Why are the memory requirements in my Kubernetes cluster more than expected?

General Discussions
1

Hi,

I’m running a cluster in AWS (EKS). I’m experiencing an issue with a pod hosting an import service (Python FastAPI endpoint). The pod restarts upon file import.

Reason : OOMKilled - exit code: 137

The initial helm pod setup assigns:

yamlCopyresources:
  limits:
    memory: "512Mi"
  requests:
    memory: "128Mi"

The file weighs 3.75 MB. Memory monitoring does not show any peak (here is a screen shot of some metrics). Locally, the container uses 137 Mo RAM.

image
image1261×727 182 KB

I eventually decided to increase memory to:

yamlCopyresources:
  limits:
    memory: "1024Mi"
  requests:
    memory: "256Mi"

And then it works.

But, I find it kind of hard/weird to accept that amount of memory to handle fairly light files.

Additional info:

  1. Cluster nodes are not under memory pressure (kubetctl describe node <node_name>):

  2. No error on the pod logs. But the restart occurs more or less when the code is loading the file:

    response = s3_client.get_object(Bucket=settings.AWS_BUCKET,Key=import_create.s3_key)
excel_content = response['Body'].read()
df = pd.read_excel(io.BytesIO(excel_content), header=1)

  3. It works on my machine (typical computer guy statement :wink: ), and the container uses around 130 Mb of RAM.

  4. The issue is not systematic. I have been able to import another file slightly bigger.

My feeling is that it might not actually be a memory issue. What do you think? How would you troubleshoot that issue? How/where would you look for more detailed information?

Cluster information:

Kubernetes version:1.33
Cloud being used: aws eks
Installation method: terraform helm