This is still happening in OpenWrt 19.07.2.
I can basically confirm this, though some details differ.
I have DS-lite as well. Luckily, my OpenWrt/4040/19.07.02 seems to be a bit more stable than Rainer's most of the time. I usually experience arbitrary hangups every one or two days. On one day, several hangups occurred shortly after each other, but this happened only once. The hangups are not related to high load, they happen "out of the blue".
Except for once, though, I cannot access the box using WiFi. (I could once, but this didn't provide any insights.) I've tried the "standard" switch configuration and others. Especially, I've tried a configuration without WAN-Port/eth1, i.e. I've put the upstream link in a VLAN of its own. Doesn't help.
I haven't tried turning off ipv6 yet, though. I have just found this thread, maybe I'll give it try later.
I don't have any diagnostics to contribute. I've kept a "logread -f" running in a terminal, doesn't show anything when the box goes down. I've had collectd gather some values (okay, that 's far from "real-time" with one sample every 10s) and there isn't anything that "announces" the hangup beforehand.
Sigh.
As I've been sick and tired of the instabilities of OpenWrt on two Fritz!Box 4040 and the Linksys EA8300, I thought a give the ZyXEL NBG6817 a try and wow this thing just works stable as I was used to from the old Buffalo WZR-HP-AG300H. I can only assume that this issue might be related to the amount of RAM. The unstable devices had just 265 MB and I could get them down easily by just watching a YouTube video, downloading somehting with wget, git or Steam. The ZyXEL has 512 MB and I think this is the reason why it is so stable (but as it has another chipset, it might be related as well). The old Buffalo router was stable from my point of view because it could not download more than 25 MB/s with its weak CPU.
I wasted a lot of time, money and nerves, finally the ZyXEL is the solution for me and I can say after 2 weeks uptime that it is worth every penny.
The IPv6 issues are gone for me after a recent commit:
We are running dual-stack OpenWrt 21.02 and have similar issues. So the described commit does not fix the situation. Maybe we should completely turn off segmentation (so ipv4 segmentation)? Any idea how to debug?
We wrote now naywatch to deal with it: