NFSv4 server is FreeBSD-13.1, problem occurs at least on Linux and Solaris clients.

The seems to be a small time window triggered by reloading /etc/exports (using SIGHUP @mountd, or 'service mountd reload') causing valid client requests getting EPERM. We can reliably reproduce the glitch by running this little Perl script on a client:

#!/usr/bin/perl

while (1) {
    open my $fh, $0 or warn($!);
    sysread $fh, my $s, 64, 0 or warn($!);
    close $fh
}

and SIGHUPing the mountd on the FreeBSD server. Many entries in /etc/exports help hitting the critical time window.

For our use case this problem has show-stopper qualities since we manage export ACLs on a per-volume basis, and we have lots of volumes.