We don't know the details exactly yet but we are moving network communication between mediawiki and swift to the service mesh to improve observability and performance (See T328872#11783346) but around 15 minutes after a pod is created, the side car for envoy start returning timeout error for connection to swift (which is annoying, we deploy it, everything is fine, then everything breaks). For now, we reverted the change to use the old way but we are gonna try again after figuring out why envoy decides to start returning timeout error (a hack could be that we just add a retry but I need to see if that would work, I don't know envoy well)