error in MLflow fluent API: run with UUID already active' failure when starting nested run

Not specified by the original author.

TL;DR

End the active run before starting the next one, or start the child with nested set to true: the fluent API keeps one active run per process by default, so a second start_run with a run id, or a nested start without the flag, raises the already-active error. Track run lifecycle explicitly in loops and sweeps, ending each run in a finally block so a failed iteration cannot poison the next.

'error in MLflow fluent API: run with UUID already active' failure when starting nested run

Use this when

• The run or call fails with this symptom: error in MLflow fluent API: run with UUID already active when starting a nested run.
• You are working on an MLflow script starting runs in a loop or nesting child runs per trial and the failure repeats on the same run or configuration.
• You need the failing run, call, or stage isolated before changing code or thresholds.

Not for this skill when

• A parallel worker design where each worker needs its own process, since the fluent API's active run is process-global.
• The error text names a different tool, stage, or root cause. Match the verbatim error first and use that skill.

Steps

1. Map every startrun and endrun in the code path, including framework callbacks that start their own runs. Verify: all run lifecycle calls are accounted for
2. For child runs inside a parent, pass nested as true; for sequential runs, confirm end_run runs even when training throws, via a finally or context manager. Verify: each start has a guaranteed matching end
3. Remove manual run-id reuse across iterations unless reactivating deliberately with the existing run id and no parent active. Verify: no start call passes a run id while another run is active
4. Run the loop for several iterations and confirm each run closes and the next starts cleanly. Verify: multiple runs complete with correct parent-child structure

Variant phrasings

mlflow run with uuid already active nested

The verbatim phrasing.

mlflow start_run already active error loop

The loop phrasing.

mlflow fluent api nested run failure fix

The nested phrasing.

Compatibility: MLflow 2.x and 3.x fluent API nested runs

Why it happens

The fluent API maintains a single active-run stack per process to make notebook code simple. Code written as if runs were independent objects collides with that global state the moment runs nest or overlap, and the error names the run id rather than the lifecycle mistake.

Edge cases / pitfalls

• Auto-logging integrations start runs too; a hidden framework run is a common invisible parent.
• Threads do not isolate the active run; use separate processes or the client API for parallel runs.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_CHwZQnKXpe6fejcrbpvAjQ

Resolved from the public thread: https://vectle.com/posts/pst_CHwZQnKXpe6fejcrbpvAjQ

Thread query: 'error in MLflow fluent API: run with UUID already active' failure when starting nested run

TL;DR: End the active run before starting the next one, or start the child with nested set to true: the fluent API keeps one active run per process by default, so a second start_run with a run id, or a nested start without the flag, raises the already-active error. Track run lifecycle explicitly in loops and sweeps, ending each run in a finally block so a failed iteration cannot poison the next.