An online endpoint returns 500 errors after a new deployment. What do you check first?
Why C is the answer
The deployment logs from the scoring container
Container logs show scoring script and dependency errors.
In detail
- A 500 usually comes from the scoring code or its environment.
- The container logs show the reason.
- Test the scoring script locally first.
Use case
An illustrative scenario, not a real incident.
An endpoint returns 500 after a release. The scoring container's logs show a missing package.
See it in different ways
stateDiagram-v2 [*] --> WhenNeeded WhenNeeded --> Online: during a request WhenNeeded --> Batch: large volume, can wait Online --> Deployments: stable URL, several deployments Batch --> [*] Deployments --> [*]
flowchart LR S0["500 errors"] --> S1["Deployment logs"] --> S2["Find the exception"] --> S3["Fix and redeploy"]
Key point: Read the scoring container logs.
Why the other options are wrong
- A. The notebook used to explore the data
- That notebook is not running in production.
- B. The name of the MLflow experiment
- An experiment name has no effect.
- D. The tags on the storage account
- Storage tags have no effect.