As I mentioned in another post ,

I wrote:

In response to what admin3 wrote:
With that said though, I am seeing something odd on my iPhone with the 2865ac's wireless, where it'll be connected but not be able to send/receive.

I've got the same symptom with ESP8285's (contained within Sonoff S26 Wifi plugs ) - but in conjunction with a Vigor AP903. They are running firmware I have written, so I've assumed that it is something I'm doing wrong ... (Hitting the "DeAuth" 'button' on the VigorAP903 recovers the situation almost instantly.

I've investigated this further and have narrowed it down to some sort of combination of devices - one of which is the Vigor AP903 (running V1.4.2). As mentioned in the quote above, the devices are Sonoff S26 WiFi plugs, which are based around ESP8285 SoCs .

I don't know if the fault is present with the manufacturer's firmware, because I bought them to modify and have never run them in their original state. My firmware uses the built in STA client to connect to an available access point (2.4GHz only) and implements a Web Server on port 80. The vast majority of the time, there is absolutely no network traffic to or from these devices...

What I found, is that if they connect to my old 2860n, or my Vigor AP802 they work fine. If they connect to the Vigor AP903 - and there is no traffic for around 75 seconds - they stop working. The Vigor shows them 'connected' , as does the ESP's debugging data: they just don't work. Leaving a job running that pings them every 60 seconds alleviates the problem. The built-in IOS keep-alive function on the AP 903 has no effect. As noted in the quote, they still seem to have enough connectivity with the AP903 that they respond to being "De-Auth'ed" - whereupon they reconnect and work for another 75 seconds.

As a workaround, I have modified my firmware, so that it sends "Hello" via UDP to Port 7 (ECHO) on the Vigor AP903, every 10 secs, which fixes the problem. (The Vigor replies via ICMP that it doesn't support the "ECHO" service and helpfully includes a copy of the data that it can't echo - you couldn't make it up )

Has anyone noticed this effect with other devices? I should imagine that most people have many clients running on their phones that 'call home' frequently - but just occasionally, I've noticed my Huawei Mate 20 Pro needs the WiFi toggling OFF and ON, otherwise it too stops working (it normally manifests as a "DNS lookup failure" - which is drivel).

only 3000 chars :?

I wrote:
only 3000 chars :?

Can't the forum software limit be increased :?: ... it's not 1995

Especially since it counts all the 'markup' information as characters as well.

I wrote:
If they connect to the Vigor AP903 - and there is no traffic for around 75 seconds - they stop working. The Vigor shows them 'connected' , as does the ESP's debugging data: they just don't work.

It's actually more nuanced than that ...
...it turns out they don't completely stop working - they actually become isolated from the LAN.

I waited for one to enter this condition and tried to access it from a Windows 10 PC connected to the same AP - and succeeded. This AP was actually connected to the 5GHz band at the time, rather than the 2.4GHz band that virtually all I.O.T. devices have to use.

It turns out that my phone can still access them as well - again, despite being connected on 5GHz. Repeated 'keep-alive' pinging from the phone (on WiFi) stops the ESP8285 from becoming inaccessible to the LAN, but will not bring it out of this state, if it has been 'allowed' to enter it.

To add: accessing the Vigor AP903 via telnet and issuing a 'Ping' of the ESP device also stops it from becoming isolated (the ping runs until manually stopped).
Also, worth noting that I have no 'isolate' options set on the AP903 - that I can find.

This bug must have taken quite a lot of coding

I wrote:
The built-in IOS keep-alive function on the AP 903 has no effect.

I was wrong about that - it does work around the issue.

I came across, what I thought was a related issue, which the ESP8266 community have been investigating for a while: https://github.com/esp8266/Arduino/pull/6889

It turns out to be a different problem, but the fix (to send periodic 'gratuitous ARPs') also works around my problem.

It got me thinking though ... I believe I have found a different, serious, reproducible problem with the Vigor AP903, that I will document (and report to Draytek) separately. This concerns ARP requests made to Vigor AP903 wireless clients. In a nutshell, they don't work!. The problem seems to independent of the actual client - I can reproduce it with a brand new HP AIO machine.

See: https://forum.draytek.co.uk/viewtopic.php?t=24568