[ML][AI Connector] Unable to create Google Gemini connector using Google vertex AI

[saarikabhasi]

Describe the bug:

Unable to create google gemini connector in AI connector flyout, Google vertex AI option.

Steps to reproduce:

1. Select AI connector
2. Choose Google vertex AI as service provider
3. Fill in all required details for create new google gemini connector
4. Click Save
5. View An internal server error occurred. Check Kibana server logs for details
6. Change to rerank task type and notice same error

Expected behavior:
Should be able to create google gemini connector with AI connector

Screenshots (if relevant):

google.gemini.connector.error.mov

Errors in browser console (if relevant):

{
    "statusCode": 500,
    "error": "Internal Server Error",
    "message": "An internal server error occurred. Check Kibana server logs for details."
}

Provide logs and/or server output (if relevant):

Logs from kibana server

Rerank task type

[2025-05-05T15:25:37.947-04:00][WARN ][plugins.actions] Failed to create inference endpoint for task type "rerank" and inference id googlevertexai-rerank-hmbrxgc1r4c. Error: status_exception
        Root causes:
                status_exception: Model configuration contains settings [{location=us-central1}] unknown to the [googlevertexai] service
[2025-05-05T15:25:37.947-04:00][ERROR][http] 500 Server Error

Text embedding

[2025-05-05T15:28:07.327-04:00][WARN ][plugins.actions] Failed to create inference endpoint for task type "text_embedding" and inference id googlevertexai-text_embedding-c8frxwa1wir. Error: status_exception
        Caused by:
                status_exception: Received an unsuccessful status code for request from inference entity id [googlevertexai-text_embedding-c8frxwa1wir] status [400]. Error message: [Gemini cannot be accessed through Vertex Predict/RawPredict API. Please follow https://cloud.google.com/vertex-ai/docs/generative-ai/start/quickstarts/quickstart-multimodal for Gemini usage.]
        Root causes:
                status_exception: Received an unsuccessful status code for request from inference entity id [googlevertexai-text_embedding-c8frxwa1wir] status [400]. Error message: [Gemini cannot be accessed through Vertex Predict/RawPredict API. Please follow https://cloud.google.com/vertex-ai/docs/generative-ai/start/quickstarts/quickstart-multimodal for Gemini usage.]

Environment

Kibana 9.1.0
Cloud deployment 9.0.0, 8.18

[elasticmachine]

Pinging @elastic/ml-ui (:ml)

[alvarezmelissa87]

There are a couple of things going on here.

1. For Google Vertex AI in the configuration the _inference/_services endpoint is returning the location field as necessary for the rerank task type - which is not necessary. That should be just for text embedding. That needs to be fixed (@jonathan-buttner is already working on a fix).

2. The model id selected here - gemini-1.5-pro-002 - isn't one of the models that corresponds to the supported task types (rerank and text_embedding, in the case of this provider). I was able to confirm that it works when using the text-embedding-004 model id with the text embedding task type.

We need to improve error messaging in Kibana for this so there is an indication to the user as to what the underlying issue is.
Error handling improvement tracking issue here. As of now, it's just a generic internal error. I can work on improving the messaging here. We can show the actual error message - especially if it's a 400 - and then add a bit of helpful text to guide the user on what to check next. So show the error message in the toast and then also something like Please ensure the form configuration aligns with the selected task type.

[jonathan-buttner]

PR for the backend fix: elastic/elasticsearch#127856