I just completed new tests.
First, I tried to add a new ZigBee device (Aqara magnet) using the « standard way »: configuration | Integration | ZigBee “Configure” | + Add Device. On purpose, the device was located beyond two thick walls and is therefore out of range of the Coordinator. A router device (Philips hue bulb) is placed mid-way between the Coordinator and the Aqara device. The add device command fails.
Second, I have used on the Router device (Philips) the “add devices via this device” command. A window similar to the “normal” add device widows open and after a few seconds, the Aqara device is correctly discovered. I tested the behavior of the device in Lovelace and it reports correctly the close/open state.
This is quite interesting because this seems to indicate that when using the “normal” add device command only the Coordinator is looking for new devices and that the search is not “pushed” to the Routers. This is may be why most people recommend adding new devices close to the coordinator. Hopefully once added in the network and placed to their final location the mesh should heal itself and the parent of the end device should be transferred from the coordinator to the router?
I am still unclear about the meaning of LQI and RSSI values (see Reported LQI values for Zigbee devices with ZHA integration for interesting question/information). As stated in this reference, it is unclear if the LQI / RSSI only refer to the last hop or if the values take in account the complete path from the ZC to the ZED. In my test, the LQI for the ZR (Philips bulb) and the ZED (Aqara magnet) are equal to 255 and this seems very high as the devices are located far through wall?
Interesting also is the fact that the connection between the router and the end device show up on the network graph! As you know the arcs that tie the nodes together display LQI value(s). On a Coordinator (ZC) to Router (ZR) connection, there are two values v1/v2? Originally, I thought these two values represented max/min values. However, by looking at the content of the zigbee.db I have found that actually these two values correspond to the LQI in each direction of the link (strange no? why different values?). On a Router (ZR) to End Device (ZED) connection, only one LQI value is displayed. This probably make sense because only the children can initiate an exchange with its parent.
As I have moved the Philips Router device in a location placed between the coordinator and the Aqara weather end device the network should normally “heal itself” and switch the parent of the Aqara end device from the coordinator to the router. By doing so, the RSI would probably change from around 100 to around 255. There are a lot of articles on the subject of ZigBee network construction/healing. If I understand correctly, in my case the Aqara end device should keep track of a working but bad link to the coordinator, and use a better link to the coordinator through the router. All the different possible paths should be displayed on the graph (extracted from the routing table kept in devices?). As I understand, the process takes time and therefore I will allow 24 hours for this process to complete. If the link is not modified automatically, I will try to force it to happen using the “add devices via this device” command.