Open glena opened 3 months ago
Seems like it would be pretty straightforward to write a Kubernetes operator that acted similarly to the self-hosted agent, dispatching deployments as Kubernetes Job
s.
An argument could be made for or against doing the work in the Pulumi Kubernetes Operator (as opposed to a new operator). A possible benefit would be a unified backend for running deployments driven by Pulumi Cloud or by the Stack
CRD.
I would advocate for a layered design based on a low-level CRD, similar to JobDefinition
, that would cause the operator to spawn a Kubernetes Job
to execute the defined job. The status block would reflect the status of the step(s). This becomes a building block to be used in two ways:
Agent
, that takes jobs from the Pulumi Cloud.Stack
CRD.I am not sure why we would need a CRD here for this, rather than just a deployment that is capable of creating Jobs.
Hello!
Issue details
We have a naive example on how to run agents in K8s (https://github.com/pulumi/customer-managed-deployment-agent/pull/3).
Some customers want proper k8s support using native k8s resources (ie: having an operator dispatching jobs)
Affected area/feature