Consider using persistent storage for flux source controller

During node rolling upgrades, source controller pod is drained, and it has to pull all the artifacts again.

This extra delay could be avoided by using persistent storage as describes in flux documentation:

This will be probably needed to introduce per-unit timeout: !2383 (merged)

Otherwise various units may experience a large delay to reconcile when source-controller pod is re-scheduled.

Edited Jan 14, 2025 by Thomas Morin