22 Srp
2014
22 Srp
'14
12:14
On 22/08/2014 10:59, Jan Včelák wrote:
Hi Jan,
I build up-to-date Knot DNS packages for EPEL 7. The
sources are in Fedora
git, so you can use them as a starting point if you want:
Thank you for this pointer. I hadn't actually noticed that Knot was in
EPEL 7 :)
My first question is: do you intend to keep Knot in EPEL fully up to
date? And how long for? If you are unable to maintain it, who will take
over your task?
I have had a look at your Knot package, and it looks very reasonable to
me. However, it's missing one thing. See below.
Yes, the usage of templates is completely reasonable.
But I'm not sure about
the .target unit. The targets are designed for services starting - I think
that it won't work for other operations (e.g., service restart). Maybe someone
can clarify it, I do not have experience with this systemd feature.
Let me explain this a bit more. Since Knot doesn't have views, and we
need to serve different zones on different addresses, we run multiple
instances of Knot on a server. The way your Knot package in EPEL is
packaged, it won't easily allow us to instantiate multiple copies of
Knot, because it only provides a single .service file, which starts a
single copy of Knot.
Now, what I would like is to have a knot@.service file, which is a
template. In this template, the ExecStart line would be something like:
ExecStart=/usr/sbin/knotd -c /etc/knot/%i
This way, I can drop multiple config files into /etc/knot, and then make
systemd instantiate multiple instances of Knot just by doing:
systemctl enable knot(a)conf1.conf
systemctl enable knot(a)conf2.conf
systemctl enable knot(a)conf3.conf
However, in order to start/stop/restart these, I would have to
individually call systemctl for each instance. There is no easy way to
start/stop/restart all of them in a single call. This is where targets
come in. By defining a knot target in a knot.target file, I can group
knot instances. Then, I can simply do:
systemctl start knot.target (to start all instances)
systemctl stop knot.target (to stop all instances)
systemctl restart.target (to stop all instances)
This setup does not exclude individual instance control. For example, if
I just want to restart a single instance, I can still do:
systemctl restart knot(a)conf1.conf
This setup is flexible enough for both the simple use case of a single
copy of knot, as well as our use case of multiple copies. For the single
copy, there is just a single instance.
I have described all this here for the benefit of other users, and to
get feedback on this idea, but I'm happy to send you some template and
target files for inclusion into the package. And if you commit to
maintaining this package in this form, then I would be happy to stop
building our own RPMs, and use yours.
Knot DNS supports startup notifications (sd_notify),
however we do not have
any feedback on this feature from our users. And there might be some space for
improvements.
I see. I don't have any experience for now either.
We also support structured logging into journal, which
is also a feature we
intend to improve - thoughts are welcome. At the moment, we just append a
field with zone name to the logged messages. Therefore, you can query journal
for messages for one zone by running:
# journalctl _SYSTEMD_UNIT=knot.service ZONE=example.com
Ok, this is useful.