Re: Test instance of our infrastructure
Dear Ian,
I just thought I could give my 2c to this idea (which I like and share).
On 28 November 2016 at 13:04, Ian Jackson
<ijackson@chiark.greenend.org.uk> wrote:
> Should we not have public test instances of all these things ?
Yes.
> I suggest we should declare (perhaps as a DEP?) a systematic scheme
> which recommends to infrastructure operators answers to the following
> questions:
(...)
>
> My starting points for answers to these questions are something like
> this:
>
> * Most of our services are addressed via domain names,
> *.debian.org or *.debian.net. Test instance of a service are
> correspondingly at *.infratest.debian.{org,net}.
In order to keep configuration changes to some files (e.g. Apache
vhost files) to the minimum, I would suggest to setup the test
instance of the service using a prefix to the DNS name.
In some environments I've seen the use of 't-' (for test environment)
and 'a-' (for acceptance environment) when defining DNS service names
and would like to suggest a similar approach here, so:
Production service: SERVICE_NAME.debian.org
Test service: t-SERVICE_NAME.debian.org
> * Test instances should talk to the test instances of other
> services. For example bugs.infratest.debian.org should track
> package data from ftp.infratest.debian.org.
Agreed. Please note that if we have a standard "naming" convention for
test instances (as above) it would make it easier to search/replace
configuration bits that refer to backends or other services and ensure
consistency between test services' configuration and productio
nservices.
> * Test instances should normally be up, but may be down or broken
> or something when they are being worked on.
I actually would suggest some test instances to be down by default (to
prevent people from pointing towards them and use them "for real") but
have the means to start them up easily by either the service managers
or DSA for testing.
> * I don't know about hosting arrangements. We should ask DSA's
> opinion. I think that test instances should run on a different
> host to the live instance, so that security bugs in test instances
> are not so much of a concern.
I think it should be OK to have test instances be on the same server
as the production service. The service manager should define its
requirements (do I need a test instance to test Debian upgrades? to
test service configurations? to test new software releases?) and to
propose what best suits its need.
For example in my case (manpages) I'd rather have both test and
production on the same server. This would make it easier to move
changes from test to production, and there are no real requirements to
test the current setup with new Debian releases. But being in
different servers if DSA believes it would be better would also be OK
.
Best regards
Javier
Reply to: