- Shipped
- September 25, 2026 at 1:23 AM UTC
- Author
- Kamo
- Commit
- 9d292ff
k1m1 reaches /mnt/worldchunks only through a hard NFS mount of k3m1. That mount has hung, so the kubelet on k1m1 cannot stat the earthdata hostPath. A rollout that landed there sat in ContainerCreating while Recreate had already stopped the old pod, which took DaemonService down for eight minutes. A k1m1 placement buys no failover: with k3m1 down, the NFS mount is dead too. Already live through kubectl patch. The image is unchanged, hence [skip ci]. DeploymentManifestTest pins the node, and pins the order of the pod's own NATS login after the shared one.