Bad service name resolution importing model in k8s

[angelocorreia27]

Action: Importing modelfiles
Error:

File "/usr/local/lib/python3.11/site-packages/requests/adapters.py", line 519, in send
raise ConnectionError(e, request=request)
requests.exceptions.ConnectionError: HTTPConnectionPool(host='open-webui-ollama.xxxx.svc.cluster.local', port=11434): Max retries exceeded with url: /api/create (Caused by NameResolutionError("<urllib3.connection.HTTPConnection object at 0x7f5cef705e50>: Failed to resolve 'open-webui-ollama.xxxx.svc.cluster.local' ([Errno -2] Name or service not known)"))
INFO: 10.32.0.1:0 - "POST /ollama/api/create HTTP/1.1" 500 Internal Server Error

kubernetes version: 1.27.2

It seems like the charts are assuming 'open-webui-ollama' as the name of the Ollama service and do not recognize a different name.

[0xThresh]

Hi @angelocorreia27, am I correct in assuming that you used the built-in Ollama config with ollama.enabled = true? Could you drop me your values.yaml file so I can try to replicate the issue?

[angelocorreia27]

Thank you for taking the time to analyze the issue. I am indeed using the built-in Ollama config with ollama.enabled = true.

However, I encountered a problem when deploying via ArgoCD. ArgoCD uses the metadata.name field from the manifest to create the service name. Since your manifest expects the service name for Ollama to be open-webui-ollama, this naming convention causes a conflict in my ArgoCD deployment setup.

Here is my current values.yaml configuration:
apiVersion: argoproj.io/v1alpha1
kind: Application
metadata:
name: not-allowing-different-name-then-sample-ollama
namespace: argo
spec:
destination:
namespace: angelocorreia27gmailcom
server: "https:xxx.xx"
source:
path: ollama
repoURL: "https://xxx"
targetRevision: HEAD
helm:
parameters:
- name: global.storageClass
value: "xx"
- name: ingress.enabled
value: "true"
- name: ingress.className
value: "xx"
- name: ingress.hosts[0].host
value: "xx"
- name: ingress.hosts[0].paths[0].path
value: "/"
- name: ingress.hosts[0].paths[0].pathType
value: "Prefix"
- name: persistentVolume.enabled
value: "true"
- name: persistentVolume.size
value: "10Gi"
- name: persistentVolume.storageClass
value: "xxx"

To work around this issue, I deployed Ollama and the OpenWebUI separately. After deploying them independently, I defined the Ollama endpoint within the WebUI configuration. Thank you.

[0xThresh]

Thanks for the extra context. I think in the case of your ArgoCD deployment, you'd want to update your clusterDomain value in values.yaml to match the service name Argo would deploy.

In the case of using a separate Ollama backend (which is the setup I use as well), you can also opt to update the ollamaUrls value to your Ollama service(s) name(s) so that it's loaded into Open WebUI correctly on startup. I have an example of that on a public repo here: https://github.com/0xThresh/self-hosted-genai/blob/main/helm.tf#L114

[angelocorreia27]

Sure. Thank you.

[0xThresh]

I found that this was a legitimate issue that resulted from one of the helper values in our chart trying to define a service name in the Open WebUI deployment while the Ollama chart's ollama.name value was still in use to name the Ollama service, causing a mismatch when you used the integrated Ollama without manually setting ollamaBaseUrls. I've implemented a fix for this in a PR I'll be submitting soon.

Thanks again for reporting it, and sorry I didn't catch it before.