sqlmesh run keeps reporting "environment is being updated" after the updating plan has failed #6157

I found this issue when I accidentally did sqlmesh plan --auto-apply to update PROD environment while sqlmesh run is running, the run detects the environment change and waits for the plan to finish.

However, if the plan fails due to some reason before the environment finalization, the run continues firing warnings like below:

[WARNING] Environment 'prod' is being updated by plan '<plan_id>'. Retrying in 30 seconds...

The plan process has already exited with an error, but the run command warning message implies that plan process is still updating the environment. The run does not resume quickly or prompts that manual recovery is required.

I understand that running against a partially applied environment is unsafe. My concern is to properly handle the failure and help the developers debug the issue, not bypassing environment finalization.

Environment

• SQLMesh: 0.236.3
• Execution: CLI
• Engine: Trino
• State: PostgreSQL
• Target SQLMesh environment: prod

Observed sequence

1. Start sqlmesh run.
2. While it is executing, run sqlmesh plan --auto-apply against the same environment.
3. Fail or abort the plan.
4. The run continues emitting the warning above even though the plan process has exited.
5. Restarting sqlmesh run does not resolve this issue (or the underlying unfinalized environment state).

Expected behavior

A failed plan should leave the environment protected, but the waiting run should provide a clear actionable outcome.

For example:

• If plan failure is recorded, run should report that the updating plan failed and exit with a nonzero status if possible.
• Otherwise, make the warning clear that SQLMesh is waiting for environment finalization, not necessarily an active process.
• Include the timeout and recovery guidance such as inspecting the failed plan and successfully reapplying it.

Actual behavior

The run repeatedly reports that the environment "is being updated" without distinguishing an active plan from state left behind by a failed plan.

The default timeout is six hours, which makes this appear to be an indefinite hang (firing warnings for every 30s).

Initial investigation

In the installed SQLMesh 0.236.3 pkg:

• Context.run() waits for a true value of environment_state.finalized_ts.
• The warning is based on the stored environment record, not the current live plan-process.
• Plan application updates the environment record before virtual-layer promotion.
• Environment finalization happens afterward.
• A failure between those stages can therefore leave subsequent runs waiting (for the default 6 hours).

Is retaining this unfinalized state the intended recovery mechanism? If so, could the waiting message and failed-plan recovery guidance be improved?