It turns out this issue as deeper than just UPS. The kubernetes `watch` actually closes automatically on all CRDs. This fixes that by listening for the stream close event and reopening it.