Both kind of. I have a relay running on a server with 2GB of mem, added a metadata updater service to my relay’s docker compose stack, which I know is very mem intensive as it scans all relays and does some calculations. 

Updater service memory climbs to about 6-700 MB under peak load but was hanging onto all that. As soon as I up the memory to 8GB automatic garbage collection kicks in and keeps memory below 200 after runs. 

https://github.com/UTXOnly/metadata_updater