Delete AKS cluster failed

[serbrech]

I first thought we have a bug in AKS for old clusters, but I could delete it just fine in the portal or CLI.

I think you're just not including an api-version in the delete command for some reason.

{
  "error": {
    "code": "NoRegisteredProviderFound",
    "message": "No registered resource provider found for location 'westus2' and API version 'MISSING' for type 'managedClusters'. The supported api-versions are '2017-08-31, 2018-03-31, 2018-08-01-preview, 2019-02-01, 2019-04-01, 2019-06-01, 2019-08-01, 2019-10-01, 2019-11-01, 2020-01-01, 2020-02-01'. The supported locations are 'eastus, westeurope, francecentral, centralus, canadacentral, canadaeast, uksouth, westus, westus2, australiaeast, northeurope, japaneast, japanwest, eastus2, southcentralus, northcentralus, southeastasia, australiasoutheast, ukwest, southindia, centralindia, eastasia, koreasouth, koreacentral, southafricanorth, brazilsouth, germanynorth, switzerlandnorth, switzerlandwest, germanywestcentral, uaenorth, norwayeast, norwaywest'."
  }
}

[serbrech]

tried to figure out where I could track this issue, but I'll wait for you to see this and point me in the right direction.
A drawback of generated codebase, it's hard to figure things out ;)

[lawrencegripper]

Interesting! thanks for raising this

@stuartleeks I you have a good starter for 10 on this one?

[github-actions]

Flagging stale issue. Actions will close this issue in the next 5 days unless action is taken.

[lawrencegripper]

Still need to look into this one. Things a bit hectic without childcare at the moment so behind on stuff.

[lawrencegripper]

So looks like this isn't directly related to the auto-gen code. Have a working theory I'm looking at.

Currently the ARMClient uses the Providers API call to get back all the supported API versions. It then rather simplistically builds up a map of RP->Version or if it can't adds Missing as the API Version. My guess is something went wrong with populating the provider dictionary or a race occurred and the provider dictionary was still being populated while the delete was called.

azbrowse/pkg/armclient/armclient.go

Lines 207 to 213 in dd44134

	func GetAPIVersion(armType string) (string, error) {
	value, exists := resourceAPIVersionLookup[armType]
	if !exists {
	return "MISSING", fmt.Errorf("API not found for the resource: %s", armType)
	}
	return value, nil
	}

azbrowse/pkg/armclient/armclient.go

Lines 242 to 252 in dd44134

	resourceAPIVersionLookup = make(map[string]string)
	for _, provider := range providerResponse.Providers {
	for _, resourceType := range provider.ResourceTypes {
	for _, apiVersion := range resourceType.APIVersions {
	if !strings.Contains(apiVersion, "preview") {
	resourceAPIVersionLookup[provider.Namespace+"/"+resourceType.ResourceType] = apiVersion
	break
	}
	}
	}
	}

Few things we could do better here:

1. Better error handling when an API version is missing. Showing a status message, returning "MISSING" is the wrong thing to do here.
2. Check that the API Version is available in the location of the resource. It's possible that an API version may be picked which isn't available in a region as the cache is built up simplistically.
3. Check that population of the Provider dictionary has been completed before allowing calls.

[lawrencegripper]

Had a quick test and with current version I can delete an aks cluster successfully. This likely points to the race condition with the provider call taking too long meaning the version lookup fails.