SEPURI-SAI-KRISHNA commented on issue #72144:
URL: https://github.com/apache/airflow/issues/72144#issuecomment-5440131327

   Thanks, the EMR catch is real and I've fixed it.
   
   **`EmrContainerOperator` / `EmrContainerSensor`:** you're right, and the 
miss was worse than one site each. Both build the trigger inside a conditional 
expression, so there are *two* constructions per site, four in total, all 
passing none of the three. My sweep only looked at `trigger=<Call>` and skipped 
`ast.IfExp` entirely, which means my invariant test would have skipped them 
too. The test now resolves the `trigger=` expression to every construction it 
can evaluate to, and a defer site whose trigger is a bare reference is asserted 
against an explicit allowlist rather than silently ignored. That last part 
catches one more: `EksPodOperator` passes a pre-built `EksPodTrigger`, which is 
a `KubernetesPodTrigger` and out of scope.
   
   `EmrContainerTrigger` also needed its hook's `virtual_cluster_id`, so it 
overrides `_hook_parameters` rather than relying on the default, which is the 
escape hatch the base class keeps for exactly this.
   
   **`SageMakerNotebookOperator`:** agreed, excluded. I'd reached the same 
conclusion from the trigger side, `SageMakerNotebookJobTrigger` is a plain 
`BaseTrigger` whose hook is addressed by execution name, and your reason from 
the operator side is the same boundary.
   
   **Ordering / crash risk:** agreed, and it's why this is one PR rather than a 
series. The trigger signatures and the call sites move together, so there is no 
intermediate commit where an operator forwards a parameter its trigger cannot 
accept.
   
   **Signature count:** we're measuring different things, and your number is 
the right one for classes. My 54 was *defer sites* that could not be fixed at 
the call site. Counting classes instead: of the 45 this PR touches, 31 accept 
none of the three, 12 already accept `region_name`, and 2 accept some other 
subset. So ~30 fully closed is correct; I should have said "sites" more clearly 
in the issue.
   
   Net effect after the fix: 113 of 114 trigger constructions forward the hook 
configuration, the remaining one being the SageMaker case above.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to