Softly prefer large CPU workers for Traefik replicas
This commit is contained in:
1 parent
b490637de3
commit
b9ad237fc3
2 files changed
+16
No files matched your search
@@ -18,3 +18,10 @@ because the additional Service uses cluster-wide endpoint routing.
|
||||
Traefik CRDs and Gateway API CRDs are cluster-wide prerequisites. Keep the
|
||||
existing CRDs during adoption and upgrades; they are not safe cleanup targets.
|
||||
Application-specific ingress resources remain with their owning application.
|
||||
|
||||
## Placement
|
||||
|
||||
Traefik softly prefers workers labeled `capacity.brunner.ninja/cpu-class=large`.
|
||||
The Xeon currently has that label. The existing required anti-affinity keeps
|
||||
the two replicas on separate nodes. If the Xeon is unavailable, Kubernetes
|
||||
places its replacement on another worker without changing this configuration.
|
||||
@@ -14,6 +14,15 @@ resources:
|
||||
cpu: 700m
|
||||
memory: 3072Mi
|
||||
affinity:
|
||||
nodeAffinity:
|
||||
preferredDuringSchedulingIgnoredDuringExecution:
|
||||
- weight: 100
|
||||
preference:
|
||||
matchExpressions:
|
||||
- key: capacity.brunner.ninja/cpu-class
|
||||
operator: In
|
||||
values:
|
||||
- large
|
||||
podAntiAffinity:
|
||||
requiredDuringSchedulingIgnoredDuringExecution:
|
||||
- labelSelector:
|
||||
|
||||
Reference in new issue
Block a user