Hi, I’m having. a problem with plugin version in python. I create a

pulumi_kubernetes.Provider

using the code below

k8s_provider = Provider(
        name,
        kubeconfig=k8s_config,
        opts=ResourceOptions(depends_on=[node_pool]),
    )

where

k8s_config

are generated from gcp cluster created before this. And all of the k8s resource use this Provider to deploy. Currently, the Provider on stack is using pulumi_kubernetes version 4.5.0, but when I do

pulumi preview

or

up

on my local machine, it keep try to revert it back to version 3.30.2 (which I presume it’s the first version that I installed on my machine). And this triggered the replacement of Provider and all kubernetes resources that depend on that Provider. Below is samples from

pulumi preview -j

{
            "op": "update",
            "urn": "urn:pulumi:prod::foundry-r::pulumi:providers:kubernetes::xxx",
            "oldState": {
                "urn": "urn:pulumi:prod::foundry-r::pulumi:providers:kubernetes::xxx",
                "custom": true,
                "id": "5de2c95a-0f6a-4b59-8a8a-6f3cabb7726e",
                "type": "pulumi:providers:kubernetes",
                "inputs": {
                    "cluster": "xxx",
                    "context": "xxx",
                    "kubeconfig": "[secret]",
                    "namespace": "default",
                    "version": "4.5.0"
                },
                "outputs": {
                    "cluster": "xxx",
                    "context": "xxx",
                    "kubeconfig": "[secret]",
                    "namespace": "default",
                    "version": "4.5.0"
                },
            },
            "newState": {
                "urn": "urn:pulumi:prod::foundry-r::pulumi:providers:kubernetes::xxx",
                "custom": true,
                "type": "pulumi:providers:kubernetes",
                "inputs": {
                    "kubeconfig": "[secret]",
                    "version": "3.30.2"
                },
            },
            "diffReasons": [
                "cluster",
                "context",
                "namespace",
                "version"
            ],
            "detailedDiff": null
        },

This will replace replace everything on kubernetes including namespace, which is not ideal. I’ve tried to delete all pulumi plugin, delete all python package in pip, fix version in requirement.txt, requirement-dev.txt, pyproject.toml. But it still tried to use the old version anyway. My question is 1. Where does pulumi command pull these version from? I’ve seen it even installed 3 different version at the same time in ci logs. 2. Can I fixed the version number so that all the dev use the same version as the ci pipeline? 3. Is there a better way of doing this? It seems really fragile that everything will be replace if I change some version number. Even if I manage to fixed the version number, it still means that I cannot upgrade it later, otherwise it’ll replace everything.

Running

pulumi about

should give details on dependencies. 4.5.0 was yanked (release removed), updating to 4.5.1 should fix the plugin

Looks like delete the whole venv and do

pip install requirements.txt

do the trick. But am I doing this right? I feel like I lock myself to one plugin version. Because if I ever to update the plugin version. It’ll replace everything on my cluster. Are there any best practice doing this?

My understanding of the issue is that it's the preview being over-cautious. To play it safe, you can do

pulumi up --target 'provider urn'; pulumi refresh

Thank you very much, I’ll try this. Is there an issue about this on github that I could follow?