Hi OP @mitchcze,
We've seen this problem a few times all ready. We've got sites with bridged 60GHz Ptmp and Ptp setups, worked fine. This recent site we had, with a simple Ptmp setup (2 stations), started losing connectivity like you described. The odd thing about it is that there's mtk switches behind the stations, and we can ping it! Traffic seems to pass - there's a baseload of camera traffic and we see it traversing. From the mtk switches, we can ping the station in front of it and vice versa.
We've tried multiple fixes as per directed by mtk support:
one is turning on mdmg-fix. Only available through cli
/interface w60g set 0 mdmg-fix=yes
. No detailed information other than it improves 60GHz reliability (EDIT: checked again, apparently it does this: Experimental feature working only on wAP60Gx3 devices, providing better point to multi point stability in some cases)
second is applying an admin-mac on the station themselves.
third is changing STP protocol mode to none.
There are no consistent effect on the problem from the solutions above. They recommended upgrading it to a new version (at that time), 6.47.6, the problem went away! Now for the recent site, we've downgraded it to the same version, the problem is still there. We tried the above fixes, nothing happened. We even tried to work around it for now. Had a netwatch running, pinging the stations and once it drops, disable and re-enable. I know it's a messy workaround at the moment but we're a bit desperate. Lol
We reckon it's an arp issue, as the station interfaces would all have the same mac address dynamically. It creates an arp conflict somewhere down the pipe. We're not sure really, we are still experimenting but left it on the back burner for a while. Hope you've found a solution you can share.