Based on my weeks of code (and operating instance) study on lemmy_server, I strongly suspect that peer servers upgrading to 0.18 is causing a swarm of federation activity that lemmy.ml is getting as peers go down and back up.
Based on my weeks of code (and operating instance) study on lemmy_server, I strongly suspect that peer servers upgrading to 0.18 is causing a swarm of federation activity that lemmy.ml is getting as peers go down and back up.
oooooh, probably all those messages that never went through are starting to slog through the system, and a million people worth of backlog are getting slogged through.
Lemmy 0.18 (and earlier) has no ability to save the federation queue, when the server stops it forgets what it had in retry to send. I suspect the problem is more the other servers going down causing lemmy.ml to keep crashing on outbound content it is sending. There are hundreds of small servers out there subscribed to lemmy.ml’s communities.
Seems like a rather bad design, if a long queue is neither saved nor sufficiently isolated to not disturb the actual usage.