It's nice if the clock is more-or-less right before services start.
bind/named resolver doesn't resolve if the clock is horribly out
(DNSSEC?) and it doesn't fix itself after a systemctl restart chronyd,
you need to actually bounce named. Without named things like
'ntp.pool.fedora' don't work, so there's a catch-22.
postgresql can be unhappy on boot. This is clock related, but I can't
replicate it 100% yet.
journald will rotate logs as soon as chronyd syncs the clock. So any
logs from bootup (1970) are lost. That's another reason to get the
clock +- correct as soon as possible.
I'm currently using a cron script to touch a file every 10 minutes, and
read that on bootup (before chronyd), and I've added a
'Requires=touchClock' to some systemd services.
The script is smart enough to refuse to run if it'll move the time