I think its a really good idea from the kubernetes executor perspective.
There are a lot of elements that should be automated when launching a pod
(i.e. a bash command wouldn't need the same amount of compute as building a
model with scikitlearn)
@Alexander
> don't you think that you give too
much
I think it's a good intention, but don't you think that you give too
much power to operators to configure how executeros should acts like?
Why executtor itself can't decide which config to apply per queue
basics? It's a question of ops domain, not devs.
--
,,,^..^,,,
On Sun, Aug 6, 2017 at 8:2
Basic proposal:
Change the queue column in the TaskInstance table from a varchar(n) => pickle.
Additionally rename the column from queue => executor_config.
Why?
Currently the `queue` column is only used for the CeleryExecutor since the
concept of a `queue` doesn't make sense for other execut