Specifically the database backend. Is the polling for notifications causing more database load? From my personal testing, lemmy.ml has the same performance problems with 0.17.4 that it does with 0.18.1 and I haven’t seen anything in the code changes that are that significant with database.
Lucky yours did delivery inbound to lemmy.ml - it is the outbound queues of the servers with lots of active communities that is still failing badly in 0.18.0
At this point, on important topics like this, I go to the original server to make sure I’m not missing posts or comments, as Lemmy to Lemmy is really faltering.