SecOps Logging - Databricks
Note
Databricks takes up to two hours to populate inference tables from serving AI models: https://docs.databricks.com/aws/en/generative-ai/agent-evaluation/monitoring-agent-framework#limitations
Admin-view user logs: https://learn.microsoft.com/en-us/azure/databricks/admin/account-settings/audit-logs
Admin-view system tables: https://learn.microsoft.com/en-us/azure/databricks/admin/system-tables/
All log tables are restricted to administrators only unless indicated otherwise in the Notes column:
| Location | Source | Log | Type | Description | Method | Notes | Integration with Splunk |
|---|---|---|---|---|---|---|---|
| Databricks | REST API | Model Build | Conda package downloads | Logs from building the conda environment in which the model runs, including any installed packages. | Build Logs | ||
| Databricks | REST API | Model Service | Container serving information | Logs from the container hosting the model. | Serving Logs | ||
| Databricks | ??? | Model Events | Deployment status | The current availability of a given model endpoint to receive queries | \<Unknown> | ||
| Databricks | AI Gateway | Model Monitoring | User Request | The text of the query submitted by an individual user | Inference Tables | >2 hour delay (see above), available to permissioned users | |
| Databricks | AI Gateway | Model Monitoring | User Request Time | The time when a user submitted a request to a model | Inference Tables | >2 hour delay (see above), available to permissioned users | |
| Databricks | AI Gateway | Model Monitoring | Response Latency | The speed at which the model was able to respond to a request measured in milliseconds | Inference Tables | >2 hour delay (see above), available to permissioned users | |
| Databricks | AI Gateway | Model Monitoring | Error Logging | Any issues encountered with performing inference on a given request are logged | Inference Tables | >2 hour delay (see above), available to permissioned users | |
| Databricks | AI Gateway | Model Monitoring | Model Traces | The MLflow model trace represented in JSON (includes request, documents, response, and metadata) for a given model interaction | MLflow Trace Search |
>2 hour delay (see above), available to permissioned users | |
| Databricks | Audit | Network Access | Access Time | When the outbound network access queries occurred | System Tables - Outbound Network | ||
| Databricks | Audit | Network Access | Destination Name | Where outbound network access queries were destined | System Tables - Outbound Network | ||
| Databricks | Audit | Network Access | DNS (if destination was DNS) | Which DNS destination was used to perform an outbound network access query | System Tables - Outbound Network | ||
| Databricks | Audit | Network Access | Storage Location (if destination was Storage) | Which Storage destination was desired by an outbound network access query | System Tables - Outbound Network | ||
| Databricks | Audit | Job Events | Creation Time | The time at which a job was created | Audit Logs - Jobs | ||
| Databricks | Audit | Job Events | Deletion Time | The time at which a job was deleted | Audit Logs - Jobs | ||
| Databricks | Audit | Job Events | Created By | The user that created a given job | Audit Logs - Jobs | ||
| Databricks | Audit | Job Events | Deleted By | The user that deleted a given job | Audit Logs - Jobs | ||
| Databricks | Audit | Job Events | Start Time | The time at which a job was run | Audit Logs - Jobs | ||
| Databricks | Audit | Job Events | Job Status | The success or failure status of a job | Audit Logs - Jobs | ||
| Databricks | Audit | Job Events | Job Permissions | Triggered when a user attempts to change a job's permissions | Audit Logs - Jobs | ||
| Databricks | Audit | Groups Events | Add to group | Records when admins add a user to a group | Audit Logs - Groups | ||
| Databricks | Audit | Groups Events | Create group | Records when admins create a group | Audit Logs - Groups | ||
| Databricks | Audit | Groups Events | Remove group | Records when admins remove a group | Audit Logs - Groups | ||
| Databricks | Audit | Secrets Events | Create secret scope | Event where user creates a secret scope | Audit Logs - Secrets | ||
| Databricks | Audit | Secrets Events | Delete secret scope | Event where user deletes a secret scope | Audit Logs - Secrets | ||
| Databricks | Audit | Secrets Events | Get secret | Event when user requests value of a secret | Audit Logs - Secrets | ||
| Databricks | Audit | Secrets Events | Put secret | Records when user writes the value of a secret to a scope | Audit Logs - Secrets | ||
| Databricks | Audit | Model Serving Events | Create model endpoint | Logs creation of a model serving endpoint | Audit Logs - Real Time Serving | ||
| Databricks | Audit | Model Serving Events | Update model endpoint access | Shows the result of permissions changes to a given model serving endpoint | Audit Logs - Real Time Serving | ||
| Databricks | Audit | Model Serving Events | Delete model endpoint | Logs delete of a model serving endpoint | Audit Logs - Real Time Serving | ||
| Databricks | Audit | Model Serving Events | Enable model endpoint | Shows when a user enables a model serving endpoint for requests | Audit Logs - Real Time Serving | ||
| Databricks | Audit | Model Serving Events | Disable model endpoint | Shows when a user disables a model serving endpoint from receiving requests | Audit Logs - Real Time Serving | ||
| Databricks | Audit | Model Serving Events | Update model endpoint | Shows when a user updates a model serving endpoint to use a different model | Audit Logs - Real Time Serving |