[chrony-dev] forced name reresolving by default?

[Miroslav Lichvar <mlichvar@xxxxxxxxxx>]

I'm wondering if the --enable-forcednsretry option should be enabled
by default.

I've noticed that on Fedora 15 the getaddrinfo function doesn't return
EAI_AGAIN even with empty /etc/resolv.conf, not sure if that's how it
is supposed to be or if it's another glibc bug. I think it used to
return that code, but there were race conditions with services that
start networking and write to /etc/resolv.conf, so I had to use the
forcednsretry option anyway.

Is there a system where we can rely on the EAI_AGAIN code in all
situations including boot w/o networking, etc? Is that even possible?

The disadvantage of enabling the option is that if someone specifies
an invalid name in the config, chronyd will be trying forever to
resolve the name and that will block the daemon for shorts periods of
time as we don't have an asynchronous resolving yet. (timestamps for
incoming NTP packets are taken from the IP_PKTINFO message so the
clients will see only large time spent in handling of the requests,
the network delay shouldn't be affected by this)

IIRC ntpd used to have a similar option, but current versions seem to
do it by default, with some special handling of EAI_NONAME and
EAI_NODATA if EAI_AGAIN was seen before.

What do you think?

-- 
Miroslav Lichvar

---
To unsubscribe email chrony-dev-request@xxxxxxxxxxxxxxxxxxxx with "unsubscribe" in the subject.
For help email chrony-dev-request@xxxxxxxxxxxxxxxxxxxx with "help" in the subject.
Trouble?  Email listmaster@xxxxxxxxxxxxxxxxxxxx.