This is mostly likely related to the service discovery problem we’re working on: how do fly.io deploys work? - #2 by kurt
When you deploy, our edges are sometimes 2 min slow detecting the new VMs (and noticing the old one is gone). For apps running a single instance, this causes errors during deploys.
The messages you’re seeing are from the lax edge and the iad worker. They come in pairs because a connection hits lax, it sends it to where the now-terminated VM was (iad, in this case), and then both ends of the proxy emit an error.
We’re working on this, for now the workaround is to scale to ~3 VMs.