Docs we will need

[Richard Laager]

On 2/4/19 12:07 PM, Hal Murray via devel wrote:
> Another complication with getting started after a building/site wide power 
> loss is that getting time needs DNS and the local caching DNS server may be 
> waiting for valid time.

The resolver really shouldn't be waiting for network time, as doing so
will create these chicken-and-egg problems.

The only thing in DNS that depends on time is validating DNSSEC
expirations. If you don't have a RTC, one possible answer is to turn
that off--either all DNSSEC validation or just ignoring expirations.
BIND, for example, has options for both.

If your system doesn't have an RTC, you may have to take various extra
steps, depending on your configuration. I don't think we need to solve
all of these problems for people. Even if you want to cover the common
case of a Raspberry Pi running Raspbian, I don't think it's necessary to
cover every hypothetical case where someone has turned on every possible
security feature and has no RTC.

And while you might be able to maintain some documentation on the topic,
you _can't_ solve some of these problems in ntpd without horrible
abstraction violations. Keeping an application-level DNS cache, as you
proposed, is something I would argue is a horrible abstraction
violation. As an admin, I want to be able to reason about DNS
independently from reasoning about ntpd. If ntpd is doing non-standard
things behind my back, that is some combination of annoying and scary.

-- 
Richard