builds which require more memory than a given buildd still try that buildd

[Jani Monoses]

These packages currently time out after 300 seconds of no stdout activity on armel, probably because swapping

Trying them on the new armel builders which have more memory may get them past the point of failure.

However retrying from LP UI does not allow selecting the machine so it is likely wasteful to keep retrying as these builds take > 5 h
so manual intervention is needed.

https://launchpad.net/ubuntu/+source/haskell-src-exts/1.9.0-2build1/+buildjob/2147354
https://launchpad.net/ubuntu/+source/trilinos/10.4.0.dfsg-1build1/+buildjob/2100348
https://launchpad.net/ubuntu/+source/ants/1.9+svn532-5/+buildjob/2004367
https://launchpad.net/ubuntu/+source/washngo/2.12.0.1-4/+buildjob/2009067

[Julian Edwards (julian-edwards)]

This is not a buildd bug, the dispatching is done in launchpad code.

[Jani Monoses (jani)]

I was under the (mistaken?) impression that the armel build servers have varying amounts of RAM. If that is not the case then it does not really matter where the build is restarted I guess. Not easy to tell, since the only public info about the builders are their name and arch.

[Julian Edwards (julian-edwards)]

On Thursday 20 January 2011 16:37:23 you wrote:
> I was under the (mistaken?) impression that the armel build servers have
> varying amounts of RAM. If that is not the case then it does not really
> matter where the build is restarted I guess. Not easy to tell, since the
> only public info about the builders are their name and arch.

Yes, but the choice of builder is made on the Launchpad side.

In the meantime, we've doubled the timeouts so we should see fewer failures.

[Jani Monoses (jani)]

Can the packages mentioned in this bug be manually given back with even more timeout as a one-time try to get them built for natty? For oneiric the build servers will hopefully be more capable and these issues will no longet manifest.

[Jani Monoses (jani)]

The real issue here is some packages take too long to build and with the current ARM builders no reasonable amount of giving back and timeout increases can make this work. They should be built when the Panda cluster is set up.