Dual stack network is not detected with several servers #4996
Comments
|
After a few months, I realized there's an option I guess this makes the issue even more minor, because a server host can just determine the type of network once and then change the config. However, if there are plans in the future to investigate how this happens (otherwise I guess it should be closed), I gathered some stats from when I tried to restart the servers. They are below. I made 294 attempts to start a server for it to be dual-stack, as follows: start the server, if it doesn't have dual stack network, repeat the procedure (or stop it, in 15 cases). Most server restart attempts were grouped together (when I restarted all the servers), their results could be dependent in some way. There were 839 server runs in total. 419 (50%) runs ended up detecting only IPv4. Of 279 attempts ending with server having dual-stack network, 39% finished after one run, 57% - after at most two, 70% - after at most three runs. 8% attempts took at least 7 runs, and in two occasions, it took 11 and 12 restarts, respectively. The 2.x server which used a different config folder failed to detect dual-stack network only once. Generally it seems like servers which are less used have a better dual-stack detection rate. |
Description
I'm running a few servers on the same machine that has both IPv4 and IPv6, and since around the end of 2022 with high probability not both of them are detected. That is, when starting the server, there's around 35% probability that IPv6 won't be available, around 35% probability that IPv4 won't be available, and otherwise dual stack network is detected.
In all those cases, the machine can successfully ping anything else in the network using any of protocols, and STK is the only service unable to detect something. The probabilities don't depend on almost anything (like server config, server type, port), except that two servers restarted at the same moment are a bit more likely to detect the same type of network. Sometimes I'm lucky to restart servers with 1.5x number of restarts, but for one specific server there might be needed up to 10-20(-rarely 40-...) restarts to make it detect both IPvX, and so most of the time I stop when there's at least IPv4 detected.
While I'm not typically using the standard version of the STK servers (though I never modified networking code, especially in the part of connection), I tested the standard version a few times throughout this time and it fails in the same way. At some point, I tried to bisect to find a commit where it breaks (both for the version I use and for the standard STK) but it failed for both on all commits made in a few last years. I suspect the problem is in the settings. Upgrading the system (Ubuntu server) did not help on either of occasions. There are no other major apps running on the machine.
I have no idea about how detection is done, to be honest. At some point I thought it's done using stun servers, and then I found out that linuxreviews stun server doesn't respond. Removing it and adding google stun servers slightly improved the situation, but now sometimes I get a strange warning
STKHost: Failed to get enough stun servers using SRV record.that otherwise doesn't affect anything. Logs are listed below.Strangely enough, starting a single 2.x server (which has another config folder) alongside all the 'buggy' servers is OK and it detects dual stack network almost always (haven't tried doing it 1000 times, though). Starting a single server per machine is also OK in the same fashion. Bots have no problems connecting to servers no matter which type of network the server detected (which is kind of logical).
I acknowledge that while server hosting is not very popular, the issue only touches directly a few people at max, but it's still quite annoying because it either causes people with non-dual stack network to not be able to join, or forces to spend more time restarting the servers (and makes it impossible to restart automatically, which means that a properly updated server requires presence in person — and sometimes it's hard to guarantee that).
Steps to reproduce
Start several servers with different ports on the same machine.
Configuration
STK release version: latest git but probably almost any commit works
STK Source (PPA, distribution package, official binary, etc.): git pull
System: Ubuntu Server 20.04, 22.04
Graphics card, CPU: ???
Additional information
Typical success log (IPv4 detected, IPv6 detected):
Typical 'fail but no one cares about IPv6' log (IPv4 detected, IPv6 not detected):
Typical fail log (IPv4 not detected, IPv6 detected):
Note that I never experienced situations when both protocols were not detected.
Used stun servers (different from default only by removing linuxreviews and adding google I think):
The text was updated successfully, but these errors were encountered: