cross-posted from: https://lemmy.g97.top/post/761
cross-posted from: https://lemmy.g97.top/post/723
Hi! I spawned my own instance of lemmy on my server and I discovered new things about how lemmy and federation works, and I have a lot of doubt. I don't know exactly if those doubts are problems of my implementation of if they are normal, so!
- My main account is on lemmy.world and I see that new posts from communities I follow show up before on lemmy.world and then on my instance. Is it normal?
- With comments happens the same thing and they are slower to "sync". Why?
- If a community has been never discovered from the search form with the full format !community@instance, it will never appear on my instance. This means that is not possible to search for an argument (i.e. steam deck) and finding all the posts and communities about it. Is this normal or a feature that we/you would like to see in future/is adaptable to the concept of the fediverse? Because if I am on a big instance with a lot of users maybe I found that specific community or post, but on smaller instances like mine it will never appear If I don't know the exact name.
- I created a community on my instance and subscribed it from lemmy.world but I don't see any post nor are they in sync. Why? https://lemmy.world/c/announcements@lemmy.g97.top vs https://lemmy.g97.top/c/announcements.
- From my instance I am unable to follow lemmy.ml communities (they are pending, usually on lemmy.world the pending status is faster)
- I am unable to search for communities on Kbin.social, and when I try I see this log message of type "couldnt_find_object: error decoding response body: missing field
propertiesat line 1 column 206" from my docker instance:
2023-06-20T22:02:16.056226139Z 2023-06-20T22:02:16.055937Z ERROR HTTP request{http.method=GET http.scheme="https" http.host=lemmy.g97.top http.target=/api/v3/ws otel.kind="server" request_id=8211e6a4-2b30-4f8c-98b3-d93843a0e293 http.status_code=101 otel.status_code="OK"}: lemmy_server::api_routes_websocket: couldnt_find_object: error decoding response body: missing fieldpropertiesat line 1 column 206 2023-06-20T22:02:16.056276976Z 0: lemmy_apub::fetcher::search::search_query_to_object_id 2023-06-20T22:02:16.056286500Z at crates/apub/src/fetcher/search.rs:17 2023-06-20T22:02:16.056293804Z 1: lemmy_apub::api::resolve_object::perform 2023-06-20T22:02:16.056300316Z with self=ResolveObject { q: "[!leagueoflinux@kbin.social](/c/leagueoflinux@kbin.social)", auth: Some(Sensitive) } 2023-06-20T22:02:16.056307712Z at crates/apub/src/api/resolve_object.rs:21 2023-06-20T22:02:16.056314152Z 2: lemmy_server::root_span_builder::HTTP request 2023-06-20T22:02:16.056320693Z with http.method=GET http.scheme="https" http.host=lemmy.g97.top http.target=/api/v3/ws otel.kind="server" request_id=8211e6a4-2b30-4f8c-98b3-d93843a0e293 http.status_code=101 otel.status_code="OK" 2023-06-20T22:02:16.056351870Z at src/root_span_builder.rs:16
- I have a lot of warnings in the lemmy log of type "Error encountered while processing the incoming HTTP request: lemmy_server::root_span_builder: Header is expired" such as:
2023-06-20T21:58:12.484449111Z 2023-06-20T21:58:12.484275Z WARN Error encountered while processing the incoming HTTP request: lemmy_server::root_span_builder: Header is expired 2023-06-20T21:58:12.484510012Z 0: lemmy_server::root_span_builder::HTTP request 2023-06-20T21:58:12.484517559Z with http.method=POST http.scheme="https" http.host=lemmy.g97.top http.target=/inbox otel.kind="server" request_id=caf194c5-cac3-4c37-a29c-577d65deb050 http.status_code=400 otel.status_code="OK" 2023-06-20T21:58:12.484525578Z at src/root_span_builder.rs:16 2023-06-20T21:58:12.484530286Z LemmyError { message: None, inner: Header is expired, context: "SpanTrace" }I have more questions/doubt but for now this is enough I think! Thank you!
Do you happen to know how the server-to-server connections are managed? I'm not too familiar with it, but it seems like HTTP/3 might provide some benefits for server-to-server communication.
Also, regarding queuing federation messages, I'm curious if packages like Kafka or Pulsar have been considered? They aren't typically used over HTTP, but it doesn't seem like it would be too hard to adapt, and the stream retention policy could be set to allow consumers to pick up older records as they have capacity (to avoid the issue around servers getting out of sync. The consumer would know the queue offset for each stream it was consuming and could pick up records as it has capacity, provided it doesn't fall so far behind that the records expire). Publishers could provide separate topics for different message types to allow consumers to prioritize activity types (for example, prioritizing receiving replies over up/down votes). Also servers could potentially use cluster replication (Mirror Maker) to handle moving activity records from one server to another (again, HTTP-only would be an issue here), and each server could then consume the federation activity messages locally from its own queue.
Kafka/Pulsar support have strong scaling support, so adding capacity for federation messages should be fairly straightforward.
I've only used Kafka once, and I'm completely unqualified to operate an instance of any complexity, by in general my experience with it was pretty good.