I have an GKE autopilot cluster running, and every now and then it randomly starts preempting pods, which doesn’t make sense in these cases. The last few days I have issues with pods being terminated early by GKE (Autopilot) the pods have the annotation cluster-autoscaler.kubernetes.io/safe-to-ev…

Hi, The main issue is that Autopilot spins up a machine especially for that specific pod, then kills the pod pretty soon after it started, just because it didn’t spun up the correct machine, or something else is going wrong. Request and Resource Requests and Limits are both fine for these pods. Thi…

Hi @paul-aldora , we are experiencing pretty much the same problem: We have a GKE Autopilot cluster that starts our problem scenario with a few nodes (for example 3) We schedule a few GPU workload pods (let’s say 2) which causes 2 new GPU nodes to be provisioned. The konnectivity-agent-autoscaler …

Hi @paul-aldora . Im looking into this Quick question is the pod deployed as a standalone pod? Or is it managed by a controller like Deployment, Statefulset, Daemonset?

Hi, We’re experiencing the same issue @paul-aldora is describing. We’re running a CronJob in a GKE Autopilot cluster and (almost) every week pods randomly get preempted, which is unexpected given that these pods (part of a cronjob) constitute the only workload in the cluster. Context: The cronjob…

@paul-aldora can you please upgrade to a newer version of GKE ? I think this is a known issue https://cloud.google.com/kubernetes-engine/docs/release-notes#June_16_2025

To which version do you recommend upgrading? I don’t see how this feature is related to the issues. Feature: For clusters running GKE version 1.32.4-gke.1236000 or later, the cluster autoscaler can scale down nodes by evicting Pods in the kube-system namespace that have no Pod Disruption Budget (P…

I’ve tried scheduling a sacrifice pod with lower priority. That saves our critical pods most of the time but not always. I’ve first updated our Autopilot cluster from 1.32.6-gke.1096000 to 1.33.2-gke.1240000 which is the newest version in the Regular release channel. No help. Then I tried switchin…

@Stefan_van_Kessel sounds like a support question. Go ahead and ask support. i cannot help you debug this way :slight_smile:

We have the same problem. Our pods are rather standard, using the default general-purpose compute class, requesting 0.5-1 vCPUs and 1-2 GB of RAM per pod. Until July, everything was fine, pod eviction was very rare - happening 1-2 times a month, and only with a few services, which is fine. However,…

GKE Autopilot and preemted pods

Google Cloud Serverless Applications

partomatl October 2, 2025, 9:42pm 17

We’ve been facing the same problem since our GKE Autopilot cluster was upgraded from 1.32 to 1.33.

Are you in contact with Google support? Have they acknowledged the issue?

Topic		Replies	Views
Large node resource reservations by gke-system-balloon-pods in GKE Autopilot Serverless Applications gke	1	338	March 3, 2026
Autopilot cluster keep restarting nodes Serverless Applications gke	7	491	March 26, 2026
GKE autopilot - fixing misbehaving autoscaler Serverless Applications gke	15	1255	June 5, 2025

GKE Autopilot and preemted pods

AI Suggested topics