[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Giving back to a particular buildd



Hi Kurt,

Am Dienstag, den 29.06.2010, 18:30 +0200 schrieb Kurt Roeckx:
> On Tue, Jun 29, 2010 at 11:30:01AM +0200, Joachim Breitner wrote:
> > Am Montag, den 28.06.2010, 23:26 +0200 schrieb Kurt Roeckx:
> > > On Mon, Jun 28, 2010 at 08:36:18PM +0200, Joachim Breitner wrote:
> > > > the Haskell team maintains (besides a lot of unproblematic) a handful of
> > > > package that require a lot of resources (mostly RAM) when building. On
> > > > some architectures, these packages can only be built by certain buildds,
> > > > while they are failing on others. So far, I have observed buildd admins
> > > > just repeatedly giving back packages in question until the "right"
> > > > buildd machine picked it up.
> > > 
> > > What I have noticed on hppa is that ghc is just taking 100% CPU
> > > time all the time.  Trying it on a different buildd is unlikely
> > > to change the result, and setting the timeout higher is also
> > > unlikely to be helping.  All hppa buildds have 4 GB ram or more and
> > > I've only seen ghc use like 250 MB.  I have no idea what the problem
> > > is, but just retrying clearly isn't helping.
> > > 
> > > Not sure what the status is on other arches.
> > 
> > We have identified and worked around one issue¹ and indeed, packages
> > like haskell-src-exts and  haskell-regex-tdfa now build at least on
> > hppa.
> 
> I still see issues with things like darcs.

and pandoc, yes. The fix is applied on a per-package basis (adding a
flag to the compiler), and I first wanted to observe whether it actually
helps. I’ll notify the darcs maintainer that he should consider applying
the patch there as well.

> > Other arches we now hit hard limits ("virtual memory exhausted" on
> > s390, cc being killed which is usually an OOM sign). I hope there is no
> > other principal problem (like the one we just worked around) left.
> 
> 
> > ¹ When ghc grows large, fork() takes longer than the time between two
> > timer interrupts, causing fork() to be restarted again and again. This
> > is fixed in the next ghc release. Until then, it turned out that
> > disabling the timer completely does actually work.
> 
> I think I still see the timer going off on hppa.

With what package? Sorry if I was unclear in the previous mails; the
work-around has been applied to:
 * haskell-src-exts
 * haskell-regex-tdfa
 * washngo

And it yet needs to be applied for to
 * pandoc
 * darcs
 * xmonad-contrib
 * agda

Greetings,
Joachim

-- 
Joachim "nomeata" Breitner
Debian Developer
  nomeata@debian.org | ICQ# 74513189 | GPG-Keyid: 4743206C
  JID: nomeata@joachim-breitner.de | http://people.debian.org/~nomeata

Attachment: signature.asc
Description: This is a digitally signed message part


Reply to: