update_manifests.sh generates wrong endpoint names

[eero-t]

This is for today's HEAD of GENAIInfra.

Use-case:

• Update manifests: cd helm-charts; ./update_manifests.sh
• Check generated service endpoint names: cd ../microservices-connector/config/manifests/; grep ENDPOINT *.yaml
• Deploy generated manifests to cluster and check resulting service names: kubectl get svc

Expected outcome:

• Generated ENDPOINT variables match service names for given endpoints, and services work

Actual outcome:

• Service logs are full of timeouts / queries failing
• Service names in ENDPOINT variables are prefixed with the calling service name, but actual service names are not
  • E.g. TGI_LLM_ENDPOINT variable value in llm-uservice.yaml manifest is http://llm-uservice-tgi instead of actual http://tgi (.default.svc.cluster.local)

[eero-t]

This seems to be a bug in the script, not in the Helm charts?

When I generate manifests directly with helm template chatqna chatqna --skip-tests ... things match, both ENDPOINT variables and actual service names are prefixed with chatqna-.

[lianhao]

The generated manifest files from ./update_manifests.sh are for GMC USE ONLY. I wouldn't suggest to use them directly. If you want to deploy each component microservice separately, please use helm charts under helm-charts/common directly.

@eero-t you're absolutely right. The devil is in the ./update_manifests.sh. We'll discuss with GMC folks in next cycle to see if we can make the generated manifest files usable for other users

[eero-t]

I see. "Generate manifests from Helm Charts" section in the directory README only talks about looking into update_manifests.sh, not about using it... :-)

[eero-t]

Btw. the table in that README has has wrong link name ("Code generation") for chatqna.

[yongfengdu]

Fixes in README to avoid confusion

343