[OpenAFS-devel] AFSDB record changes and pre-existing clients

[spacefrogg]

As far as I remember, the client does no caching at all=2E So caching of DN=
S records should be the responsibility of the machine's DNS caching resolve=
r=2E Those normally adhere to TTLs=2E This said, you have to plan for a sof=
t migration when using DNS anyways, as DNS caches interpret TTLs from their=
 last fetch=2E So, the first fetching cache determines the start of request=
s going to your new servers, the last determines the end of requests going =
to the old ones=2E Setting very short TTLs on the old records first (but no=
 shorter than 5 Minutes, different story), helps keeping this window short=
=2E Of course, after shortening the TTLs, wait until the old TTLs have run =
out or not all caches might have seen=2E

Hope this helps=2E