[epic] Initial kubectl operator
plugin updates for operator-controller and catalogd
#1424
Labels
epic/kubectl-operator-plugin
epic
supporting-tools
v1.x
Issues related to OLMv1 features that come after 1.0
With the imminent 1.0.0 releases of operator-controller and catalogd, we should circle back to our
kubectl-operator
plugin, update it to work with operator-controller/catalogd APIs and make those interactions the default experience (deprecating the OLMv0 implementation).One of the most important new features the kubectl plugin should include is a way to query the contents of the
ClusterCatalog
objects on a cluster. This is a tricky problem to solve because the networking and authentication stacks between an off-cluster client and the on-cluster catalogd service are often environment-specific. Potential options include:kubectl port-forward
, but this requires users to have permission to get information about the catalogd pod (and possibly its service) in order to setup the port-forward connection.kubectl proxy
, but no client authentication to the in-cluster service is supported. This would be a problem if we ever want to require client authentication in order to access catalogd from outside the cluster. Are there any advantages ofproxy
overport-forward
that we should consider?NodePort
- this configures each node to start a listener on the specified port for the specified protocol, and all nodes proxy connections to the in-cluster service. Downside is that it requires a port reservation, and clients would need a way to discover the node IPs and port assignment (not insurmountable, but something we'd have to consider carefully).Because all of these options have pitfalls, we may need to design a flexible solution that enables distributions that include OLM to select which of these to support. And ideally the client could automatically discover which technique to use. Since clients will be required to query for
ClusterCatalog
to know what is even available to query, perhaps catalogd could include external URLs in theClusterCatalog
status?In addition to the challenges with exposing catalog content off-cluster, we should also implement some or all of:
We should definitely plan to implement a really nice UX for discovering the current state of what is actually present on the cluster. We can implement custom printer columns, sorting, etc. that can really enhance a user's understanding of the state of the system.
There may be additional iterations on this effort (for e.g. as we discover possible benefits of new printer columns or something else), but for this initial iteration we plan a functional breakdown to give solid experiences in stages. This may enable other features (for e.g. service account permissions evaluations).
These phases are captured as sub-issues, and duplicated here for those expecting a table:
The text was updated successfully, but these errors were encountered: