NVIDIA / holodeck / 22947311788

54%

Build Type

push

github

Committed by web-flow

Commit Message

fix: pre-release validation fixes for kubeadm, networking, and CTK (#716)

* fix: use local IP for kubeadm init health checks (single-node)

kubeadm v1.33+ validates API server health via the control-plane-endpoint
URL. When set to EC2 public DNS, this can timeout because the public DNS
isn't routable from within the instance during init. Use the node's private
IP for init and include the public DNS in cert SANs for external access.

Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

* fix: HA kubeadm init uses local IP to avoid NLB chicken-and-egg

kubeadm v1.33+ health checks reach the API server via the control-plane-
endpoint. With NLB, this creates a deadlock: NLB can't route until API
server is healthy, but kubeadm won't report healthy until NLB routes.
Fix: init with local private IP, include NLB DNS in cert SANs, then
update kubeadm-config and admin.conf to reference NLB after init.

Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

* fix: create nvidia-container-runtime symlink on RPM platforms

nvidia-container-toolkit provides nvidia-ctk but may not provide the
nvidia-container-runtime binary on RPM distros. Container runtimes
(containerd, CRI-O) require it. Create symlink if missing.

Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

* fix: truncate NLB target group name to AWS 32-char limit

AWS enforces a 32-character limit on target group names. Long holodeck
cluster names could exceed this, causing NLB creation to fail.

Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

* test: add explicit K8s version to RPM test configs

Without an explicit version, RPM tests pull the latest K8s release which
may be incompatible with the test infrastructure.

Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

* fix: use public IP for control plane endpoint when no LB

determineControlPlaneEndpoint returned PrivateIP when no LoadBalancerDNS,
making kubeco... (continued)

Run Details

44 of 113 new or added lines in 6 files covered. (38.94%)

1 existing line in 1 file now uncovered.

2891 of 5822 relevant lines covered (49.66%)

0.56 hits per line