I believe problem 3 (MultiXact SLRU buffer cache lock contention) can occur even if you don't nest transactions in your code. We traced the problem back to fore

I believe problem 3 (MultiXact SLRU buffer cache lock contention) can occur even if you don't nest transactions in your code. We traced the problem back to foreign keys, when rows are updated Postgres internally does SELECT .. FOR KEY SHARE on referenced tables. This requires MultiXacts. I feel that we miss benchmarks and observability tools for the low level locks which are essentially single threaded, if you have lock contention on them the system will grind to a complete halt. If you can recompile, raising the size of this buffer cache can help.