Docker Engine 29.7's overlay networking breaks every Swarm task without IPv6
I asked a Swarm service for twenty replicas on an overlay network and got zero. Not slow, not partially up. Zero, forever, with the exact same command that gives me twenty on the engine version one release behind it. I'd gone looking at the Docker Engine 29 release notes for something smaller: a line in the 29.8.0 changelog about spreading out the daemon's periodic overlay-network gossip so it…
On October 9, 2023, the latest version of Docker Engine, 29.7, was released, but it broke Swarm tasks on overlay networks that lacked IPv6 support. This issue caused all Swarm tasks on such networks to come up with zero running instances, regardless of the number requested. To test the problem, a reporter created a Swarm service with 20 replicas on an overlay network named "gossip-net" using the Alpine Linux 3.20 image.
On Docker Engine version 29.6.2, the service came up with 20 running tasks as expected, but on Docker Engine 29.7 and later, the service returned 0 running tasks. The reporter confirmed the issue by running the same commands on different versions of Docker Engine and observed the same behavior. The problem was not due to any bind conflicts or other factors, as the reporter tested various configurations.
The only distinguishing factor was the absence of IPv6 support on the host machine, as confirmed by checking the system's IPv6 settings. Docker Engine 29.6.2 handled the situation by completing the join over IPv4, while newer versions of the engine (29.7 and later) failed to determine the transport's address family, leading to the "address family" error and the inability to join the overlay network.
The reporter attributed the issue to a change in the Swarm overlay network driver's handling of VXLAN data plane address families when IPv6 is not available.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
