That's exactly my experience, as well.
First I've put it off as "my internet connection being unreliable".
But, since almost everything works fine, but mock builds continue to fail
(mostly because of "Failed to synchronize cache for repo 'foo'", where
foo is either fedora, updates, or rawhide),
I now think this really is some infra problem.
Also, I think some of the problems that happen in COPR are caused by
its very burst-y behavior:
https://copr.fedorainfracloud.org/status/stats/
For example, today, some of my builds were pending for over 7 hours,
during which time almost no running build completed, but pending tasks
blew up from the average 0-10 to over 140, since no pending tasks
moved to running at all, during that time, from what I can tell.
Then, a lot of builds (about 35) were moved to the "running" state at
once, and some of them failed with connectivity issues, again - which
doesn't surprise me, when there are possibly hundreds of mock builds
triggered by COPR at about the same time, obviously overwhelming the
system serving the repositories.
Well, that was caused by copr backend becoming unresponsive.
So, no jobs were processed until a copr maintainer was able to reboot it.
The repo issues seem to be our mirrorlist containers sometimes throwing
a 503 when they shouldn't. ;( This is made worse by dnf never retrying
them, so if it happens once thats it. We are investigating and trying to
fix whatever is causing this.
kevin