[Azure] [OpenAI] Enhance Logs Dashboard

[lucian-ioan]

Proposed commit message

WIP, enhancing Azure OpenAI integration dashboard in preparation for GA.

Improvements:

1. Controls: Model Deployment Name, Model Name, Model Version
2. Lens (by title): Number of Requests, Total Average Latency, Response Codes, Response Codes over time, Average Latency, Request and Response Length.

Unfortunately Azure masks the last octet of the IP so we can't get exact data about specific users.

Checklist

• I have reviewed tips for building integrations and this pull request is aligned with them.
• I have verified that all data streams collect metrics or logs.
• I have added an entry to my package's changelog.yml file.
• I have verified that Kibana version constraints are current according to guidelines.

Author's Checklist

• [ ]

How to test this PR locally

Related issues

Screenshots

Before:

After:

[elasticmachine]

🚀 Benchmarks report

To see the full report comment with /test benchmark fullreport

[muthu-mps]

Adding more control filter is more useful but we need to also verify how the model details are captured for different logs here.

Controls: Model Deployment Name, Model Name, Model Version

Audit

• The audit log will not have the model specific information. When we apply any of the control above this data will be blank in the dashboard. This may not impact since user will be aware that the audit events may not relate to the model.
• Lets check what happens with the Request Response logs.

Request Response

• This logs comes in two different format.
• The request response logs add the model fields only for the successful requests. Not all the error requests logs the model fields.
• When filtering based on the model fields this will show only the successful request and you won't be able to see the failed requests. Requests with 400 or 404 errors will not be listed.

@lucian-ioan - What do you think, If you consider this filters adds more value then lets move them to the second level after the resource controls.

[lucian-ioan]

@muthu I believe the control groups for model information make sense even if in some cases the lenses will be empty. Users might want to benchmark different types of models against each other easily.

I agree that the model controls should come after the subscription ones, fixed that.

Removing old dashboard and adding new screenshot shortly.

[muthu-mps]

Please add the dashboard image to the /img directory. Otherwise the change looks good!

[elasticmachine]

💚 Build Succeeded

• Buildkite Build
• Commit: fb32a1b

History

• 💔 Build #11881 failed a79a488
• 💚 Build #11867 succeeded 3d70611
• 💚 Build #11846 succeeded 80e931d

cc @lucian-ioan

[elastic-sonarqube]

Quality Gate passed

Issues
0 New issues
0 Fixed issues
0 Accepted issues

Measures
0 Security Hotspots
No data about Coverage
No data about Duplication

See analysis details on SonarQube

[elasticmachine]

Package azure_openai - 0.2.0 containing this change is available at https://epr.elastic.co/search?package=azure_openai