Fix 2.7.0 db migration job errors

[LipuFei]

Related to #33651

Fixes 2 issues when running Airflow 2.7.0 python 3.11 on Kubernetes with CeleryKubernetesExecutor and Sentry.

• A circular import from kubernetes_executor.py
• Do not validate SQL when Sentry is initialising. It's still in the configuration phase.

I've tested this on my Kubernetes cluster and it works.

The only doubt I have is the validate=False change. I'm not sure if this is the intended use.

[potiuk]

Nice catch !

[ephraimbuddy]

How do we prevent this from happening again? I think there's a bug somewhere because it was made to work when imported directly. See

airflow/airflow/__init__.py

Lines 79 to 86 in 2ae1c10

	# Things to lazy import in form {local_name: ('target_module', 'target_name')}
	__lazy_imports: dict[str, tuple[str, str]] = {
	"DAG": (".models.dag", "DAG"),
	"Dataset": (".datasets", "Dataset"),
	"XComArg": (".models.xcom_arg", "XComArg"),
	"AirflowException": (".exceptions", "AirflowException"),
	"version": (".version", ""),
	}

[potiuk]

I think it's difficult to prevent - or detect upfront. This was fairly specific to having sentry integration enabled. The only way I can think of is to run our "test migration" job in CI to run will all possible options (like sentry) enabled. All options that can potentially follow different "flow" of imports.

Unfortunately the way how our configuration/imporrs are done does not make it easy to avoid/detect.

[gil-tober]

@LipuFei
Does this also closes #31442?

[LipuFei]

@LipuFei
Does this also closes #31442?

Yes, I will close #31442.

Sorry, thought it was my ticket... This doesn't close #31442.

[ephraimbuddy]

I think it's difficult to prevent - or detect upfront. This was fairly specific to having sentry integration enabled. The only way I can think of is to run our "test migration" job in CI to run will all possible options (like sentry) enabled. All options that can potentially follow different "flow" of imports.

Unfortunately the way how our configuration/imporrs are done does not make it easy to avoid/detect.

I was thinking that the import should work given that we implemented a lazy import for AirflowException so that if it's import directly it works as seen here

airflow/airflow/__init__.py

Lines 79 to 86 in 2ae1c10

	# Things to lazy import in form {local_name: ('target_module', 'target_name')}
	__lazy_imports: dict[str, tuple[str, str]] = {
	"DAG": (".models.dag", "DAG"),
	"Dataset": (".datasets", "Dataset"),
	"XComArg": (".models.xcom_arg", "XComArg"),
	"AirflowException": (".exceptions", "AirflowException"),
	"version": (".version", ""),
	}

, same way we import DAG directly.