I caught a ~1.5GB aphlict on secure.phabricator.com after about 24 hours / 100K connections.
We currently have a ~5.4GB aphlict on notify001 after about 16 hours.
On notify001 this isn't really an issue since the thing just restarts after exploding, I think. However, this can create memory pressure on hosts like secure001 where the process can't have the whole server.
I'm pretty sure this didn't previously happen (I think I recall months of uptime for aphlict last year?) so there may have been, e.g., a node update which changed something we depend on. Whatever the issue is is likely a problem on our side, but it may not have been exposed before.