rns.recipes

On the github release page for v1.1.5, there's also a pre-release build of 1.1.6. This build contains a lot of modifications to the core Transport I/O logic that are intended to make everything much more reliable under conditions such as those seen lately with high levels of DDoS / storm / spam attempts on the public backbone. This is still a work in progress.

There may be bugs and unintended weirdness, so any help testing this will be much appreciated. Especially on public backbone nodes that see a lot of weird shit.

The Transport I/O refactoring also lays the groundwork for a free-threaded version, which will allow RNS to run much more efficiently on multi-core machines, so catching any strangeness early here is very useful.

Also, does anyone know what's up with all those "!a83eb2df" LXMF announces going round lately? Seem to come in bursts, haven't got the foggiest where that's from.

Also, please spread the word if you know someone who could help try this out on a couple of nodes. Any feedback is welcome here or via my LXMF address.

i think those nodes might be faulty styrene nodes, however they could also be someone's mass produced LoRa nodes (maybe russians or ukranians?) unfortunately i don't have LoRa nodes to test on, have you posted this to the matrix chat?

Hm, interesting, hadn't seen that before. Might be.

Nope, I'm not on matrix anymore. Feel free to post it there if you want.

styrene has a general issue with spamming announces because styrened sets the default limit to like 10 minutes or something and it keeps flooding the net, also the russians and ukranians have been active setting up reticulum nodes, i've also seen a few nodes with the same (or at least very similar) prefix announcing themselves as police officer nodes

The !a83eb2df nodes match the Node ID naming convention for Meshtastic. I cross-referenced a handful of the announces with meshmap.net and sure enough they showed as Meshtastic nodes. My guess is that someone's trying to create some kind of Meshtastic bridge ... in a very inefficient or questionable way.

I'll see to testing 1.1.6 this week.

Yes, they seem to be meshtastic nodes, but atm it seems to just spam, without adding any functionality. At least nobody ever replied after I messaged a few of them. They seem to originate from around Helsinki. If someone is motivated one could try to message the Finnish meshtastic discord and ask about it if.

Beleth is running 1.1.6 now. Only one odd error so far, but looking good. Still getting hammered with clients timing out and/or resetting connections, but that's not an rnsd issue of course.

I am running rns 1.1.6 pre-release, Python 3.14.4 with GIL. Looking good. Path requests are getting resolved again and all annonces are coming in again when Nomadnet is connected to a Backbone Interface in Gateway mode.

Nice to hear. Yeah, seeing a lot of nodes disconnecting / reconnecting as well, but that's probably just to be expected with how many are connecting at this point. Most of the work here is just preparation for the no-GIL version, so I wouldn't actually expect it to work well in free-threaded mode yet, but we're getting there.

Either way, I just uploaded a v2 of the 1.1.6 pre-release, with even more optimizations and a few more fixes. It significantly cuts the memory usage as well, but you may have to remove ~/.reticulum/storage/tunnels if you don't want to wait for the eventual cleanup there. Also quite a bit of logic made more robust in terms of background tasks and such. Should be a significant improvement on slower hardware with many connections.

I'll be testing this over the next few days as well. Any observations are very much welcome and appreciated.

Updated to 1.1.6 last night, 8 hours later it had dropped from around 500 tcp connections to 100, I'd been averaging 500 all week so I restarted rnsd. After a few hours it hit 2000. I have not seen that many connections since the first couple of days after standing up backbone interface discovery. Logs are same same, if a connection drops you get routing and epoll errors but everthing else is fine, lxmd is not starved for peers and NN is resolving. All good.

rns 1.1.6-3 pre-release, Python 3.14.4 with GIL up and running. Looking good.

rns 1.1.6-3 pre-release, Python 3.14.4 with GIL up and running. Looking good.

(Forgot my "Name" above)

Great to hear! Let me know if anything weird happens.

hey mark you could use PYUV for backbone interfaces instead of epoll, it acts as a translation layer between epoll, kqueue and IOCP and autoselects the best one according to the OS, that way backbone interfaces would support basically every major OS

Bad idea. Can't use pyuv, as that would add a massive external dependency to RNS, including compiled C code, etc, etc. Anything that close to the networking core needs to stay native to RNS.

Okay, 1.1.6 is out. Thanks for the help everyone!

Signing off for this one, but if you do find any weirdness in 1.1.6, you can leave a comment here, and I'll read it.