canonical-ci-engineering team mailing list archive

[Vincent Ladeuil <vila+ci@xxxxxxxxxxxxx>]

>>>>> Andy Doan <andy.doan@xxxxxxxxxxxxx> writes:

    > + Francis and Vincent
    > I don't know enough about package building and PPA's to know what the right
    > thing to do is. Given the limitations described by Ursula below, I want to
    > confirm whether or not we really need clean PPA's?

Yes, we do need clean PPAs.

At the very least, we want PPAs that can be relied upon for:

- not providing a package that hasn't been built yet for the user needs,

- providing an up-to-date package to the user.

    > If we need clean PPA's we might need the PPA-assigner to have 3 states:
    >   clean-and-available
    >   dirty-and-available

No doubt.

    >   locked

Assigned would be less controversial IMHO. It also implies handling
*who* is using the PPA, since when, something else I can't think of
right now ;)

    > We could then have some async/background logic that could get a PPA from
    > dirty-and-available to clean-and-available. Or some other approach?

That sounds like the best approach to me, especially if we don't control
how long it takes to clean a PPA.

This will require some tuning to ensure we always have enough clean PPAs
available.

This, in turn means we should be ready to handle an empty
pool gracefully. By gracefully I mean two things:
- users of the PPA assigner will in turn fail gracefully,
- the error message should be crystal clear for both the cause and the
  possible remedies from the user pov.

  Vincent