space-wizards
Integration tests are very leaky
Not sure when this was introduced, but we're running into this downstream as it appears to be hitting the Github Actions OOM killer of ≈ 10 GB so presumably Content isn't far behind from hitting the same problem. I'm not smart enough to diagnose the problem, but at the least it doesn't seem to be from the last engine update. It's also possible this is a long-standing issue that was just never noticed since memory usage in tests is relatively low priority.
Here's what I get when profiling space-wizards/space-station-14@1b43f6efd41188c88a4744c87024943224b811c2:
It seems like it's almost all entity prototype dictionaries on the client. (I think.)
You have dotTrace open, get it to show a retention path for the relevant data.
I think I got red herring'd as actually comparing snapshots shows these are the sources of the new bytes:
The B2 trees are retained from:
It also looks like what I'm able to profile myself on macOS is somewhat limited as it seems only profiling memory via process attach for unit tests seems to work, so I can't use full allocation mode or get some of the call tree. (This is apparently skipping 300 MB of objects according to an out of screenshot warning.)
I'm not sure if this is actually relevant, but I don't really understand how sampled memory allocation profiling works so I'm including it anyway. Presumably EntityTest is only a contributing factor regardless since disabling it shows a similar ramp of memory usage, which makes sense.