Does anyone with experience with Elixir/Phoenix in production know how common of a problem this sort of memory leak is? Would it be more common with websocket connections since they are persistent and not automatically build around request/response, so async is not built in?
We struggled with oom errors in prod but that was because some very early code was not honoring some basic “don’t do this”, like don’t do repeated Binary appends to an xml that had a gigantor base-64 encoded pdf embedded (use iolists!!). They seem to have disappeared.
The biggest danger is a legacy of linux’s memory overprovisioning BS. I’m looking forward to a BEAM written in zig with failable allocators everywhere.