Repository navigation
TL;DRThe multicast plugin reflects LAN mDNS one-way onto the internal ContextSince 2020, Supervisor unconditionally runs the multicast plugin ( Originally it reflected mDNS bidirectionally between the host interfaces and the internal Why deprecateConsumer analysis. None of the components that need mDNS obtain it through the plugin. Home Assistant Core runs on the host network with its own zeroconf. Observed effect of the 2022 change. The removal of outbound reflection in February 2022 amounts to a natural experiment. The traceable fallout over four years consists of two issues ("AirPrint support", "Bonjour not working") on the CUPS app's repository, neither connected to the plugin nor escalated. The plugin's own tracker collected one issue between February 2022 and February 2026, an unrelated feature question. The reports that were filed concern interference instead: multicast storms, ChromeCast breakage, Roon interference, Alexa problems with high CPU (plugin-multicast#137, open). Users affected by these have no supported way to disable the plugin and have resorted to tagging dummy images over it. A more recent observation: host-network apps that listen on all interfaces — matter.js among them — receive each LAN mDNS frame twice, once on the physical interface and once as the reflected copy on the Remaining functionality. What the plugin still provides is one-way copying of LAN mDNS onto the bridge, and that with caveats. Reception is unreliable by construction: the repeater joins the multicast group only on the The pimd attempt (2021–2024). plugin-multicast#17 proposed integrating pimd to "support all multicast protocols", driven by RaspberryMatic users whose Homematic IP wired gateways (HmIPW-DRAP/HmIP-HAP) use UDP multicast on 224.0.0.1/224.0.0.120. The attempt (PR plugin-multicast#61) got pimd compiled and running (after enabling Plugin usage itself is not measurable: it runs on every installation and its work is passive packet copying, which leaves no signal in telemetry. A public comment period and a gradual rollout are therefore the available probes; the plan below is shaped accordingly. ProposalDeprecate gradually behind a feature flag, with each phase independently shippable and reversible until the last:
If phases 1–3 surface users who need actual reflection, the escape hatch is an opt-in community app (avahi with This discussion is also part of the probe: if you know of an app or setup that relies on LAN→bridge mDNS reflection today — a bridge-network app discovering LAN devices via mDNS without Replacing the use casesThe plugin's original intention splits into two replaceable capabilities: making app services discoverable on the LAN (mDNS announcement), and giving apps access to multicast traffic in general (which reflection and routing both failed to deliver). Each has a targeted successor. mDNS announcements: systemd-resolved DNS-SDEvery concrete demand we can find for LAN-facing mDNS from apps is announcement, not reflection: addons#4742 (Samba registering A separate proposal (link to be added) covers this: a Supervisor runtime API modeled on the existing discovery API, publishing app services through systemd-resolved's DNS-SD registration — a Notably, that API is not a sequencing prerequisite for this deprecation: the plugin has not carried announcements out to the LAN since 2022, so removing it takes nothing away here. The two can proceed on their own timelines. General multicast: L2 presence via macvlan/ipvlanFor everything beyond announcing services — mDNS browsing, SSDP, Homematic-style UDP multicast, and multicast protocols we haven't thought of — the pimd chapter taught us the structural answer: link-local multicast cannot be reflected safely or routed at all; the app needs to actually be on the network. Today that means References
Changelog
|
Replies: 1 comment
|
Discussed, approved 14th Sept |
Discussed, approved 14th Sept