Describe here the purpose of your PR.
The CLI’s createCommandEvent call (the one that logs Uploading run metadata... from TrackableCommand) is the only step in the analytics upload flow without a retry wrapper. A single transient 5xx from the server fails the CI command and surfaces as Failed to upload run metadata: unknownError(502), forcing the user to re-run their CI job. Artifact uploads (MultipartUploadArtifactService) already wrap their work in RetryProvider.
This PR injects RetryProviding into CreateCommandEventService and runs the API call inside runWithRetries, so transient failures (502/503/504, network errors) are retried up to three times with backoff before bubbling up. Same pattern as the existing artifact-upload path.
How to test locally
Document how to test your changes locally
- Point the CLI at a server that returns a 502 once and then 200 (e.g. a stub or a temporarily-down endpoint). Run any tracked command and confirm
Uploading run metadata... succeeds without surfacing unknownError(502).
- Inject a mock
RetryProviding in a unit test to assert that createCommandEvent calls the underlying API up to 3 times on failure.
- Sanity check: a 4xx (e.g. unauthorized) still surfaces as before — the retry will exhaust quickly given the tiny backoff and throw the same typed error.
T