Possible regression: hackney doesn't seem to respect pools
After some reports about larger RAM requirements in 2.1 I checked on lain.com, and sure enough, sometimes the BEAM process would use 1gb+, which seems to be too much (the normal amount is <300mb). The process isn't always this large, and usually shrinks down after a while.
Turns out that it seems to be a problem with hackney pools: When I do anything federation related, ~600 connections are opened immediately and nearly fill up my ram. By doing a few things (faving a few posts), i can crash my instance (with 2gb ram).
Switching to gun instead of hackney did fix this behavior, and ram usage stayed at the normal level throughout all federation.
It seems that somehow the pools that we set do not work for hackney. I'm not too familiar with the code and couldn't find any obvious issue.