Warr1024

I'm not seeing consistently higher CPU consumption during problems (though it's hard to tell, when MT is usually hovering around 100% no matter what) so it's actually not entirely inconsistent...

I did some significant database optimizations, and it seemed to be good again for a while. Then, just a few minutes ago, my players reported that it had gone haywire...

34% Lua time wouldn't account for step time going up 400% to 2000%. Sounds like it's not caused by added compute workload, but some kind of problem with a wait.

> And if you use dtime (...) to determine how much work you do in the current step, you can easily make lag spikes persist longer Yeah, I've done a...

I did a bunch of database maintenance stuff, like vacuuming and such, and it improved things to some extent. When the server is working well, it's working quite well (possibly...

It looks like I've got the issue mitigated enough on my server that it's effectively not a problem there anymore. The fact that there's a performance difference between 5.8 and...

After a few days of quiet, I caught another meltdown. It seems like the meltdown condition is separate from the overall IO-bound issue. I also watched iotop for a while...

Well, after tracking down a lot of little problems that had a small impact, I did find one huge issue.  That one long orange "C" part there is in...

> if you have at least one registered callback. is that the case? Indeed I do. In fact it was for a mechanic of this game that I needed the...

Confirmed that it's NOT from vectors being passed to me via register_on_liquid_transformed: I still get that huge vector constructor thing in my flame graph even when that callback is getting...