I am having trouble passing sensitive information to my flow's run configuration. Having:

secret = PrefectSecret("SOME_SECRET")
flow.storage = S3(bucket="some-bucket", secrets=["AWS_CREDENTIALS"])

flow.run_config = KubernetesRun(
    image="<http://registry.gitlab.com/<some_image>|registry.gitlab.com/<some_image>>",
    labels=["prefect-14"],
    env={
        "PREFECT__LOGGING__LEVEL": "DEBUG",
        "SECRET": secret
    },
    image_pull_secrets=["image-puller"],
)

and then running:

flow.register("ProjectName")

I get:

TypeError: Object of type PrefectSecret is not JSON serializable

Any thoughts on this?

Secrets need to be JSON Serializable. Is your secret a list or dict?

It's a string saved on Prefect Cloud

I think if it’s outside of the Flow, you need to do

PrefectSecret(…).run()

. Could you try that?

Running the

PrefectSecret

run

method solves the

TypeError

. However, I can see in the console that it returns the value of SOME_SECRET.

You’re right that it’s not. The PrefectSecret is really meant to be used in a Flow context. What are you using the secret for here that it needs to be passed as an env variable? Is it something that is needed to start the Flow run?

from prefect.engine.executors import DaskExecutor
import coiled 

executor = DaskExecutor(
    cluster_class=lambda: coiled.Cluster(**{
        "software": "somesoftwareenviron",
        "shutdown_on_close": False,
        "name": "prefect-executor",
        "n_workers": 1, 
        "credentials": "local",
    })
)
flow.executor = executor

I'm trying to provision a Coiled cluster. I have a kubernetes agent sitting on a K8s cluster.

I’m not sure I’m following. Are you looking to pass the secret into

credentials

there?

No, the

credentials

argument is not the actual credentials, but is used to indicate which set of credentials to rely on In the docs:

Ah ok. Sorry, I’m not following how the Coiled cluster is related to the PrefectSecret?

In order to provision the Coiled cluster I need to authenticate using a token. I'd like to pass in the token as a

PrefectSecret

.

Ahh ok I understand. So I believe passing it through the RunConfig will have the secret exposed in some way. You can try adding it to your storage:

flow.storage = S3(bucket="some-bucket", secrets=["AWS_CREDENTIALS", "SOME_SECRET"])

and I think this will fetch it from Cloud when Prefect is loaded. Will try it in a bit.

Thanks, I'm gonna try this and get back to you.

Thanks for letting me know. I think you’re good but I’ll look into this more also.

@Michail Melonas you might have tried this already, but after you added

secrets=["COILED_TOKEN"]

to Storage, I'm wondering if you passed the secret from

prefect.context.secrets

into your custom function for creating the Coiled cluster: i.e.

def coiled_cluster():
    import dask
    dask.config.set({"coiled.token": prefect.context.secrets.COILED_TOKEN})
    return coiled.Cluster(**{
        "software": "somesoftwareenviron",
        "shutdown_on_close": False,
        "name": "prefect-executor",
        "n_workers": 1,
    })

flow.storage = S3(bucket="some-bucket", secrets=["COILED_TOKEN"])
flow.executor = DaskExecutor(cluster_class=coiled_cluster)

This pattern follows ideas from passing the "main image" (i.e. Docker image specified on run config for the Agent / flow runner) into the DaskExecutor (which manages the task runner) at runtime (see https://docs.prefect.io/orchestration/flow_config/executors.html#specifying-worker-images) I'm planning to test this pattern in the weekend, but wondering if you've tried this already (before you came to your Docker ENTRYPOINT solution)

Hey, @Chris L.. I have not tried this approach to passing in the secret. Let me know what you find

I believe this should work, just code the token might be exposed as plain text here

@Kevin Kho hi Kevin, could you kindly clarify where and when this code might be exposed? My assumption is that it shows up once on the agent as part of the process that starts up the executor. I don’t imagine the secret is saved to hard disk (or to any logs) on the agent after the flow run completes and the executor is torn down?

Not 100% sure here but my best answer is: It’s not saved to hard disk yep and you just have to be careful about not logging it. I think this is fine because it only runs on the client, but passing data to Dask workers in general does not encrypt stuff. So some people have concerns that this is being passed over the network