Docker 29.8.2 bug: failing swarm service cuts healthy containers off overlay network

On Docker 29.8.2, each failed container start on an overlay network calls Leave twice for one Join, so a node with N containers loses the network after N+1 failures. Two healthy containers went NO-CARRIER within 20 seconds while still shown as running; only a daemon restart fixed it. No release with a fix yet; upstream report is moby/moby#53834.

• Each failed start does one Join but two Leaves, driving the count negative
• A node with N containers loses the overlay network after N+1 failed starts
• Containers go NO-CARRIER yet docker ps still shows them running
• Only systemctl restart docker restores them; no fixed release exists