Released v1.3.5:
Features
• Direct Incus connections: Connect without a pre-configured incus remote using --server-url.
• Remote bind mounts: Pass host bind mounts through to remote Incus servers per-volume via x-incus-compose.remote: true or globally with --allow-remote-mounts. docs
• Unmanaged external networks: External networks pointing to unmanaged host interfaces (such as br0 with MANAGED: NO) now automatically attach using nictype: bridged / physical / macvlan with parent: <name>. docs
• Directory seed updates on up: Seeded directory bind mounts (x-incus-compose.seed: true) now re-seed on every up, updating matching files while leaving existing volume data untouched.
Fixes
• Set lxc.environment.hooks on Incus < 7.6 so containers with a non-standard PATH (e.g. ghcr.io/nix-containers/images/*) don’t break host LXC hooks.
• Flush network aliases to raw.dnsmasq even when all services set container_name equal to their service name (#206).
• Validate down --rmi values (local / all) and service arguments passed to down (#220, #221).
ic-dns in v1.4.0: opt-in or opt-out?
Our next major release (v1.4.0, planned right after Incus 7.0.2 LTS) introduces ic-dns incus-compose/docs/root/dns.md at develop · lxc/incus-compose · GitHub, a dedicated DNS sidecar that handles service discovery and DNS routing for managed containers.
We originally planned to ship ic-dns as opt-out (enabled by default) in v1.4.0, and we’re confident in its stability.
However, since routing all container DNS through ic-dns is a significant architectural change, we’re considering making it opt-in in v1.4.0 and switching to opt-out in a later release.
What do you prefer for v1.4.0 — enabled by default (opt-out) or opt-in first?