Flaky matter over thread connections with battery powered Aqara devices and Eero routers

Configuration Matter/Thread
1

I've been having an ongoing issue with two Eeros (6E gateway + 6E acting as a wireless extender). Any battery-powered (sleepy) Matter-over-Thread device that parents to the bridge goes unavailable in Home Assistant while still answering ICMPv6 pings... i.e. it's reachable at the IP layer but HA can't sustain the Matter subscription across that path.

Physically disconnecting the bridge forces the devices to re-attach to the gateway, and they immediately work and stay stable. The gateway is closer/stronger (~10 ft vs ~20 ft), so I don't know why they attach to the bridge in the first place, or why they never reparent... though I think that's because the Thread link to the bridge stays "up" from the device's perspective... so nothing triggers a re-scan.

My one mains-powered Aqara wall switch (a Thread router, always-on, not parented) never has the problem (and it's in the same room as one of the sensors that becomes unavailable), so it's specific to sleepy end devices.

Anyone encounter an issue like this before? For now I guess I'm just resorting to disconnecting the bridge for a minute when it happens. It's not incredibly frequent... I can have these devices up for a week at a time.

2

I’m just investigating a similar problem and have come across this. I’ve got a handful of IKEA matter sensors, and three Eero 6+ routers in a WiFi mesh. I’ve seen that devices can go offline in HA’s Matter server display, but still be pingable. In my case it’s been both battery/sleepy devices and mains/routing devices (actually USB, but you know what I mean). The best fix was rebooting the Eero devices, although anything that caused the network to take a good hard look at itself seemed to help for a while. I’ve tried both Apple and HA as Matter controllers, but the effect seemed the same. I’ve heard suggestions that there’s a problem with the Matter service discovery (mDNS?) messages from Eero devices, although I don’t have enough knowledge to prove/disprove that.

However, I also happen to have some Nanoleaf Shapes that have Thread Border Router functionality, so my latest test is to disable Thread on the Eeros and enable it on the Shapes so I just have a single border router. It’s too soon to be conclusive, but things are running OK so far.

Are you still rebooting devices when they fail, or have you come up with a better solution?