diff options
| author | Aleksandr Nogikh <nogikh@google.com> | 2025-04-16 12:25:32 +0200 |
|---|---|---|
| committer | Aleksandr Nogikh <nogikh@google.com> | 2025-04-17 08:16:37 +0000 |
| commit | 2a6ededbf54a9f8ac036ad0ebfc673934f93fde9 (patch) | |
| tree | f861c782b9e6799aa6cb88ef8d6e38051c2d08bf | |
| parent | 229db4cff1245e6c493ba0cb4d53e799a041015e (diff) | |
syz-cluster: enforce pod to node assignment
On GKE, we use a separate node pool that supports nested virtualization.
Taints and tolerations (that were used before) only make sure that no
other pods are scheduled there, but are not enough to make sure that the
pods that do need nested virtualization will end up there.
Use nodeSelector to force the affinity.
| -rw-r--r-- | syz-cluster/overlays/gke/kustomization.yaml | 11 |
1 files changed, 4 insertions, 7 deletions
diff --git a/syz-cluster/overlays/gke/kustomization.yaml b/syz-cluster/overlays/gke/kustomization.yaml index bdf53c431..ed2846a29 100644 --- a/syz-cluster/overlays/gke/kustomization.yaml +++ b/syz-cluster/overlays/gke/kustomization.yaml @@ -13,7 +13,7 @@ resources: patches: - target: kind: WorkflowTemplate - name: boot-step-template + name: (boot|fuzz)-step-template patch: |- - op: replace path: /spec/templates/0/tolerations @@ -24,12 +24,9 @@ patches: effect: "NoSchedule" - target: kind: WorkflowTemplate - name: fuzz-step-template + name: (boot|fuzz)-step-template patch: |- - op: replace - path: /spec/templates/0/tolerations + path: /spec/templates/0/nodeSelector value: - - key: "workload" - operator: "Equal" - value: "nested-vm" - effect: "NoSchedule" + cloud.google.com/gke-nodepool: nested-vm-pool |