#kubernetes (2022-09)

https://github.com/aws/containers-roadmap/issues/185#issuecomment-1234850299

does anybody know where i can find details around how memory is throttled when k8s is running under cgroupsv2 with the feature enabled? I’m trying to wrap my head around what it means for memory to be treated as a “compressible” resource.

Hey guys,

Facing an Issue running jenkins-operator on kubernetes cluster v1.22.11-eks with Datadog monitoring agent. The DD agent is injecting some ENV vars in the jenkins-instance when created ex: DD_AGENT_HOST & DD_ENTITY_ID. The above ENV vars are causing the operator to restart the jenkins-instance pod in a loop.

Did anyone of you have used jenkins-operator next to a monitoring agent like Datadog, Newrelic, etc?

Can someone help me on understanding an anti-affinity policy I am creating for an application? I’m trying to guarantee each new replica will be deployed in a different AZ, so I came up with the following block:

affinity:
  podAntiAffinity:
    preferredDuringSchedulingIgnoredDuringExecution:
      podAffinityTerm:
        labelSelector:
          matchExpressions:
          - key: app.kubernetes.io/name
            operator: In
            values:
            - ${app-name}
       topologyKey: topology.kubernetes.io/zone

So, my thinking is:

1. I am creating an Anti-Affinity policy, meaning I want to avoid, if possible (thus the preferred element), assigning the pod to a node according to certain rules
2. I am using podAffinityTerm and labelSelector, meaning I am looking for labels in pods that are already assigned to my cluster
3. I am looking for pods with a label common to all my application pods (the scheduler will use the information in matchExpressions, key, operator, and values)
4. I am telling the scheduler to look for a different zone (as defined in topologyKey) if the criteria is met
5. I know that if all the zones are already filled with pods of my application, then the scheduler will be free to assign to any node (because no other rules are defined, and I defined a soft rule for the anti-affinity) So is my understanding correct?

Hi folks, I’m Anirudh. I used to work on K8s core controllers in the past, and for the past 2y I’ve been working on a platform called Signadot to test microservices in K8s at scale. Rather than stamping out new copies of infrastructure, the approach we took is to make use of request-level tenancy and dynamic request routing to isolate environments. This makes it possible to get lightweight environments which can share resources with each other while isolating at the request level, similar to how the “copy-on-write” model works for memory. Just launched on Product Hunt and would love to get feedback if you have a few minutes to spare. TIA!

Does anyone know how to troubleshoot a pod stuck on ContainerCreating? logs and events does not show anything that might be telling.. really weird

I am trying to get the aws-ebs-csi-driver helm chart working on a EKS 1.23 cluster.

The message I am getting from PVC events

failed to provision volume with StorageClass "gp2": error generating accessibility requirements: no topology key found on CSINode

The CSI topology feature docs say that:

• The PluginCapability must support VOLUME_ACCESSIBILITY_CONSTRAINTS. • The plugin must fill in accessible_topology in NodeGetInfoResponse. This information will be used to populate the Kubernetes CSINode object and add the topology labels to the Node object. • During CreateVolume, the topology information will get passed in through CreateVolumeRequest.accessibility_requirements. I am not sure how to configure these points.

if anyone here has kicked the tires on https://acorn.io/ – I’d be interested to hear your thoughts. edit: “Acorn is a containerized application packaging framework that simplifies deployment on Kubernetes”

EKS question. How does one use pod security groups to connect the traffic between the ALB SG and the pod SG? Using the ALB Ingress

I’ve got the traffic between the pod and RDS SGs working fine, but the traffic between the ALB and the Pod is only permitted when I do the following:

• open TCP 4200 on the pod security group from the VPC CIDR

• open TCP 30141 on the pod security group from the VPC CIDR

Any combination of allowing those ports from the ALB SG doesn’t work. That also includes the ALB ingress shared SG Where 4200 is the container port, and 30141 is the service NodePort

edit—- Got this working. Needed to open the following

• open TCP 4200 on the pod security group from the node security group ID

• open TCP 30141 on the pod security group from the node security group ID

Anyone have experience with different service meshes? I have gotten Istio, Linkerd, and NGINX Service Mesh working, but when I tried Consul, cannot get off the ground. Their community is not too responsive sadly.