jjj333-p
commented
9 months ago changed the name of this because i think its actually a really cool idea to try to catch and report on some of the errors but i dont think it should be just federation errors (albeit those are the most common)
i honestly wonder if it could (should) even automatically try to run a download state when it spots a state reset (something that actually shows up in logs before those auth errors if you search for it but its easy to miss), but i think definitely whenever it sees something like that it should report on it
as for your last bit, honestly incoming federation is easier to catch because its usually accompanied with something similar to dendrite issue #3202 with a "bad signature" error and usually some back-filling. I'm not entirely sure how to go about detecting outgoing federation issues as i havent experienced them myself, the thing ive more experienced is it not receiving events from any clients until things are rebooted and that doesnt produce errors, it just has the m.typing edu events and never logs about an event sent to roomserver. of course my idea is for unexpected errors it just posts that it saw the error so that would i suppose take care of anything infrequent that does produce an error.
anton-molyboha
Well, these are two requests in one, but they have a similar motivation: in both cases the symptom is that you don't see some of the messages in a room/rooms. Unless the affected counterparty is matrix.org, it is easy to not even notice the problem. The bot could monitor dendrite logs, and warn the admin if something looks off.
In my experience, state desync manifests itself with a large number of "Error authing soft-failed event" messages in the dendrite log. The bot could monitor the log, and warn the admin if it sees such error repeatedly for the same room_id.
I'm not sure how to catch broken federation. Outgoing federation problems will probably show up in the log as well. Incoming federation -- in the worst case -- can be caught statistically: if we are used to see X messages per day from a remote HS, and suddenly that number drops to zero, we probably have a problem.