You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We have a problem of adopting fluxcd-opeator, because it expects user to explicitly specify cluster domain.
In fact user might have different domains for their clusters, and requirement for explicitly specifying the domain can be ommited which simplifying user experience and lowering down their cognitive load:
For the implementation, this default value should be removed:
The cluster domain is not hardcoded, we just provide a default, in the same way upstream Flux does it. Users can specify a different value in the FluxInstance custom resource.
Also simplifying user experience would result in major performance issues as every single Flux request between controllers will result in x5 DNS calls. Given that upstream Flux does not allow users to remove the cluster domain, neither will the operator.
We did use single DNS name back when we started Flux v2 and it didn't work on GOV private cloud hosted on GKE and AKS clusters, as CoreDNS is configured there to skip resolving names without the cluster domain.
We're preserving all the helm charts and images in our repo, just in case to implement on-fly fixes on them:
The fluxcd helm chart was fixed to not use FQDNs
Currently fluxcd is the only application which requires this.
Just noticed the problem with singlenames, we use kubeapps dashboard which is deployed in different namespace, it takes url from status.artifact.url which is now looking like:
I0617 13:30:45.550692 1 repo.go:653] +indexOneRepo: [cozystack-apps], index URL: [http://source-controller/helmrepository/cozy-public/cozystack-apps/index.yaml]
failed due to: Get "http://source-controller/helmrepository/cozy-public/cozystack-apps/index.yaml": dial tcp: lookup source-controller on 10.96.0.10:53: no such host
Activity
stefanprodan commentedon Jun 16, 2024
The cluster domain is not hardcoded, we just provide a default, in the same way upstream Flux does it. Users can specify a different value in the FluxInstance custom resource.
stefanprodan commentedon Jun 16, 2024
Also simplifying user experience would result in major performance issues as every single Flux request between controllers will result in x5 DNS calls. Given that upstream Flux does not allow users to remove the cluster domain, neither will the operator.
kvaps commentedon Jun 16, 2024
Okay how about using single DNS name?
Looks like go client produces the same amount of DNS requests as in case with FQDN:
stefanprodan commentedon Jun 16, 2024
We did use single DNS name back when we started Flux v2 and it didn't work on GOV private cloud hosted on GKE and AKS clusters, as CoreDNS is configured there to skip resolving names without the cluster domain.
kvaps commentedon Jun 16, 2024
Roger that. I will think how to workaround this issue.
We probably can add the second cluster domain into coredns configuration.
Thanks for expansion
stefanprodan commentedon Jun 16, 2024
How did you used Flux until now? The Flux CLI, TF provider and the community Helm chart all are setting the cluster domain.
kvaps commentedon Jun 16, 2024
We're preserving all the helm charts and images in our repo, just in case to implement on-fly fixes on them:
The fluxcd helm chart was fixed to not use FQDNs
Currently fluxcd is the only application which requires this.
stefanprodan commentedon Jun 16, 2024
If you're using the operator, then it's super easy to drop the cluster domain, just do:
kvaps commentedon Jun 16, 2024
@stefanprodan it seems this is the most elegant solution. Thank you!
kvaps commentedon Jun 17, 2024
Just noticed the problem with singlenames, we use
kubeappsdashboard which is deployed in different namespace, it takes url fromstatus.artifact.urlwhich is now looking like:http://source-controller/helmrepository/cozy-public/cozystack-apps/index-4220e6b6ac4dc145f454f2681cf608bb9f1526288639994b2f83660e61160535.yamland obviously can't reach it
full status: