Game voice: reverse proxy/UDP filtering issue investigation

nta · 2021-07-06T06:29:38.350Z

Does this also include logs where users have not had any connection issue? I’m not seeing any UDP disconnects or even any reconnects at all in at least one of them.

nta · 2021-07-06T20:15:58.334Z

8 posts were split to a new topic: Native audio: @manzarek’s investigations

BradFitt · 2021-07-06T10:49:13.011Z

No, that’s all from affected players.

BradFitt · 2021-07-06T10:52:33.044Z

Did your external mumble server get protected by OVH Firewall? It doesn’t even matter if you are using external server in my opinion. If your external server is also protected and everything is fine, you can simply use the internal server.
I never tried 3D voice, and I believe most of users are using native voice.

BradFitt · 2021-07-06T10:54:00.575Z

Correction, Cloudflare Spectrum is L4, not L7, I believe.

nta · 2021-07-06T20:19:16.820Z

3 posts were split to a new topic: Game voice: unspecified crashes

manzarek · 2021-07-06T12:01:52.396Z

BradFitt:

you can simply use the internal server.

Not at 800+ players, you need murmur

nta · 2021-07-06T12:14:11.972Z

Yeah, then you probably have an entirely different issue since these users didn’t even actually try to reconnect or still didn’t set the developer flag.

nta · 2021-07-06T12:14:38.598Z

Yeah, then you probably have an entirely different issue since these users didn’t even actually try to reconnect or still didn’t set the developer flag.

gottfriedleibniz · 2021-07-06T13:56:34.390Z

FlyingPit are these, or any, players on canary? Get the sense of correlation != causation given the reports of other canary related issues that have sprouted in the past few days.

Based on the CPU load screenshots shown for RSM, I do not see much of an issue (well, not a hill I want to die on) trying to nickle-and-dime clock cycles (et. all) elsewhere by using an external murmur. However, ETL traces of fxserver under load (w/ mumble) seem valuable: quack has shown interest in another round of optimizations.

Finally, I think I understand manzarek’s issue. A potential fix for this also blends into another feature request.

BradFitt · 2021-07-06T19:13:04.850Z

I see. Are you suggesting the reverse proxy causes some other issues, and those issues affect the mumble server?

BradFitt · 2021-07-06T19:16:26.055Z

gottfriedleibniz:

FlyingPit are these, or any, players on canary? Get the sense of correlation != causation given the reports of other canary related issues that have sprouted in the past few days.

Based on the CPU load screenshots shown for RSM, I do not see much of an issue (well, not a hill I want to die on) trying to nickle-and-dime clock cycles (et. all) elsewhere by using an external murmur. However, ETL traces of fxserver under load (w/ mumble) seem valuable: quack has shown interest in another round of optimizations.

Finally, I think I understand manzarek’s issue. A potential fix for this also blends into another feature request.

AFAIK, no. Most of them are staying with the release channel. And also, this issue has been discovered for a while. I don’t believe it’s related to the recent canary issue.

nta · 2021-07-06T20:20:33.837Z

I cleaned this topic up a little so it’d not end up like the last attempt at a voice mega-thread. Please don’t reuse existing topics for tangentially related issues, nor should one hesitate to make a new topic for even uncertain issues - as long as it doesn’t turn into a ‘same issue’-fest without constructive discussion.

nta · 2021-07-06T12:13:38.388Z

ok then did you even try reporting traces of a stand-alone fxs or such?

BradFitt · 2021-07-07T09:19:04.656Z

Exactly same thing happened when I use external mumble server with reverse proxy. However, everything is fine with external mumble server on another server.

nta · 2021-07-07T09:41:50.431Z

The message you’re replying to was a reply to @manzarek above, I’m not sure how it ended up out of order and not being a reply anymore.

BradFitt · 2021-07-07T10:32:53.450Z

Alright… Do you want me to perform any other test regarding this issue then?

nta · 2021-07-07T10:33:47.847Z

Ask @gottfriedleibniz or so, I’m not sure what could still be problematic after the recent switching around of crypto lookups.

BradFitt · 2021-07-07T18:52:35.364Z

Alright… Looks like this will be a permanent problem unless some magic appears.

nta · 2021-07-07T19:36:32.247Z

pcaps may help especially if you can isolate them to a standalone voice server having this same issue by the way, as long as they also contain a user initially starting to have this issue (unless it’s not a source endpoint issue, then it’s a bit useless if there’s no ssl key logging)