Problem
The current Helm chart version vllm-stack (e.g. 0.1.9) sets the default router memory limit to:
routerSpec:
resources:
limits:
memory: 500Mi
In real cluster scenarios, this leads to:
OOMKilled- Exit code 137
- Router entering
CrashLoopBackOff
500Mi is insufficient for typical production-stack usage.
Observation
The GitHub repository appears to already contain an increased memory default for the router, but the published Helm chart (0.1.9) still renders 500Mi.
Running:
helm show values vllm/vllm-stack --version 0.1.9
still shows 500Mi.
So the Helm release does not yet include the updated default.
Suggested Fix
Increase the default router memory limit to at least:
requests:
memory: 1Gi
limits:
memory: 2Gi
or remove the hard memory limit and leave it configurable.
Workaround
Users can override the value manually in their values.yaml, but the default leads to immediate instability in many setups.
Thanks for the great project!
Problem
The current Helm chart version
vllm-stack(e.g. 0.1.9) sets the default router memory limit to:routerSpec:
resources:
limits:
memory: 500Mi
In real cluster scenarios, this leads to:
OOMKilledCrashLoopBackOff500Mi is insufficient for typical production-stack usage.
Observation
The GitHub repository appears to already contain an increased memory default for the router, but the published Helm chart (0.1.9) still renders
500Mi.Running:
helm show values vllm/vllm-stack --version 0.1.9
still shows
500Mi.So the Helm release does not yet include the updated default.
Suggested Fix
Increase the default router memory limit to at least:
requests:
memory: 1Gi
limits:
memory: 2Gi
or remove the hard memory limit and leave it configurable.
Workaround
Users can override the value manually in their
values.yaml, but the default leads to immediate instability in many setups.Thanks for the great project!