[ofiwg] replacing travis CI

[Jason Gunthorpe]

On Mon, Jun 21, 2021 at 01:58:19PM -0400, Doug Ledford wrote:
> On Mon, Jun 21, 2021 at 11:06 AM Jason Gunthorpe <jgg at ziepe.ca> wrote:
> >
> > On Mon, Jun 21, 2021 at 04:44:29AM +0000, Hefty, Sean wrote:
> >
> > > > We pay a small nomial Azure fee to host a private container registry
> > > > and the CI runs its images out of this.
> > >
> > > I'll look into Azure.  Who is 'we' in the we pay?  OFA?
> >
> > We share the Azure tenant and billing with the UCF
> >
> > > OFA is supposed to have it's own testing cluster.  It would be nice
> > > if that could run this sort of CI.
> >
> > IMHO the managed services are a lot better than trying to run your own
> > hosted stuff with no staff...
> 
> The FSDP has staff ;-)
> 
> The goal for the FSDP is to be able to do all of this and to be able
> to test directly on bare metal.  I know we talked about this in the
> past, but you have to use Azure as the control service and feed it
> back to azure runners on local machines in order to get the bare metal
> testing you want, yes?  Something like that could be done at the FSDP
> too, but we haven't prioritized that.

Yah, UCX is doing that:

https://dev.azure.com/ucfconsort/ucx/_build/results?buildId=19941&view=results

The test box with 'XX on worker' is running on someone's on-prem
infrastructure.

Wow, I'm amazed how much that has grown.. There must be > 100 servers
in there now.

If FSDP wants to donate some VMs to run pyverbs tests on real HW then
this would be the way to plug it into rdma-core CI.

Jason