Host on wifi unreachable from two DL-WRX36 APs but not from 2.4G

Installing and Using OpenWrt
1

I have a strange problem where a linux host becomes unreachable via ping unless I reboot it.

The reason it is strange is that if I flip from 5G to 2.4G on my client, after 10 seconds or so I can reach it! This is regardless of which AP I am associated with on my network (I have two). My client IP on the network doesn't change. I can also reach it fine from the wired network.

The host in question is a rpi4, running home assistant in docker. It is associated with dumb AP1, and has a static IP, it is connected to 5G wifi.

If I connect a client to 5G wifi on either AP1, or AP2, I can't ping it.
If I connect a client to 2.4G wifi on AP1, or AP2, it works fine.
if I used wired ethernet, I can ping it.
It îș pingable in network diagnostics on both APs.

Both dumb APs are running openWRT and both 5G and 2.4G are connected to "lan" so there should be no difference. 5G has no other connectivity issues that I can see. e.g. I can connect to other hosts on the local network and anything externally.

If I go and reboot the rpi, it will be fixed on 5G, until the next time !
If I restart the 5G wifi that the pi is connected to, it will be fixed again until the next time.

The APs are DL-WRX36 devices with different builds, one stable one snapshot.

If the behaviour was local to the AP that is associated with the RPI I would be more comfortable that its a firmware bug that occurs when the client is on the same wifi radio as the host itself (like a client isolation happening when this is not turned on). But because I see the same issue from the other AP, across the lan, that is weirding me out. How could a defect in the firmware affect a client, associated on a different AP entirely, but only if the client is on 5G not 2.4G.

I know I am going to need to share my configs so will do that shortly, just wondering where to start, feels like some routing table type of glitch is happening. I have vlans set up, but the SSID to network mapping ('lan') is the same on 2.4 and 5, this is not related to my guest network.

I figure I am going to need a wire tap for this one at some stage !

Thoughts welcome, will share configs shortly.

2

I have another pi elsewhere on the network, connected to AP2 on 5G wifi. It could not ping the host in question.

But, once I was able to ssh on the host in question and then ping this other pi, the first ping was > 1000ms but after that worked, then this other pi has a route now to the host.

Feels like some kind of switch / routing problem that can be healed with a packet in the opposite direction. But I cannot put my finger on why it only affects the 5G side of the client network yet, across both dumb APs, and not ethernet or 2.4G.

Ideas still welcome.

3

OK I have a strong set of theories emerging.

Googling around, raspberry pi wifi is notorious for not responding to ARP requests because the adapter goes to sleep. Most people work around this by wiring it in to ethernet. Others seem to have had success by disabling power saving mode.

I have disabled wifi power saving on two of my pis now and we will see how they go.

I don't know why this wasn't a problem before BUT -- in my prior network perhaps broadcast arp packets were making it across both 2.4G and 5G and now they aren't.

The reason the pi is pingable "after a while" on the 2.4 network is I think home assistant is making regular API calls to a device(s) that are on 2.4 Ghz wifi, so perhaps all the required devices on 2.4 learn and update their arp tables when the pi makes its own broadcast arp to find its target , but the 5GHz clients just don't see this broadcast happen.

This still may be indicative of an issue on DL-WRX36 with broadcast arp requests not making it to 5Ghz clients. It could be a side effect of WMM mode - I saw something online where one user solved his raspberry pi arp issues by disabling APSD ?

Oh and here sounds like the issue:

Multicast to Unicast on the DL-WRX36 is reported to fix it, but I found that this breaks DHCP for clients that are bridged (at least, using an airport express as ethernet to wifi bridge)

Edit: because I only saw this issue on 5G, I have re enabled multicast to unicast on 5G only, and am using 2.4G for my bridged device.