Under which category would you file this issue?
Airflow Core
Apache Airflow version
3.2.2
What happened and how to reproduce it?
Following an update to Airflow 3.2.2 logs are no longer being uploaded to remote destinations such as S3.
What you think should happen instead?
Remote logging is working fine in Airflow 3.2.1
Task logs should be uploaded to remote destinations and displayed in UI
Operating System
Linux
Deployment
Other
Apache Airflow Provider(s)
No response
Versions of Apache Airflow Providers
apache-airflow-providers-common-compat==1.15.0
apache-airflow-providers-common-io==1.7.2
apache-airflow-providers-common-sql==2.0.0
apache-airflow-providers-standard==1.13.1
apache-airflow-providers-cncf-kubernetes==10.17.1
apache-airflow-providers-mysql==6.6.0
apache-airflow-providers-databricks==7.15.0
apache-airflow-providers-openlineage==2.17.0
apache-airflow-providers-datadog==3.10.4
apache-airflow-providers-http==6.0.2
apache-airflow-providers-pagerduty==5.2.5
apache-airflow-providers-amazon==9.29.0
apache-airflow-providers-ftp==3.15.0
apache-airflow-providers-sftp==5.8.0
apache-airflow-providers-trino==6.6.0
apache-airflow-providers-cube==1.0.0
apache-airflow-providers-slack==9.10.0
airflow-provider-great-expectations==1.0.0
Official Helm Chart version
Not Applicable
Kubernetes Version
No response
Helm Chart configuration
No response
Docker Image customizations
None
Anything else?
Following warning is logged:
{
event: Unable to find AWS Connection ID 'aws_default', switching to empty.
filename: base_aws.py
level: warning
lineno: 639
logger: airflow.task.hooks.airflow.providers.amazon.aws.hooks.s3.S3Hook
timestamp: 2026-06-09T21:10:22.495664Z
}
The root cause of this issue is the connection preloading behavior on the worker.
Following work done in this PR for Two-token mechanism for task execution, the worker, at the start, now gets a workload scoped long-lived token which does not have access to connections and so the connection preload logic gets None for connections and caches it here.
A potential fix is simple one-liner: do not cache None allowing fall-through logic to hit actual connection get call here. This connection get call will succeed following task execution because the worker token is updated to execution scoped token via the API client's _update_auth() hook once a task is run/started.
Are you willing to submit PR?
Code of Conduct
Under which category would you file this issue?
Airflow Core
Apache Airflow version
3.2.2
What happened and how to reproduce it?
Following an update to Airflow 3.2.2 logs are no longer being uploaded to remote destinations such as S3.
What you think should happen instead?
Remote logging is working fine in Airflow 3.2.1
Task logs should be uploaded to remote destinations and displayed in UI
Operating System
Linux
Deployment
Other
Apache Airflow Provider(s)
No response
Versions of Apache Airflow Providers
Official Helm Chart version
Not Applicable
Kubernetes Version
No response
Helm Chart configuration
No response
Docker Image customizations
None
Anything else?
Following warning is logged:
The root cause of this issue is the connection preloading behavior on the worker.
Following work done in this PR for Two-token mechanism for task execution, the worker, at the start, now gets a
workloadscoped long-lived token which does not have access to connections and so the connection preload logic getsNonefor connections and caches it here.A potential fix is simple one-liner: do not cache
Noneallowing fall-through logic to hit actual connection get call here. This connection get call will succeed following task execution because the worker token is updated toexecutionscoped token via the API client's_update_auth()hook once a task is run/started.Are you willing to submit PR?
Code of Conduct