10:42 a.m.
I haven't updated my rawhide box for a while and now I started a yum
update and it takes ages to update the 680 packages. I noticed that
ldconfig is run after a package is installed (because this packages have
it in %post). Why can't yum simply skip this and only do a ldconfig at
the end?
this would save a lot of time in updates like this and would also be
good for anaconda (update from FCX -> FCY)
11:07 a.m.
>>>> "d" == dragoran
<dragoran(a)feuerpokemon.de> writes:
d> Why can't yum simply skip this and only do a ldconfig at the end?
The sorting gets pretty hellish; some of the packages which need an
ldconfig run might be needed for scriptlets of other packages. I'm
sure it's technically possible to lower the number of ldconfig runs
assuming that the dependency information of every single package is
completely correct, but I imagine it wouldn't be terribly simple.
- J<
11:27 a.m.
Jason L Tibbitts III wrote:
>>>>> "d" == dragoran
<dragoran(a)feuerpokemon.de> writes:
>>>>>
d> Why can't yum simply skip this and only do a ldconfig at the end?
The sorting gets pretty hellish; some of the packages which need an
ldconfig run might be needed for scriptlets of other packages.
any example? I have
never saw such a thing before (this does not mean
that it can't exists)
I'm
sure it's technically possible to lower the number of ldconfig runs
assuming that the dependency information of every single package is
completely correct, but I imagine it wouldn't be terribly simple.
- J<
12:12 p.m.
On Sat, 2006-07-22 at 17:42 +0200, dragoran wrote:
I haven't updated my rawhide box for a while and now I started a
yum
update and it takes ages to update the 680 packages. I noticed that
ldconfig is run after a package is installed (because this packages have
it in %post). Why can't yum simply skip this and only do a ldconfig at
the end?
this would save a lot of time in updates like this and would also be
good for anaconda (update from FCX -> FCY)
This isn't something yum is choosing to do. When the install is actually
occurring yum has handed everything off to rpm-lib.
if you want to change this you have to chane it in rpm.
-sv
12:22 p.m.
seth vidal wrote:
On Sat, 2006-07-22 at 17:42 +0200, dragoran wrote:
> I haven't updated my rawhide box for a while and now I started a yum
> update and it takes ages to update the 680 packages. I noticed that
> ldconfig is run after a package is installed (because this packages have
> it in %post). Why can't yum simply skip this and only do a ldconfig at
> the end?
> this would save a lot of time in updates like this and would also be
> good for anaconda (update from FCX -> FCY)
>
This isn't something yum is choosing to do. When the install is actually
occurring yum has handed everything off to rpm-lib.
if you want to change this you have to chane it in rpm.
-sv
ok than change it in rpm (there should be no diff.) the result would be
in both cases faster updates.
12:42 p.m.
On Sat, Jul 22, 2006 at 07:22:31PM +0200, dragoran wrote:
ok than change it in rpm (there should be no diff.) the result would
be
in both cases faster updates.
Bad idea either. What about the following:
package A is installed/upgraded
%post A: runs ldconfig
package B is installed/upgraded
%post B: uses utility of A that needs a library of A
There is no magic intelligence that could see if the %post install
can be postponed, because a subsequent installed package may need
its result (even that of ldconfig) during its installation.
B.t.w., did you measure what (percentage of) time could be won when
collecting the ldconfig calls? I really don't think it's worth the
pain...
--
-- Jos Vos <jos(a)xos.nl>
-- X/OS Experts in Open Systems BV | Phone: +31 20 6938364
-- Amsterdam, The Netherlands | Fax: +31 20 6948204
10:15 a.m.
Jos Vos wrote:
On Sat, Jul 22, 2006 at 07:22:31PM +0200, dragoran wrote:
[...]
B.t.w., did you measure what (percentage of) time could be won when
collecting the ldconfig calls? I really don't think it's worth the
pain...
tell me how and I will be happy to test and show results ;)
10:55 a.m.
dragoran wrote:
Jos Vos wrote:
> On Sat, Jul 22, 2006 at 07:22:31PM +0200, dragoran wrote:
>
> [...]
>
> B.t.w., did you measure what (percentage of) time could be won when
> collecting the ldconfig calls? I really don't think it's worth the
> pain...
>
>
tell me how and I will be happy to test and show results ;)
On top of my head, write a script that samples 'ps' a few times per
second, this should give you a good approximation on the total runtime
of ldconfig over a long yum operation...
12:58 p.m.
dragoran(a)feuerpokemon.de (dragoran) writes:
> B.t.w., did you measure what (percentage of) time could be won
when
> collecting the ldconfig calls? I really don't think it's worth the
> pain...
>
tell me how and I will be happy to test and show results ;)
create a wrapper around /sbin/ldconfig which measures the time. E.g.
| # mv /sbin/ldconfig{,_}
| # cat <<EOF >/sbin/ldconfig
| #!/bin/sh
| exec time /sbin/ldconfig_ "$@"
Enrico
12:58 p.m.
dragoran napsal(a):
ok than change it in rpm (there should be no diff.) the result would
be
in both cases faster updates.
rpm actually already performs this optimization in
the special case of
%scritpletmacro -p /sbin/ldconfig
(not when using
%scriptletmacro
/sbin/ldconfig
).
Mirek
6491
days inactive
6492
days old
9 comments
7 participants
participants (7)
-
Denis Leroy -
dragoran -
Enrico Scholz -
Jason L Tibbitts III -
Jos Vos -
Miloslav Trmač -
seth vidal