Inform admins via webui that the lander_service_wrapper can't get a PPA

[Ursula Junque]

Just discussed this issue. The rough solution we want to pursue:

1) the lander_service_wrapper that handles getting a free ppa should loop forever until it gets a ppa. This will allow us to only have a single artifact for the ticket (ie the lander won't keep re-launching the job).

2) We include service status into the webui. One of these would be the health of the ppa-assigner (ie ERROR: no free ppa's)

[Francis Ginther (fginther)]

This needs a bit of thought about the best action. Under normal circumstances, it could take a while for a ppa to become free after it's been recycled.

On second thought, I don't think outright failing the ticket is the right answer. Something should keep trying until a sufficiently long enough timeout is reached, or do we wait forever.

[Francis Ginther (fginther)]

The ppa_assigner should provide ppa availability in it's status API (which the web-ui will eventually call).

[Ursula Junque (ursinha)]

I agree it could have a timeout instead of failing right away, but we should update the ticket accordingly so the user knows what's going on. On my setup (that led to this bug) there were no PPAs available for the provided launchpad account, if only there was a way to check if the lp user has suitable PPAs in the beginning and fail right away it would be good (pretty sure that's not possible, though).

[Chris Johnston (cjohnston)]

Just curious, should the ticket fail because a free PPA can't be found? I
don't think we should be failing tickets for infra problems like this. It
isn't the tickets fault, and if we fail the ticket, isn't there a good
possibility that the next tickets will encounter the same issues?

[Chris Johnston (cjohnston)]

On Wed, Feb 26, 2014 at 4:28 PM, Francis Ginther <
<email address hidden>> wrote:

> This needs a bit of thought about the best action. Under normal
> circumstances, it could take a while for a ppa to become free after it's
> been recycled.
>
> On second thought, I don't think outright failing the ticket is the
> right answer. Something should keep trying until a sufficiently long
> enough timeout is reached, or do we wait forever.
>

Unless we can ensure that the problem is ticket related, and that the next
ticket won't fail too for the same reason, I would think we should wait
forever.

[Andy Doan (doanac)]

Just discussed this issue. The rough solution we want to pursue:

1) the lander_service_wrapper that handles getting a free ppa should loop forever until it gets a ppa. This will allow us to only have a single artifact for the ticket (ie the lander won't keep re-launching the job).

2) We include service status into the webui. One of these would be the health of the ppa-assigner (ie ERROR: no free ppa's)