At HEAD, I think they should ~never need restarts after repository edits, since we'll now reload the repository in bin/repository update. This only landed like last week though.

The major one I am aware of is when configuration changes. This seems hard to do surgically (e.g., we'd basically just have to say "restart the daemons" after every config change, since there are probably like 150 options that can affect them). But maybe we should.

A simple way to do this would maybe be redirect to "?you-edited-something=true" after a config edit, and throw a "Hey, remember to restart the daemons!" banner on that page?

We could try to get the daemons to refresh configuration, but there are limits to their ability to react to configuration changes without phd restart (for example, they can't change their pidfile location midstream).

But maybe we should.

I don't know what that was about, I don't favor trying to separate daemon-affecting config from non-daemon-affecting config.

Overall, the "hey, remember to restart the daemons" banner seems most reasonable to me here.

A maybe cleaner approach to this is:

• Add a function to hash all of the config values.
• When starting a daemon, insert the config hash into the daemon log table (this requires a new column).
• Add a setup check to compute the current web config hash and compare it to the daemon config hashes.
• If there are running daemons with a bad hash, raise a config warning: "you should restart the daemons, their configuration is out of date..."