[Question] Karpenter controller memory usage increases with Pod AGE and stabilizes around 85%

[EigoOda]

Environment

• Karpenter version: v1.2.1
• Kubernetes version: v1.32
• Cloud provider: AWS
• Installation method: Helm chart

---

Observation

We’ve observed that the Karpenter controller Pod’s memory usage gradually increases over time (correlating with Pod AGE) and then stabilizes around 85% of the memory limit.
This behavior is consistent across restarts and environments.

---

Questions

1. Is this expected behavior related to Go’s GC soft limit (GOMEMLIMIT = 90% of MEMORY_LIMIT)?
2. Is there any known issue or potential leak in Karpenter v1.2.x related to controller cache or informer memory growth?
3. Would upgrading to v1.3.x or later improve memory usage stability?

---

Additional Context

We confirmed that:

• The controller performs normally and does not get OOMKilled.
• Restarting the controller resets memory usage back to baseline.
• No abnormal increase in CPU usage or logs.

It seems like Karpenter’s memory usage just reaches a steady state near its GC threshold, but I’d like to confirm whether this is by design or if there’s ongoing optimization in newer versions.

Thanks in advance for clarifying!