[rjgoyln]

Summary

The polling loop only exits once the compute profile reaches the status the operation waits for. An operation type the trigger does not recognise never reaches one, so the trigger polls indefinitely instead of reporting the bad input. No shipped operator passes such a value, so this is hardening rather than a fault reachable from a Dag.

The Invalid operation event meant to cover the case sits after the loop, unreachable, as do the two timeout events beside it — the triggerer's deferral timeout has always superseded those.

• resolve the expected status before polling and emit Invalid operation there
• drop the unreachable post-loop branches
• narrow the missing-profile error to ValueError

The status each operation waits for is unchanged, and every event the trigger can still emit keeps its payload. ValueError never leaves the trigger: the same except Exception turns it into the same error event, and the operator raises on event["message"] without inspecting the type.

Touches the same method as #72695 but different lines; the two merge cleanly in either order.

Tests

test_run_invalid_operation fails against the pre-change trigger, which yields the missing-profile error instead. The two test_run_create_success cases pass either way — they cover create operations the existing tests left untested.

---

Was generative AI tooling used to co-author this PR?

• Yes — Claude Code (Opus 5)

Generated-by: Claude Code (Opus 5) following the guidelines

[rjgoyln]

Thanks for catching these! Both points are addressed — I updated the docstring to document all supported operations and fixed the invalid-operation message to include the rejected value. The test expectation has been updated accordingly.