Hi, I am using MetalLB under Talos (0.11) with K8s 1.21. The cluster has been running on MetalLB 0.9.6 for months w/o issues. I have just attempted to update to 0.10 but found that it misbehaves as explained in this ticket.
The problem:
In 0.9.6 I would see a single "nodeAssigned" event per LB; in 0.10.2 I see a flood of them, roughly one per LB every 2 seconds, which generates hundreds of events every minute.
LBs themselves continue to work, but clearly something is not right.
My configuration/environment:
- MetalLB 0.9.6 (working) and 0.10.2 (nodeAssigned spam)
- Installation method: by manifest
- L2 mode
- kube-proxy in iptables mode (Talos default)
- Talos 0.11.5 with K8s 1.21.3 under Linux 5.10.52
- Calico CNI 3.19.0
My config:
apiVersion: v1
kind: ConfigMap
metadata:
namespace: metallb-system
name: config
data:
config: |
address-pools:
- name: ingress-public
protocol: layer2
addresses:
- 10.99.99.122/32
auto-assign: false
- name: ingress-foo
protocol: layer2
addresses:
- 10.99.99.123/32
auto-assign: false
- name: ingress-internal
protocol: layer2
addresses:
- 10.4.99.0/32
auto-assign: false
- name: default
protocol: layer2
addresses:
- 10.4.99.1-10.4.99.31
Special cluster settings: The MixedProtocolLBService=true api-server feature gate is set, and one LB on the cluster serves both TCP and UDP. The issue appears to be unrelated to this setting however.
Instead of updating I have also tried to delete the metallb-system namespace and recreate it from scratch with no change.
If you believe that this issue is not due to a configuration/cluster at my end I would appreciate your input. I can dedicate some time to diagnosing and fixing this issue w/ code contributions if needed.
Hi, I am using MetalLB under Talos (0.11) with K8s 1.21. The cluster has been running on MetalLB 0.9.6 for months w/o issues. I have just attempted to update to 0.10 but found that it misbehaves as explained in this ticket.
The problem:
In 0.9.6 I would see a single "nodeAssigned" event per LB; in 0.10.2 I see a flood of them, roughly one per LB every 2 seconds, which generates hundreds of events every minute.
LBs themselves continue to work, but clearly something is not right.
My configuration/environment:
My config:
Special cluster settings: The
MixedProtocolLBService=trueapi-server feature gate is set, and one LB on the cluster serves both TCP and UDP. The issue appears to be unrelated to this setting however.Instead of updating I have also tried to delete the metallb-system namespace and recreate it from scratch with no change.
If you believe that this issue is not due to a configuration/cluster at my end I would appreciate your input. I can dedicate some time to diagnosing and fixing this issue w/ code contributions if needed.