[CentOS] About mysql upgrade

Thu Apr 28 16:53:42 UTC 2016
Dennis Jacobfeuerborn <dennisml at conversis.de>

On 28.04.2016 17:58, Lamar Owen wrote:
> On 04/28/2016 08:45 AM, Sergio Belkin wrote:
>> I've found some issues upgrading mysql, some people recommends run
>> mysql_upgrade. I wonder why such a script is not run from scriptlet of
>> mysql-server rpm.
> Back in the Dark Ages of the PostgreSQL RPMset (PostgreSQL 6.5), early
> in my time as RPM maintainer for the community PostgreSQL.org RPMset, I
> asked a very similar question of some folks, and I got a canonical
> answer from Mr. RPM himself, Jeff Johnson.
> The answer is not very complex, but it was spread across a several
> message private e-mail thread.  The gist of it is that the RPM
> scriptlets are very very limited in what they can do.  Trying to do
> something clever inside an RPM scriptlet is almost never wise.  The key
> thing to remember is that the scriptlet has to be able to be run during
> the OS install phase (back when upgrades were actually supported by the
> OS installer that is now known as anaconda). Quoting this introductory
> section:
> On August 18, 1999, Jeff Johnson wrote:
>> The Red Hat install environment is a chroot. That means no daemons,
>> no network, no devices, nothing. Even sniffing /proc can be problematic
>> in certain cases.
> Now, I realize that that is OLD information; however, anaconda is still
> doing the same basic chrooted install, just with a prettier face.  You
> cannot start a daemon in the chroot, since many things are simply not
> available to the scriptlets when installed/upgraded by anaconda. 
> Scriptlets have to work in an environment other than 'yum update.'  And
> also note that this is a very different situation than Debian packages
> live in; RPM scriptlets are essentially forbidden from interactivity
> with the user; Debian's equivalent are not so hindered.  At least that
> was the rule as long as I was an active packager.
> Further reference a WayBack Machine archive of a page I wrote long ago:
> https://web.archive.org/web/20010122090200/http://www.ramifordistat.net/postgres/rpm_upgrade.html
> And leaving you with this thought, again from Jeff Johnson:
> On August 18, 1999, Jeff Johnson wrote:
>> Good. Now you're starting to think like a packager  Avoiding MUD is
>> *much* more important than attempting magic.
> ....
>> The bottom line is you shouldn't attempt a database conversion as
>> part of the package install. The package, however, should contain
> programs
>> and procedures necessary to do the job.

The real reason something like mysql_upgrade isn't run automatically
hasn't really much to do with the technicalities of RPM but the fact
that mysql_upgrade might kill your server if you don't know what you are
If for example you have a big table that takes up 70% of your storage
and mysql_upgrade needs to convert it to a new format then it will
create a copy of that table in the new format first and then delete the
old one...except of course that this would require 140% of disk space to

So independent of the packaging technology used major changes like these
should never be done automatically and instead always be handled by an
admin who knows what he is doing.