Re: m68k arch falling ever behind
On Wed, Mar 24, 2004 at 05:25:32PM -0600, Chris Cheney wrote:
> It appears m68k is getting more behind in building packages as each day
> passes. The actual graph shows peaks and valleys, but the overall trend
> appears to be slanted upward fairly steeply. It is currently up to a
> 269 package backlog. Is anything being done to rectify this situation?
Nothing that I know of from official side.
>From my point of view, there are 3 060 machines and a fairly speedy 040
buildd that are forced to be inactive.
To get some advance, one of the 060s is doing a gcc-snapshot build to take
some load off the buildds.
BTW:
m68k is not the only arch with some backlog:
listdir/alpha_stats:Needs-Build : 0
listdir/arm_stats:Needs-Build : 5
listdir/hppa_stats:Needs-Build : 45
listdir/i386_stats:Needs-Build : 0
listdir/ia64_stats:Needs-Build : 0
listdir/m68k_stats:Needs-Build : 277
listdir/mips_stats:Needs-Build : 205
listdir/mipsel_stats:Needs-Build : 118
listdir/powerpc_stats:Needs-Build : 0
listdir/s390_stats:Needs-Build : 0
listdir/sparc_stats:Needs-Build : 165
Quite surprising is the sparc port, imho.
Another interesting number is how many packages are building rigth now:
listdir/alpha_stats:Building : 126
listdir/arm_stats:Building : 169
listdir/hppa_stats:Building : 49
listdir/i386_stats:Building : 2
listdir/ia64_stats:Building : 74
listdir/m68k_stats:Building : 49
listdir/mips_stats:Building : 123
listdir/mipsel_stats:Building : 98
listdir/powerpc_stats:Building : 72
listdir/s390_stats:Building : 216
listdir/sparc_stats:Building : 71
Fetching a huge bunch of packages is IMHO not the most clever idea. Either
some buildds are fetching dozens and dozens of packages in one bunch or some
buildd admins don't have much time to sign and upload packages (or handle
the logs at all).
> One of the packages I maintain, taglib, was uploaded on Mar 10 and has
> yet to even be attempted. It built on most other arches the day it was
> uploaded. I just checked and its still in the middle of the queue on
> m68k, Also I just noticed that some of my other KDE packages are even
> further down in the queue. :< Which means they all have little hope of
> ever being built since the buildds use LIFO ordering. IMHO This is an
> example of why it is not a good idea to build in LIFO order. LIFO
> ordering can be good if your buildd system is capable of ever catching
> up, but as seen with several of the archs, not just m68k, this happens
> very rarely.
It's not the hottest news that LIFO handling is not the most clever thing
to get things done. Again, I don't know of any plans from the official side
to solve these problems.
All I can say:
the backlog that currently exists on m68k, is not the fault of the arch
itself. There *are* problems with the current buildd infrastructure at
large and anyone that denies that, should wake up. I don't think the above
number of needs-build and building show a buildd infrastructure without
problems. These number exists over a longer term, not just 3 or 4 days.
Yes, you're are not the only one who's not satisfied by the current
situation.
--
Ciao... //
Ingo \X/
Reply to: