| 15 Apr 2025 |
@emma:rory.gay | can draupnir keep up with 500 messages/s? | 23:04:26 |
Cat | That isnt a homeserver size issue. Thats a the bot isnt supposed to scale out that far. | 23:04:53 |
uep | in a sense it doesn't matter.. it's already losing if it has to react and publish more messages to deal with them | 23:05:02 |
@emma:rory.gay | not a single protocol can do redactions without sending more traffic | 23:05:22 |
uep | the fact that attackers send 500m/s as the threshold where the responses are overloaded is more effect than it is cause | 23:06:04 |
emily | I agree when you're talking about people in the room | 23:06:08 |
Cat | Emma Array Redactions is whats being refered to. | 23:06:19 |
emily | I'm not sure "nixos.org Matrix works when nixos.org infra is down" is very valuable to us | 23:06:23 |
Cat | The fact you have to currently fight Symetric warfare. | 23:06:29 |
emily | beyond like, having a place for the infra team to coordinate for that emergency, which could just be a room on another HS | 23:06:33 |
dgrig | It's indeed a cool patry trick, which works amazingly well when you trust all the participants. It works great for the rooms I have with friends where no moderation is needed. It also works great for a closed federation like the one element wants to sell with their clients. Doesn't work that great for a public room anyone can join | 23:06:56 |
uep | In reply to @emma:rory.gay not a single protocol can do redactions without sending more traffic correct, which is why more moderation features as distributed policy is needed in a system that's built on distributed delivery | 23:06:57 |
emily | like nobody considers it a big problem that when discourse.nixos.org goes down people can't use the forum. we just keep it up. since as a project we already rely on a lot of infra | 23:07:02 |
@emma:rory.gay | they dont, but that's an unreasonably low message rate for a server as small as mine | 23:07:15 |
uep | and then we get back to relying on reputation systems and other mechanisms | 23:07:24 |
Cat | https://github.com/matrix-org/matrix-spec-proposals/pull/4084 | 23:07:49 |
@emma:rory.gay | im not going to because im not hostile, but i totally could spam hundreds of messages per second here | 23:07:52 |
uep | sure, the point is it's not about the number, it's about the distributed nature of the problem and the response mechanism | 23:08:10 |
@emma:rory.gay | probably get a few thousand if not tens or hundreds of thousands in before even hitting any kind of spam ban thing you might have set up | 23:08:34 |
Cat | Also Emma just so you know. Bots shouldnt scale that far. Wrong API. | 23:08:53 |
Cat | Appservices yes regular bots fuck no. | 23:09:00 |
Dandellion | sure, you could get a bunch of backup servers and move over, but that is a lot less elegant | 23:09:07 |
@emma:rory.gay | not wrong api | 23:09:16 |
Dandellion | with other-room-membership join rules and knocking one can build protection at ingress points and it would be pretty effective | 23:09:21 |
@emma:rory.gay | appservies use the same API | 23:09:21 |
uep | In reply to @emma:rory.gay im not going to because im not hostile, but i totally could spam hundreds of messages per second here and mjolnir's response to that is a basic rate limit, that lets you configure the number of messages, but not the time interval, which is fixed at a minute. | 23:09:21 |
Cat | You shouldnt be expected to keep up with HS wide Federation line rate on /sync | 23:09:25 |
@emma:rory.gay | yes, which is easily circumvented lol | 23:09:44 |
Cat | Wait thats not configurable in Mjolnir. I keep forgetting how bad things are over in Mjolnir land on that front. | 23:10:08 |
@emma:rory.gay | nor is it in draupnir, wdym | 23:10:21 |