[sles-beta] Suggestion to speed up installs/Upgrades/patching for "future" SLE

Josef Reidinger jreidinger at suse.cz
Mon Jun 23 02:34:13 MDT 2014


On Mon, 23 Jun 2014 18:25:52 +1000
"Darren Thompson" <darrent at akurit.com.au> wrote:

> Frederic
> 
> I'm seeing the "multiple dracut initrd regenerations" during SLES12
> beta9 so it's still very much evident in SLES12.
> 
> According to Matthias (you replied to his email), the issue was
> already considered and the spedup was rejected in favour of maintaing
> package stability (in the event of a crash during upgrades).
> 
> Using RPM "transaction" does seem like a good option, perhaps some
> more discussion between the 'coders' may be warranted.
> 
> Darren
> 
> 

Well,
let me explain how it works in SLE11. There was a special ENV variable,
that all rpm respect in its post install script. This variable said if
it is run in YaST. If such variable is set, then all such stuff is
skipped and YaST2 itself generate all needed stuff.
It sounds pretty elegant, but there is some nasty drawbacks.

1) YaST runs all post install hooks even if such packages are not
installed or if it is not updated, so YaST was slow for simple install
of single package

2) YaST need to be in sync what packages do, this means YaST have to
have same code as is in rpm post install. It is duplicite and we have
trouble to keep it in sync and what is worse it is quite hard to detect
if such situation happen resulting in nasty bugs. This is main reason
why we drop it.

For me RPM transaction is much nicer solution if it work reliable even
if upgrade failed in middle of install ( I do not know how this exactly
work ).

Josef


More information about the sles-beta mailing list