Re: Heads Up: cirrus-ci is shutting down June 1st

Hi,

On 2026-06-03 19:36:29 -0400, Andres Freund wrote:
> > Concretely: I propose that we bail out of the setup step if the
> > repository isn't postgres/postgres, just for the initial committed
> > version, and then we can test what this actually does in practice to
> > our downstream forks. If I'm being overly paranoid, we can immediately
> > remove it; else we can add an opt-in. But adding it after the fact
> > won't protect anyone who synced up in the interim.
> 
> We clearly would need to have an opt-out from that from the get-go, otherwise
> I couldn't even test that things are still working before merging, and
> postgresql-cfbot won't work...  Thomas has it otherwise mostly ready to go
> (this thread actually is being tested automatically already).


Attached is a possible implementation of this.  If the PG_CI_ENABLED
repository variable is not set to 1, we run as little CI as possible.

To make that less confusing, emit a summary whenever we skip running CI, with
a message explaining how to enable CI.  See e.g. the bottom of
https://github.com/anarazel/postgres/actions/runs/26926523027

Unfortunately this summary is only visible when logged into github.


Attached is a squashed version of my earlier changes. In addition:

- src/tools/ci/README was updated with the knowledge that the workflow
  may or may not be enabled in a fork

- a few typos etc were fixed

- the change above was added as a third patch


I'm on the fence whether we want the third change or not.

I'd like to push the CI support tomorrow morning.  Perhaps othes will chime
about whether we should require this explicit opt-in or not...

Greetings,

Andres Freund