Is it possible to extend

dagster.OpExecutionContext

to add a member to the

context

without using a resource definition? I'm considering ways to implement some metrics and pretty much want the dependency to always be available just like the

log

member is in the context.

hey @Mike Atlas I think that would be pretty difficult since the dagster code creates and provides the

context

object for each op/asset and there isn’t a way to customize how that context is created. one thing you could consider is writing a custom decorator that wraps

@op

or

@asset

that always adds a

required_resource_key

for the member you want to add. then that would be available to all of your ops and assets

Nice idea @jamie thanks. Got it... yeah, I've already gone off and started implementing my metrics object as a resource, since it's the lowest friction way I can see to achieve it

yeah i could see that being useful. We’re doing some work on the resource system in early 2023, so if you want to open a GH feature request for the global resource idea we’d definitely consider it as part of the revamp

will do ✅

yeah that might work. I’m not deeply familiar with custom loggers, so I don’t know what, if any, issues you might run into with that. @owen might be able to help a bit

User-defined custom log levels are not supported, and calls to, e.g.,

context.log.trace

or

context.log.notice

will result in hard exceptions at runtime.

hmm maybe not ahaha

yeah that all sounds pretty unpleasant. if the custom decorator method seems possible for you i’d give that a try! we did something similar at my previous company and it worked well

I think my enhancement request will just be to support a generic telemetry provider, such as https://opentelemetry.io/docs/instrumentation/python/