Cory Snider

It looks like in that case the embedded DNS server is doing the right thing: upon failing to get an answer from the first upstream server, it falls back to...

Nope, _I_ was looking at the wrong logs. 🤦 The `NXDOMAIN` response from the wrong logs looks to actually be from an independent request which got round-robin routed to udp:1.1.1.1...

I reproduced the issue locally. The name `s3.amazonaws.com.` is just the right length to offset one of the answer records to be cut off at the end of its `length`...

The `TestIntegration/TestFileOpCopyIncludeExclude/worker=dockerd` buildkit integration test is failing, but it is also failing on master so is unlikely to be a Go 1.20 problem. Everything else is green; I think we're...

> * Was the `distribution/xfer: fix download fencepost bug` commit intended to be part of the go1.20 related work? Yes, through a long causal chain. golangci-lint had to be bumped...

@thaJeztah you make a very compelling argument. I've captured that [in the repo wiki](https://github.com/moby/moby/wiki/Bumping-the-Go-toolchain-version) for the benefit of the next person to take on the task of bumping the Go...

#44887 stalled because its various iterations either did not fix this issue or fixed the issue by breaking something else.

> We have subnets with 18 quintillion available IPv6 addresses, each container could just have its own and not have to share ports with anyone, and people _still_ suggest NATing...

> Also, it would mean that containers could no longer access the host IPs. Containers can be connected to more than one network at a time. Connecting to a `bridge`...

> Then another approach I had was dump all the data about a network (`docker network inspect net`), disconnect all containers from that network, delete the network, re-create the network...