@task(
    requests=Resources(
        cpu="3",
        mem="3Gi",
    ),
)
def foo():
    ...



@workflow
def my_wf():
    foo().with_overrides(
        requests=Resources(
            cpu="5",
            mem="5Gi",
        ),
    )

It just took me a while to figure out why this task pod ends up requesting 3 cpus/3Gi of memory instead of 5/5 which I would have expected. The reason is that the task is registered with limits=requests and then the override requests are larger than the original limits which is prevented here. I find this rather unintuitive. I think we should either automatically override the limits as well or log a warning that the user should override the limits too. Do you agree that this is unintuitive and if yes, which option do you think is better?

Would also find that unintuitive. Probably the harder bit is the other case of:

@task(
    requests=Resources(
        cpu="3",
        mem="3Gi",
    ),
    limits=Resources(
        cpu="3",
        mem="3Gi",
    ),
)
def foo():
    ...



@workflow
def my_wf():
    foo().with_overrides(
        requests=Resources(
            cpu="5",
            mem="5Gi",
        ),
    )

As there’s probably no way to know whether

limits

have been set in the first place?

The reason is that the task is registered with limits=requests

Right, I think this is the case especially if no

limits

are specified. In any case, I'd + the idea of warning the user. I think it's a bit more deterministic at scale than automatically overriding the limits.

As there’s probably no way to know whether

limits

have been set in the first place?

If no limits have been set explicitly, the requests are taken also as limits.

they are in propeller i think?

What do you think the behaviour should be here?

I think is on admin

Added to tomorrow's contributors meetup agenda 👍🏽