[helm] don't use vllm service name

[poussa]

Description

If vllm k8s service name is vllm bad things happen and the server fails to comeup with error message:

ERROR 11-26 09:29:11 engine.py:366]     lambda: int(os.getenv('VLLM_PORT', '0'))
ERROR 11-26 09:29:11 engine.py:366] ValueError: invalid literal for int() with base 10: 'tcp://10.100.120.175:80'

See vllm note about the service naming

Issues

na see above.

Type of change

List the type of change like below. Please delete options that are not relevant.

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds new functionality)
• Breaking change (fix or feature that would break existing design and interface)

Dependencies

na

Tests

Manually tested.

[eero-t]

Service fullname is always prefixed with the Helm parent chart name. Only if you helm install vLLM component directly, it's just vllm.

Agent component already relies on that service name being $release-vllm, and there are couple of not-yet-merged PRs that also rely on it.

=> You need to add check that updates fullname only when it actually equals vllm. And a comment with link to the doc (https://github.com/vllm-project/vllm/blob/main/docs/source/serving/env_vars.rst) would also be nice.

[yongfengdu]

I see 2 places in current code are using vllm.fullname

1. testpod https://github.com/opea-project/GenAIInfra/blob/main/helm-charts/common/vllm/templates/tests/test-pod.yaml
2. llm-uservice https://github.com/opea-project/GenAIInfra/blob/main/helm-charts/common/llm-uservice/templates/configmap.yaml#L19
   Please also delete the comments for not using vllm as the release name in vllm README https://github.com/opea-project/GenAIInfra/blob/main/helm-charts/common/vllm/README.md#installing-the-chart

[yongfengdu]

If so, the testpod will have to do this check too.
Do you know a better way to reference the service name in test pod? currently we use {{ include "vllm.fullname" . }}, the same as used in service name.

=> You need to add check that updates fullname only when it actually equals vllm. And a comment with link to the doc (https://github.com/vllm-project/vllm/blob/main/docs/source/serving/env_vars.rst) would also be nice.

[poussa]

Not needed.