We have identified an additional data plane component that has been sometimes failing to handle node restarts for customers whose data planes are hosted on AWS. As a necessary part of the update process, the components will be a subjected to a restart during this maintenance. There is a small chance that tasks that are in the "queued" state in Airflow when this restart occurs will fail. Such failures will respect retries if you have them configured for the task.
The two hour window of this maintenance represents a period of time during which clusters may be updated, but for each individual Airflow deployment, the window of potential disruption will be less than 10 minutes.
Posted on
Nov 04, 2024 - 22:00 UTC