Comment 10 for bug 1899902

Using a long zone file and limiting the resources of the system I was using to test these changes (just limiting lxc CPU cycles) I found a set up where I can constantly hit the issue.

As Andreas pointed out, we are seeing a race condition here.

After forking, bind9 signals that it is ready by returning on the parent process. This leads systemd to mark the service as ready. Finally nginx starts running.

After bind9's parent process exits, bind9 reloads the response policy zones, creating a small gap where a name in the rpz would not be resolved. If the nginx name checks happen during that gap, nginx will fail to start.

I will change the tests to just add a record to the fake name and then we should have a sound test.

Thanks, Andreas and Dan :)