Split per-run activity + early biosim_run_id DB write (convergence PR2.5)

[jcschaff]

PR2.5 of docs/simulation-runs-convergence-plan.md — the follow-up the original PR2 plan called for, finally landed. See docs/workflows-architecture.md Shape C for the diagram.

Summary

The monolithic submit_biosim_simulation_run_activity bundled submit + poll + HDF5 + save in one activity, so OmexSimWorkflow could not see biosimulations_run_id until the whole thing returned. PR2.5 splits it into two activities and uses the gap between them to write the run id to the database early, via the already-canonical Temporal pattern: an activity writing to MongoDB.

Activity split (in biosim_runs/activities.py)

• submit_biosim_run_activity — cache-check + submit; returns biosimulations_run_id immediately, plus the full cached_workflow_run on cache hit (so the workflow can skip poll).
• poll_biosim_run_activity — poll until terminal + HDF5 (with 404-lag retries) + save BiosimulatorWorkflowRun; returns the saved record.

Workflow wiring

• OmexSimWorkflowInput gains optional submission_run_id: str | None = None. When set (only SimulationRunWorkflow sets it; the verify path leaves it None), the child fires update_run_status_activity right after submit, writing biosimulations_run_id onto the matching SimulationRunRecord.
• OmexSimWorkflowOutput gains biosimulations_run_id (set after submit; exposed via the workflow's existing @workflow.query). Set even when the run later ends FAILED, so callers see the id either way.
• UpdateRunStatusInput.status is now Optional — the early write carries only the run id; the terminal write carries status.
• Worker + test fixtures: register the split pair in place of the retired monolithic.

Hybrid status read

• GET /simulations/{processing_id} now reads SimulationRunRecords for the processing_id and fills biosimulations_run_id per job when the workflow query lacks it. Workflow query owns live status; DB owns the early run id metadata. Non-fatal on DB failure.

Verify path: untouched semantics

The verify path (OmexVerifyWorkflow → OmexSimWorkflow → activities) passes submission_run_id=None, so the early DB write is skipped — verification behavior is byte-equivalent. The submit/poll split is functionally equivalent to the old monolithic end-to-end.

Testing

• ruff + mypy --strict clean.
• All simulations unit tests pass (router, runs-query, parsing).
• integration_local/test_simulations_run.py (the mock-backed e2e) passes — exercises the new submit/poll split + early write end-to-end.
• Full pytest -m "not integration": 76 passed, 3 failed. The 3 failures are all pre-existing/environmental, not PR2.5 regressions:
  • test_slurm_service::test_slurm_job_submit — UnboundLocalError, needs SSH creds.
  • test_sim_workflow — real-network simdata 404 (passes on retry, also failed on main earlier).
  • test_omex_verify_workflow_GCS — real-network biosimulations.org currently returning FAILED for fresh submissions (verified the integration-marked sibling test_simulation_run_workflow fails on main today with the same pattern).
• PR2.5's verify-path code is structurally equivalent to main's monolithic (no submission_run_id → no early write; submit/poll have the same semantics).

🤖 Generated with Claude Code