Ask Julie Robertson a questionLost your password? Please enter your email address. You will receive a link and will create a new password via email.
Ask Julie Robertson a questionPlease briefly explain why you feel this question should be reported.
Please briefly explain why you feel this answer should be reported.
Please briefly explain why you feel this user should be reported.
Why does Kubernetes Horizontal Pod Autoscaler not scale even when CPU usage is high?
Autoscaling relies on metrics and resource requests, not just raw CPU usage. If the metrics server isn’t working or your pods don’t define CPU requests, Kubernetes has nothing to scale against. CPU limits alone are not enough, which surprises many people the first time they configure autoscaling. WhRead more
Autoscaling relies on metrics and resource requests, not just raw CPU usage.
If the metrics server isn’t working or your pods don’t define CPU requests, Kubernetes has nothing to scale against. CPU limits alone are not enough, which surprises many people the first time they configure autoscaling.
When autoscaling doesn’t react, the issue is usually missing data rather than incorrect thresholds.
Takeaway: Autoscaling only works when metrics and requests are both present.
See lessWhy does Terraform fail with “provider configuration not present” during destroy?
Terraform still needs the provider configuration that was used to create the resource, even during destruction. If you removed or renamed a provider after resources were created, Terraform can no longer manage them. This often happens after refactoring modules or cleaning up unused providers too earRead more
Terraform still needs the provider configuration that was used to create the resource, even during destruction.
If you removed or renamed a provider after resources were created, Terraform can no longer manage them. This often happens after refactoring modules or cleaning up unused providers too early.
Reintroducing the provider temporarily allows Terraform to finish cleanup safely.
Takeaway: Never remove a provider until all resources using it are gone.
See lessWhy does my Kubernetes service work internally but not from outside the cluster?
Internal access proves the service works, but external access depends on how it’s exposed. If the service type or networking setup isn’t correct, traffic never reaches the cluster from outside. Security rules and load balancer provisioning are frequent blockers here. Takeaway: External access probleRead more
Internal access proves the service works, but external access depends on how it’s exposed.
If the service type or networking setup isn’t correct, traffic never reaches the cluster from outside. Security rules and load balancer provisioning are frequent blockers here.
Takeaway: External access problems are almost always networking issues.
See lessWhy does my Terraform backend initialization fail with a state lock error?
Terraform is being cautious here. The state lock error means Terraform believes another process is using the state file, even if that process no longer exists. This usually happens after an interrupted run—someone closes their laptop, a CI job gets canceled, or a network connection drops during applRead more
Terraform is being cautious here. The state lock error means Terraform believes another process is using the state file, even if that process no longer exists.
This usually happens after an interrupted run—someone closes their laptop, a CI job gets canceled, or a network connection drops during
apply. Terraform leaves the lock behind to protect the state, but it has no way to know the process died.If you’re sure no one else is running Terraform, manually unlocking the state is safe. The key thing is to avoid force-unlocking while another deployment is genuinely in progress, because that’s when state corruption happens.
Takeaway: State locks are normal, and stale locks are a routine operational issue, not a Terraform bug.
See lessWhy does my Kubernetes node show NotReady after scaling up?
A new node reports NotReady until networking and system components are fully initialized. If it stays that way, the issue is almost always related to networking or permissions. Common causes include CNI plugins failing to start, blocked outbound access, or missing permissions required for node bootsRead more
A new node reports
NotReadyuntil networking and system components are fully initialized. If it stays that way, the issue is almost always related to networking or permissions.Common causes include CNI plugins failing to start, blocked outbound access, or missing permissions required for node bootstrapping. Looking at node events usually reveals whether kubelet, networking, or system pods are failing.
This is rarely a compute issue and almost never fixed by simply waiting longer.
Takeaway: Persistent
See lessNotReadynodes usually point to networking or bootstrap failures.