Hi guys, replying to all, asnwering Chris and Cy emails that I did not reply earlier. I trimmed quoting, but of course I've read your emails!
Point 1. Please let's forget about 'late' option and any other rc(8) hints and magic. As I already explained the problem can (and usually does) live outside of the particular host that does the mount. It could be a boot race of a bunch of networking equipment, it could be some other network outage, etc. Point 2. Both Chris and Cy said that this is not a bug, since it was there for so many years. Sorry, this argument doesn't buys me. It is a typical cognitive distortion named "normalization" or "desensitization," where an individual becomes so accustomed to a negative situation that they no longer recognize it as problematic. I am also affected by that, and it is very good practice sometimes to force yourself to look at something with a fresh look. With a fresh look a suggestion to hardcode IP addresses in hosts(5) or doesn't look scalable neither modern option. I totally agree that in certain setups it is the right way to do, but not always. Point 3. I got our concern on a mount_nfs(8) blocking on a DNS resolution viewed as a POLA violation. So I suggest a trade-off: let's isolate the retrying behavior only to the background mode. That would fix the problem of machine booting without mounts and is very unlikely to affect anyones POLA feelings: https://reviews.freebsd.org/D49145 As the review text notes at the end we got a problem in the libc getaddrinfo(3). Rick also noticed it earlier when making his patch. Our resolver can't tell us a negative answer versus a timeour. This definitely is a problem and I already started investigating it. But definitely out of scope of NFS. -- Gleb Smirnoff
