High memory usage leading to OOM killed on kubernetes

[farru1998]

Summary

Hey folks!

Dealing with an issue that the memory usage of NextJS app on kubernetes is increasing linearly over time.
As per our observation, it seems the garbage collector is not being triggered maybe. As a result, memory usage increases linearly, eventually reaching the pod’s memory limit and causing an OOM (Out of Memory) error.

Once this happens, the Kubernetes pod is restarted. Below is the attached memory trend before and after restart.

We are running npm start in our dockerfile as the CMD command which eventually runs:

start": "next start -p 5000"

Additional information

Next Version: 14.2.5
Node Version: 20

Example

No response

[icyJoseph]

IIRC, we have to specify max-old-space-size? https://stackoverflow.com/a/48392705

[icyJoseph]

Not this one exactly, but I know that 15.5.2 (even 15.5.1 IIRC) has a bunch of memory usage improvements (including fetch leakage)

[1jmj]

hmm, i'm on 15.5.2 and still having OOM problems. The CPU on "next-server" spikes even when no traffic and then locks the server until it runs out of memory and resets. Its in Docker.

[icyJoseph]

I think there was also some OOM issue in Kubernetes itself, I can't find it right now, but it was either an issue or discussion on this repository. Don't quote me on that.

IIRC there was a script to diagnose some of this stuff... What kind of image is it? and do you have any instrumentation/next.config.js side effects?

spikes even when no traffic

That's really not good 🤔 something else must be happening here

[1jmj]

I have a simple memory output in instrumentation. Its flat:

[instrumentation-mem] 2025-09-09T13:56:59.216Z {"rssMB":279,"heapUsedMB":107.3,"heapTotalMB":114.5,"externalMB":4.2,"uptimeS":521}
[instrumentation-mem] 2025-09-09T13:57:09.223Z {"rssMB":279,"heapUsedMB":107.3,"heapTotalMB":114.5,"externalMB":4.2,"uptimeS":531}
[instrumentation-mem] 2025-09-09T13:57:19.225Z {"rssMB":279,"heapUsedMB":107.3,"heapTotalMB":114.5,"externalMB":4.2,"uptimeS":541}
[instrumentation-mem] 2025-09-09T13:57:29.232Z {"rssMB":279,"heapUsedMB":107.3,"heapTotalMB":114.5,"externalMB":4.2,"uptimeS":551}
[instrumentation-mem] 2025-09-09T13:57:39.239Z {"rssMB":279,"heapUsedMB":107.3,"heapTotalMB":114.5,"externalMB":4.2,"uptimeS":561}
[instrumentation-mem] 2025-09-09T13:57:49.240Z {"rssMB":279,"heapUsedMB":107.3,"heapTotalMB":114.5,"externalMB":4.2,"uptimeS":571}
[instrumentation-mem] 2025-09-09T13:57:59.245Z {"rssMB":279,"heapUsedMB":107.3,"heapTotalMB":114.5,"externalMB":4.2,"uptimeS":581}
[instrumentation-mem] 2025-09-09T13:58:09.252Z {"rssMB":279,"heapUsedMB":107.4,"heapTotalMB":114.5,"externalMB":4.2,"uptimeS":591}
[instrumentation-mem] 2025-09-09T13:58:19.252Z {"rssMB":279,"heapUsedMB":107.4,"heapTotalMB":114.5,"externalMB":4.2,"uptimeS":601}
[instrumentation-mem] 2025-09-09T13:58:29.254Z {"rssMB":279,"heapUsedMB":107.9,"heapTotalMB":114.5,"externalMB":4.2,"uptimeS":611}

We are alpine and node 24

[1jmj]

Update, we think it's something in PayloadCMS so looking further into that.