Would it make sense to bundle the

seed

contents into the

scheduler

?

Since

depends_on

doesn’t do anything in docker-compose files version 3+ (and we’re on 3.7), it seems like we should be ensuring that the files are initialized properly as a dependency of the scheduler.

Coordination between the two doesn’t seem to make a ton of sense to me, but I also don’t remember if there was a practical reason why we split out

seed

in the first place.

What do you think @charles?

both server and schedueler rely on seed.

does that mess up what you just proposed?

hmm

kind of

I mean the scheduler depends on the server to run before doing anything too

Same implicit dependency, just on a different system

Server instead of seed

woof. this feels brittle to m.e

it's hard for me to understand how they could have just not replaced dependson.

From the kube perspective you run into the same problem.

No depends on

apps are responsible for their own connect/reconnect logic

i see

which is probably the same reasoning as docker-compose I imagine

so can we just leave the seed container

but then have anything that tries to connect to config respository fail / retry if it can't find the files it needs.

seems like that's the pattern we're supposed to use?

Yeah we can do that too.

That seems fine

And in kube we can do that as a one time job

instead of a deployment