[Mesa-dev] [AppVeyor] mesa master #6163 failed
Eric Engestrom
eric.engestrom at imgtec.com
Thu Nov 16 15:56:35 UTC 2017
On Thursday, 2017-11-16 14:20:03 +0000, Jose Fonseca wrote:
> On 16/11/17 14:08, Jose Fonseca wrote:
> > On 15/11/17 12:20, Eric Engestrom wrote:
> > > On Wednesday, 2017-11-15 10:35:04 +0000, Jose Fonseca wrote:
> > > > I could post a remark on AppVeyor's forums, but it sounds from the
> > > > logs the issue could be on our (freedesktop.org's) side.
> > > >
> > > >
> > > > I've reconfigured AppVeyor to use
> > > > https://anongit.freedesktop.org/git/mesa/mesa.git/ instead of
> > > > git://anongit.freedesktop.org/mesa/mesa . Perhaps fewer things can go
> > > > wrong when using HTTP.
> > >
> > > I seem to remember cloning through http was much more cpu intensive,
> > > which was an issue on the fdo servers; I'm not sure we want to do that
> > > for each commit now.
> > >
> > > We can keep that for a while though, if only to see if it makes
> > > a difference in appveyor.
> >
> > HTTP doesn't seem to make a difference:
> >
> > https://ci.appveyor.com/project/mesa3d/mesa/build/6174
> >
> > I'll revert to git protocol.
> >
> > Still no solution for these errors though.
>
> I just had an idea: is anongit.fdo an alias to git.fdo, or a separate
> server?
from a quick `dig`, git.fdo is on kemper, while anongit.fdo is on molly
>
> Could it be the problem here is that whatever is syncing git -> anongit
> leaves the tree temporarily inconsistent? If so I wonder if anything
> changed recently.
Could be, although git objects should remain valid forever since we
don't rewrite history, so the only thing I could see would be a sync
issue if molly copies the index and starts serving it before the objects
are copied.
fdo admins: is the sync between kemper and molly done via git, or some
external tool like rsync?
Jose, I think making appveyor use github might be a good idea regardless
of the fdo sync issue, as it would offload the per-commit-clone work,
and that's exactly the kind of things CDNs like github are for.
More information about the mesa-dev
mailing list