Did something recently changed with regards to precedence rules of provider attributes in the kubernetes (typescript) provider? I recently upgraded to

@pulumi/kubernetes: 0.23.1

and seeing some odd behaviour: We instantiate provider instances dynamically via

new gcp.Provider({kubeconfig: ..., namespace: 'namespace-A'})

and then use this provider instance in some resources like

new k8s.core.v1.ServiceAccount(foo, {metadata: {namespace: 'namespace-B'}}, {provider: k8sProvider})

. I would expect the serviceaccount to be created in

namespace-B

in the example above, but it actually gets created in

namespace-A

. I can create a more detailed repro if necessary but wanted to ask first if this intentional or a known issue? @creamy-potato-29402?

If I recall correctly the

namespace

property on

Provider

instances used to be ignored, and was recently changed to actually set a default namespace. That said, I agree it should not override the namespace if provided explicitly on a resource. Could you open an issue? This sounds like a bug to me.

ok makes sense. Yep, let me open an issue

Actually, that is the intended behavior, currently. https://github.com/pulumi/pulumi-kubernetes/pull/538/files#diff-5d82f691515dea3a9f870cf8d1a0fc84R74 https://github.com/pulumi/pulumi-kubernetes/pull/506#issue-265491527 Feel free to open an issue if you disagree with that design, and we can discuss further. If you don’t want the resource to be in that namespace, you can create a new provider that references the same kubeconfig/cluster without the namespace arg set.

just opened the issue https://github.com/pulumi/pulumi-kubernetes/issues/576