samba_dnsupdate: make rodc_dns_update() more robust against timing problems

samba_dnsupdate: make rodc_dns_update() more robust against timing problems

Without this we had an interesting race!

The messaging_dgm code caches connected datagram sockets based on the
destination pid for 1 second.

The fact that samba_dnsupdate constantly recreates its messaging
context (and the underlying datagram socket) means that we the winbindd
messaging context may get a stale connection. As a result sending any
message from winbindd back to samba_dnsupdate will result in
ECONNREFUSED.

That means the IRPC response from winbindd never reaches
samba_dnsupdate, which will then hit a timeout.

In turn samba_dnsupdate on the RODC times out.

This was a workaround for the problem, by having just one global
IRPC handle and thus just one messaging_dgm context.
The actual problem is solved a few commits before
("messages_dgm: Properly handle receiver re-initialization").
But we keep this as an performance optimization, which hopefully
means that the overall samba_dnsupdate is less likely to
timeout after the hardcoded 20 seconds.

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>