My original post from about a year ago with the same subject began with:
I have noticed random and sporadic WiFi / ARP problems on my network, usually after something reboots. However, the problems seem to always clear up after a while, and I'm wondering if this is "normal"?
Here I am posting a solution as a new post, because my original post has been closed to any new replies (first post was about a year ago, so apparently too old). Doing this in case anyone else runs into this and searches this forum for ARP problems, as it turns out it was an unintentionally self-inflicted problem not directly related to Orbi WiFi.
The problem was that various nodes on my network could not see a handful of other nodes, but there were different subsets of which nodes could not be seen depending on which system I checked. Simple pings would fail. In all cases, the ARP for the nodes that could not be seen was listed as 'incomplete' (Linux, macOS, and Raspbian). Even more frustrating, if I waited long enough, the ARP problems would usually resolve themselves, but then sometimes then problem would re-appear elsewhwere on the network.
Long story short, it turned out that I had a subnet mask problem, as I had a half dozen Raspberry Pis that were dual-hosted (WiFi and Ethernet), and on several of those I had not configured a proper subnet. Because of a bad subnet mask, those nodes thought that the WiFi and Ethernet were both on the same subnet, but obviously were not (one was 10.0.1.0/24 and the other was 10.0.2.0/24, but the mask was 255.0.0.0). It appears that the ARP replies got confused as to which interface was appropriate. I believe that the reason why it sporadically worked was when the ARP cache expired, sometimes that ARP response would be seen the right interface.
What made it even more confusing was that even nodes that had the proper masks could sometimes not communicate between themselves, which I don't fully understand, but everything has been working much more stable and reliable ever since I fixed the masks on those several nodes, so I presume that it was an issue of the wrong ARP responses being cached somewhere.
So it was a self-inflicted problem when I upgraded some old Raspbery Pis from a Debian configuration in which the subnet mask was an exlicit entry (separate line) to a CIDR (when Debian moved to dhcpcd.conf), and while I removed the netmask line from the configuration, I forgot to specify the network size (/24) at the end of the IP.
So just wanted to document how a simple configuration error on a handful of nodes can inject inconsistency and unreliability into the overall network. To which I am hearing some of the network gurus on this forum shouting 'Duh!'. Usually I wouldn't make such an obvious mistake, but the inconsistency of the problem is what threw me off, because it wasn't obvious where I had done something wrong, and I have 50+ nodes on my network (and the fact that I was cloning Raspberry Pis from a (bad) standard configuration helped propagate the problem further).
So just posting this in case someone else sees a similar problem, as honestly I originally hadn't even thought that this could be a subnet mask problem, so I never bothered to check the obvious.
