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

Re: Example where testing-security was used?



On Tue, May 31, 2005 at 11:25:39AM +0200, Adrian Bunk wrote:
> On Mon, May 30, 2005 at 10:56:16PM +0200, Marc 'HE' Brockschmidt wrote:
> > Hamish Moffatt <hamish@debian.org> writes:
> > > On Mon, May 30, 2005 at 11:48:54AM +0100, Mark Brown wrote:
> > >> On Mon, May 30, 2005 at 12:34:21PM +0200, Adrian Bunk wrote:
> > >>> But setting up autobuilders doesn't require a new infrastructure
> > >>> (and shouldn't require more than half a year).
> > >>> Wasn't the infrastructure a prerequisite for woody and is working?
> > >> It turned out that the central part of the existing infrastructure
> > >> didn't scale up well enough to cope with the new architectures in sarge.
> > > There are no new architectures in sarge.

> > That's right, but the buildd network has to work for both oldstable and
> > stable. potato + woody didn't need as many buildds as woody + sarge
> > will need.

> 17 -> 22 architectures is an increase, but doesn't look like a very 
> serious one.

There were never security autobuilders for potato; and security and
proposed-updates are separate queues.  So in terms of centralized load on
the wanna-build server, this is a jump from 22 (11 stable-security + 11
proposed-updates) to 33 (11 oldstable-security + 11 stable-security + 11
proposed-updates; AFAIK there is no oldstable-proposed-updates).

If testing-security is brought on-line again for etch within the year
following sarge's release (as I certainly hope it will), the peak number of
wanna-build *databases* being served by ftp-master.d.o (saying nothing of
the number of actual buildd connections) would be 66 (oldstable-security +
stable-security + proposed-updates + testing-proposed-updates +
testing-security + unstable, x 11 archs -- not counting prospective archs).
The greatest number newraff has ever really been asked to handle at any one
time up to this point has been 55 (the number we have currently), which was
only done *after* newraff's scalability problems were addressed; prior to
that, AFAIK there were only ever as many as 44 databases actively used (p-u
+ s-s + t-p-u + u, from the release of woody w/ security autobuilder support
up until this spring).

So at 44 the server was already at its limit, the release required a 25%
increase in the number of databases (and roughly the same increase in the
number of connections), and etch would have brought us up to 50% over that
limit.

-- 
Steve Langasek
postmodern programmer

Attachment: signature.asc
Description: Digital signature


Reply to: