All,
I noticed that a lot of time for us to do our continual NSD patching is
spent writing the text version of the zone file. I had a brief Jabber
chat with Wouter and he explained how this worked.
Basically, nsd-patch produces a new text version of the zone file, and
then zonec compiles that, so the text zone file is quite important. :)
I mentioned this to a colleague, and he suggested that we use a Unix
pipe rather than a temporary file for the patched zone. This would save
a significant amount of disk IO, which is usually quite slow. The
disadvantage is the administrator no longer has a text version of the
zone file to look at.
Looking at the code, it seems fairly straightforward, but I wanted some
input from NSD folks before hacking.
I think the full set of changes would be:
1. Add an option to patch a single zone by "nsd-patch". (This is
necessary because if you are going via a pipe, the zonec program
can only work on a single zone at a time.)
2. Add an option to specify the output file if you are using a
single zone in "nsd-patch" (including treating "-"
as stdout).
3. Change "nsdc" to loop across zones in a configuration file and
patch each one via "nsd-patch $zone -o - | zonec -z -" (or the
equivalent).
Assuming this actually results in a speed increase, I don't know whether
the behavior should be made default or not. I think it should be. After
all, the text zone files do not represent the "actual" state of the
zone
at any given time, and one can always use "dig" to AXFR the zone if
one
needs such a thing.
--
Shane