Cancel requests which are not possible

[leszekhanusz]

Currently, if you request an image using a model,
and this model becomes unavailable (last worker proposing it disconnected),
then you will stay at an estimated time varying between ~5 or 10 seconds,
without any indication that you're waiting for nothing.

Feature requests:

• Directly refuse any request with a model which is not available
• Cancel all the requests with a model if that model become unavailable

[db0]

There is always a possibility that the model will come online though, in face I have a plan in mind to make workers load new models as demand changes in the waiting queue.

[leszekhanusz]

ok, but currently you can request in the API any model name, even if it does not exist and it will go through.

[leszekhanusz]

Some other cases to consider:

• providing an invalid worker id should instantly return an error instead of going through
• if you ctrl-c your worker during a generation, the client will wait indefinitely instead of receiving an error or sending back the request to the horde, even if you restart your worker right away

[leszekhanusz]

And it is also the case if you try to request an image on your own worker with a width/height too big for your config

[leszekhanusz]

Basically we should follow the fail-fast principle.

[db0]

I don't disagree. All these stuff is in my plan :D

[db0]

• Request will be aborted if a bad worker_id is given
• Jobs whose workers never returned, eventually expire and restart
• models can still be specified freeform, but /check now returns a field is_possible, which is False when there's no available workers to fulfil the request (so you can notify the user to cancel)