect data". A complete backup is usually the safest before
doing *any* such upgrade.

1.  Make backups of everything and restore after the EL N+1 installer runs,
hopefully forgetting nothing of importance.

Common, especially if you plan to be stripping out unnecessary
services and configurations. You get a cleaner installation,
especially of "config" files with RPM updates leave untouched.

2.  Keep /home , /opt , /usr/local , and all other non-distribution given
applications, configurations, and data, on separate partitions that are not
touched during the EL N+1 installation process by using a manual override
selection during the EL N+1 installation process.

Yasha, you keep making these strange and incomplete dichotomies. Our
favorite upstream vendor does a good job of following the File System
Hierarchy, and keeps system configurations in well-defined locations
of /usr, *not* /usr/local, and keeping system materials out of /home/

The fun and games occurs when someone's been adding out-of-band
packages, such as customized Apache layouts /home/httpd, or anything
written by Dan Bernstein that insists on its own "special" loations in
/qmail/, /djbdns, or various other inappropriately located base
directories. (It's an old sorepoint for folks who've maintained such
packages.) It also occurs when people decide to install multiple local
versions of core utilities in /usr/local/, such as keeping 4 or 5
different manually installed Java versions in /usr/local/java.

2.1  If these are not real partitions but are handled via the Linux LVM
method, will these be respected or will these be overwritten?

Partition membership is irrelevant. It's a filesystem, and if the
partitions are mounted at installation or update time, they will
potentially be overwritten or even rebuilt from scratch. And if your
setup is complex or clever, mistakes become easy. This is why
separating bulky development areas to separate partitions or disk and
separate backup strategies are so helpful.

2.2  Is there any mechanism to force the EL N+1 installer to ignore certain
directory trees that are not real partitions (e.g., do not touch /usr/local
if /usr/local actually is on the same partition as /usr -- given that /usr
in general must be overwritten to install EL N+1)?

Unmount them, and do not let the update procedures mount or
re-partition them. Be aware that if the update of software does seek
to make changes, for example by putting files in /usr/local/ with
software from a File System Hierarchy violating 3rd party repository,
and the partitions are not yet mounted, the changes will be concealed
by the mounting.

As an aside, when I do this, I keep many things from the EL N installation,
including many configuration files in /etc so that these can be quickly
reinstalled.  I personally accomplish this by copying these to a
subdirectory of a directory for a mount point for a filesystem that will not
be touched (e.g., /home/prevsys/etc with /home on, say, /dev/sda10), using
cp -pr to save the required permissions.

I do an rsync based backup of /etc and /home to another host, plus any
relevant config directories from /opt or /usr/local as needed.

These days, I make certain that each physical partition that will be
reformatted/overwritten by the EL N+1 installer is large enough to
accommodate the typical growth (bloat?) of the new major release.

Well, yes. This is where it's helpful to simply have a complete backup
to restore from. In particular, the new SL 6 structure for naming
logical volumes with the hostname included is very helpful in virtual
environments, allowing the mounting of partitions on the hypervisor or
with the disks transferred to a new host to be much simpler and avoid
conflict with existing logical volume names. But if you're upgrading
from SL 5.x to SL 6.x, you won't inherit this. You'll need to
repartition or play serious games to get that kind of feature.

Mind you, it's often much simpler and quite effective to simply use
the "upgrade" features built into the DVD's or kicstart setups. You'll
miss out on a few features like the volume naming, but our favorite
upstream vendor has done a very effective job of preserving
configurations across upgrades without having to do a complete
rebuild, by carefully managing config file updates and providing
update tools for databases.

I have been informed that "update" will not work from EL N to EL N+1 for major releases, only for minor sub-releases (EL 6.0 to EL 6.1, etc.). Major releases require a full install, as though one were using a completely new hardware system. My personal backup technique these days is simply to buy a new disk, clone it from a working one using dd, and then if I must install to overwrite, bring up the backup disk on an unused disk controller interface connection, and copy (cp -pr, cpio, tar, etc.) files and/or hierarchies from the archived backup to the production disk. This is feasible today because of the relatively low price of the bare harddrives, and necessary in my case from my workstation or laptop because I do not have access to a high throughput channel to a disk farm server (e.g., a NAS). At low throughput, the task simply takes too much real time -- what will work for a few Gbytes in terms of real time is not feasible for a Tbyte or more.

The sorepoints you mention are unfortunate, but also real, and not only for the sort of applications you mention, but also for applications that are commercial, without source, and licensed for fee.

Moreover, during the major release upgrades, my understanding is that / , /usr, and other systems directories that are the mount points for the underlying partitions containing file systems are automatically reformatted (erased) or the upgrade will fail. I believe that I have email from Connie Sieh on this very point.

Do I understand you correctly that LVM works the same as physical partitions, and that particular logical entities (e.g., the LVM analogue of physical partitions) can be exempted during the full install of a new major release?

Reply via email to