[OpenAFS-devel] Testing (and a call for volunteers)

Simon Wilkinson sxw@your-file-system.com
Thu, 20 Sep 2012 11:52:51 +0100


On 20 Sep 2012, at 00:59, Booker Bense wrote:

> On Sun, Sep 16, 2012 at 9:53 AM, Troy Benjegerdes <hozer@hozed.org> =
wrote:
>> What about launching a couple of VM's using kvm/qemu? (this is where =
I'm
>> going at the moment)
>>=20
>=20
> In the buzzword compliant world[1], tools like vagrant are being used
> to create exactly this kind of testing.

The main problem here is that our developers use a wide variety of test =
platforms - we need 'make check' (at least) to be able to run on all of =
those platforms. make check also needs to be sufficiently fast that it =
can be a pre-commit smoke-test - a test suite that takes hours to =
execute will just never get run if that's on a pre-commit test suite.

Of course, that doesn't rule out doing all of this stuff on a =
post-commit basis (say nightly), and having it report results back. But =
someone needs to volunteer to build, develop and maintain this =
infrastructure, and we seem to be very short of actual volunteers at the =
moment.

> Spinning up a "mini-cell" with a single command would be a huge step =
forward.

My hope is that make check will be able to spin up a mini-cell on the =
developers machine. But that's limited to one ptserver, one vlserver, =
one fileserver and so on. We can't test things like volume moves, ubik =
replication and so on with only a single machine. And I don't believe =
that spinning up VMs to do so is feasible for the reasons I outline =
above.

Cheers,

Simon