Inference pool name in HTTPRoute should be truncated to 63 characters

[mnmehta]

Component

I don't know

Describe the bug

In templates/sample-application/httproutes.yaml we construct a reference to the inference pool name:

name: "{{ include "sampleApplication.sanitizedModelName" . }}-inference-pool"

However, the actual inference pool name has elsewhere been truncated to 63 characters because the name needs to be a valid DNS label. The net outcome is that the HTTPRoute object is unable to find the backend, so the the gateway returns cluster_not_found for any request.

Steps to reproduce

1. Deploy llm-d with a longish model name, e.g. using these overrides:

sampleApplication:
  baseConfigMapRefName: basic-gpu-preset
  model:
    modelArtifactURI: pvc://model-pvc/neuralmagic/DeepSeek-R1-Distill-Llama-70B-FP8-dynamic
    modelName: "neuralmagic/DeepSeek-R1-Distill-Llama-70B-FP8-dynamic"

2. Check HTTPRoute object, there will be this error:

Message:               InferencePool "neuralmagic-deepseek-r1-distill-llama-70b-fp8-dynamic-inference-pool" not found

3. Check the inferencepool, you will see the name has been truncated to 63 characters:

kubectl get inferencepool
NAME                                                              AGE
neuralmagic-deepseek-r1-distill-llama-70b-fp8-dynamic-inference   3h21m

Additional context or screenshots

No response

[nerdalert]

Hi @mnmehta Will poke around to see if there is a quick fix. We are probably too tightly coupled to the model name via the model service.

[nerdalert]

@mnmehta at your leisure would you PTAL at #314. I don't have the hardware atm to push the model you are trying. If you can't just let me know and I'll figure out a test. Ty!

[sumansuhag]

name: "{{ include "sampleApplication.sanitizedModelName" . }}-inference-pool"
But actual InferencePool name is truncated elsewhere to 63 characters.

Effect: HTTPRoute fails to bind to the correct backend, causing cluster_not_found errors at runtime.

Option 1
name: "{{ include "sampleApplication.sanitizedModelName" . | trunc 50 }}-inference-pool"

OR

{{- define "sampleApplication.inferencePoolName" -}}
{{ include "sampleApplication.sanitizedModelName" . | trunc 50 }}-inference-pool
{{- end }}
Bug confirmed — name in HTTPRoute exceeds DNS label limit.

🔄 Suggest using a helper like sampleApplication.inferencePoolName to centralize the name generation and apply consistent truncation (≤ 63 chars). That way, both inferencepool and HTTPRoute always reference the same valid name.

[nerdalert]

name: "{{ include "sampleApplication.sanitizedModelName" . }}-inference-pool" But actual InferencePool name is truncated elsewhere to 63 characters.

Effect: HTTPRoute fails to bind to the correct backend, causing cluster_not_found errors at runtime.

Option 1 name: "{{ include "sampleApplication.sanitizedModelName" . | trunc 50 }}-inference-pool"

OR

{{- define "sampleApplication.inferencePoolName" -}} {{ include "sampleApplication.sanitizedModelName" . | trunc 50 }}-inference-pool {{- end }} Bug confirmed — name in HTTPRoute exceeds DNS label limit.

🔄 Suggest using a helper like sampleApplication.inferencePoolName to centralize the name generation and apply consistent truncation (≤ 63 chars). That way, both inferencepool and HTTPRoute always reference the same valid name.

Hi @sumansuhag! I like your suggestion better than my PR. Would you be willing to open a PR? Ty.