bhouse-nexthop opened a new pull request, #118: URL: https://github.com/apache/cloudstack-kubernetes-provider/pull/118
Lets the CCM in `deployment.yaml` start on a fresh cluster, before the CNI and CoreDNS are up. CloudStack CKS ships this manifest in its binaries ISO. | Change | Why | |---|---| | `hostNetwork: true` | There is no pod network until the nodes are initialised, and the CCM is what initialises them. | | `dnsPolicy: Default` | Use the node's resolver. CoreDNS isn't running yet, and `ClusterFirstWithHostNet` would point at it. As a result, the `api-url` can no longer be a cluster Service name. | | Tolerate the `not-ready` and `network-unavailable` NoSchedule taints | Nodes carry these until the CNI is up. | | Tolerate the `control-plane` and `master` NoSchedule taints, and prefer control-plane nodes | Control-plane nodes may be the only ones available. `master` covers clusters older than 1.25. | | Declare `containerPort: 10258` | On the host network the CCM binds 10258 on the node. Declaring it stops the scheduler from putting two replicas on one node. | | Rolling update `maxSurge: 0`, `maxUnavailable: 1` | With one eligible node, a surge pod could never be scheduled and the rollout would hang. | | e2e comment, `docs/development.md` | Remove the claim that the pod can't use Docker's DNS, which isn't true on the host network. | The existing `uninitialized` toleration is unchanged. Testing - `go build ./... && go test ./...` - Strict-decoded the Deployment with `k8s.io/api/apps/v1`, and parsed every document as YAML. - I didn't run the kind e2e locally; it runs in CI. Fixes #117 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
