[Fixed] WIFI connectivity issues when upgrading from version 19 to 22

[cowwoc]

I just want to document this problem in case anyone else is having this same problem.

Problem description
Upon upgrading my WRT1900ACS router from OpenWRT 19 to 22, my phone's WIFI connection kept on getting "stuck" seemingly randomly. By "stuck" I mean that the phone would think it is connected, but the web browser would hang (waiting on the network) if I tried opening new pages. Every time this would happen I had to disable/re-enable the phone's WIFI and everything would be back to normal. A few minutes later, the same problem would occur.

Cause of Problem
After many tests, I realized that the connection was only hanging after 5 minutes being idle (e.g. because I was reading a long article). Upon scanning the WIFI settings I ran across a feature called Station inactivity limit which has a default value of 5 minutes.

Fix
Go into Network -> Wireless -> Edit -> Interface Configuration -> Advanced Settings -> increase "Station inactivity limit" to a large value.

Warning

Do not disable "Inactivity Polling". It does not do what you think it does.

"Station inactivity limit" drops idle connections after a certain timeout. "Inactivity polling" prevents idle connections from being dropped if they respond to the polling packet.

In my experience, "inactivity polling" is not good enough. If the router attempts to poll an idle mobile phone connection, and the phone is in power-saving mode, then it will not respond with an ACK packet. The router will drop the idle connection, but the phone won't know. The next time the phone tries using its connection, it just hangs. The response never comes back. Users are forced to manually reestablish a connection by disconnecting and reconnecting to the WIFI.

As far as I can tell, the only thing that works is increasing "Station inactivity limit" to a large value like a day. It is highly unlikely connections that are idle this long will ever come back. This avoids the hanging connection problem I discussed above.

I hope this saves others the hours I spent chasing this down

Proposal
I suggest shipping OpenWRT with a large "Station inactivity limit" value. Android phones don't handle silent disconnects gracefully and this problem is hard to diagnose. "Disassociate On Low Acknowledgement" (which is enabled by default) should be more than enough to kill dead connections.

Changelog

• 2024/06/28: Added a section explaining why it doesn't help to disable "Inactivity Polling".

[svrocket]

@cowwoc

WRT1900ACS

just want to confirm cowwoc's experience and say yeah it was exactly the same for me.

I have multiple WRT-32x's (armada-385-linksys-venom mvebu cortexa)
and openwrt v19.x sucked so bad and I lost so many hours of my life trying to figure it out, that I fell back to openwrt v18.x which has been performant and stable and wonderful for 4+ years. The problem was introduced in v19, and I have avoided 'upgrading' all this time because the upgrade is a downgrade, despite the version number getting larger.

I've been around since the days of chaos calmer & barrier breaker, across multiple generations of hardware, so this aint my first rodeo.

I'm about to try the space-fold-jump to openwrt v22 (thats why I'm back on the forum), assuming its all better now.

[cowwoc]

Thank you for the confirmation. Out of curiosity, do you have WDS set up?

My setup consists of two routers connected using WDS and for some reason I don't recall the primary router (the one that was connected to the internet modem) ever disconnecting WIFI connections. Was this your experience as well?

[svrocket]

WDS = wireless distribution system, I probably tried that in the beginning, because I was using mutiple routers to geographically extend the same SID across the property. But I quickly gave up and just named one SID "WestSide" and the other "EastSide", they provided access to the same LAN ip subnet.

In the middle phase, I stripped everything down to no firewall, single DHCP + DNS master, arpwatch & rsyslogd & continuous perl/logwatch following and counting events in the logs, eventually got to the point where I would use crond to recycle the network / wifi service 2x day just to guarantee the service was available. I used to make fun of the old IBM mainframe guys who "had to IPL every day" because IBMs RACF was a piece of poo-poo software. And yet here I was with openwrt v19.0.7, walking around with a handheld Android Wifi Survey app - like I was Star Trek Scotty with a Tricorder, taking readings from various spots in the property, powering OFF clients that were too old to support the latest protocols. I would try hostapd instead of hostapd-mini, read about legal/FCC requirement for the wireless broadcast signal...

My problem wasnt "random clients cant connect to the WiFi, or couldnt resume after Android sleepytime" my problem was "couldnt figure out why WiFi SID drops off the air", and yes I know about the feds dfs channels, the third radio in the WRT32x, I swapped hardware, in the end, writing /dev/zero over the whole flash storage media and re-installing openwrt v18.0.8|9 is what made "stability".

The only time my openwrt nodes reboot now, is when the electricity drops due to hot weather/wildfires/rolling blackouts.

Uptime is 103 days, today.