Are there any patterns for creating Pulumi resource names based on output values? I suspect the answer is "No; don't do that"... we've got a project that we'd like to drive with stack outputs from another project, but aren't really able to create resources in the second stack that have meaningful names, since that information is technically an output from the stack reference. For now, we're just auto-generating resource names (e.g.

resource-{i}

, where

i

is just a counter), but I was curious if others had any guidance to provide here. Thanks!

"No; don't do that". I frequently wrap resources in thin objects that contain their name, in order to work around this. So if I have an array of (for example) policies that I want to attach to an array of roles, but the policies are created / managed elsewhere, then I'll pass around something like a PolicyInfo object that looks like

{ name: string; policy: aws.iam.Policy }

.

so to be clear here we're consuming a stack reference to another stack where resources are created, then creating "derivative" resources from the stack outputs of the other stack. Concretely, we're creating a Confluent Cloud API key, service account, and Kafka cluster in the "other" stack and then creating a Kafka topic and Kafka ACL in the "main" stack, consuming the API credentials and Kafka cluster bootstrap server URL via the stack reference. In particular, I'd like to be able to name the ACLs such that it's clear which service account they belong to, e.g.:

for topic in topics:
        kafka.Acl(
            f"{topic.name}-{service_account_id}-{'write' if write else 'read'}-acl",
        ....  
        )

the

kafka.Acl

name must be a reified

str

, so I'd like to consume the

Output[T]

from the stack reference into a

str

(the particular value of concern here is

service_account_id

). Does that make sense?

Unfortunately there is an execution graph issue here, as stack references are not known to be resolved at the time they are requested. You know that they are, but Pulumi may have to wait for the other stack to finish upping, or wait for the values to come back from a remote backend. It's all asynchronous and can fail, so an Output is required.

hardcoding would be problematic because we'd have to know to change things in two places, the goal is to "drive" the main stack via the other stack

Can you inject the configuration / name into both projects from one place?

that's a good idea, possibly

It does work, you can safely do it. The problem is that

apply()

does not run during a preview. So previews will show that the resources will be deleted, and up will not delete them.

😬 yikes lol

Values retrieved via pulumi.Config are not Outputs, so if you can use that, problem solved?

hmmm, yeah that sounds like a promising approach

You need to use

require

not

get

, since

get

returns a Promise.

yeah, that's totally cool here though, I definitely want it to fail if the value is not present

Thanks @little-cartoon-10569!