[knot-dns-users] Knot DNS 1.5 technology preview

Hi Everyone, we're getting closer to the 1.5 release and because we're mostly feature complete now, I think today might be a good opportunity to share some of the new stuff we've created and hear out your opinions. There's a ton of improvements that I can't cover in a few paragraphs, but I've picked 3 of those that I think are the pillars for the bigger plan for this year. Just so you know, there is no tarball this time, but if you're brave enough, you can actually try everything covered here. You can find the code in our git repository under a tag "v1.5.0-alpha": $ git clone -b v1.5.0-alpha https://gitlab.labs.nic.cz/labs/knot.git 1. Query processing and code base We've known for a while that there were parts of the code that we weren't happy with. It's not just that it was getting hard to patch and read, but with we had a ton of ideas that we couldn't get implemented because of the limitations (in fact, this effort bears fruits as soon as of this release, but more on that later). One of those places was query processing and dealing with queries in general. Naturally, we've sat down and reworked this from scratch. Now, from the user perspective it's not so exciting - yes we made it RFC compliant in several cases and it's faster than ever (I'm running benchmarks as we speak, so you might want to check out our webpage later). It's still a preview and we're ironing things out, so don't take the numbers as final. For those who are interested in the details (thank you!), it's much much more important because the code lays a solid foundation for the cool stuff to come while being lightweight. How much? Around 10+k LOC which is roughly 10-15% of our code base gone and we offset that by adding almost 5k LOC worth of test cases. To put it in another words, it means it does more with less and we're not done here. 2. Separating Knot DNS library We also improved the Knot DNS build process overall. The code providing basic building blocks for the server and utilities is now held in the libknot library, which is by default linked dynamically to these components. You will also notice the libzscanner library, our high-performance zone scanner written in Ragel (with no dependencies on Knot DNS). We plan to build some other projects on top of that and provide public headers in the future (which is more likely to happen for the libzscanner - if someone is interested). But currently, you may benefit from a smaller size of the binaries. 3. Pluggable query processing modules + auto forward/reverse records New features are always a compromise - usually between ease of use, complexity and resource usage. But what if we could make the core as lightweight as possible and let the user choose? The new processing API is built like a Lego (don't step on it though), it breaks down the processing into independent parts that together form a plan for query processing. So what we did was, we've created an interface for loadable modules and let it add/remove or change the query processing plan as the module sees fit. This allows to keep complexity at bay (for those who don't want extra stuff) and provide a playground for new ideas. We actually have a first module - it can synthesize per-answer forward/reverse records for IPv4/IPv6 addresses from a template in configuration file (see 4.11 in the documentation on further information). For the impatient, an example speaks the best: We have an IPv6 reverse zone we need to take care of, but the problem is the address space is too big to generate a zone file. So we create an empty zone, put in static records we want and let the module to synthesize other records per-answer (if they fit in the template and aren't present in the zone): # Configuration: 1.6.b.0.0.0.0.0.0.2.6.2.ip6.arpa { query_module { synth_record "reverse dynamic- example. 400 2620:0:b61::/52"; } Now if we load Knot DNS, and kdig for a reverse record, we get a following reply: $ kdig PTR 1.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.1.6.b.0.0.0.0.0.0.2.6.2.ip6.arpa. ;; QUESTION SECTION: ;; 1.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.1.6.b.0.0.0.0.0.0.2.6.2.ip6.arpa. 0 IN PTR ;; ANSWER SECTION: ..snip.. 400 IN PTR dynamic-2620-0000-0b61-0000-0000-0000-0000-0001.example. Forward records can be dealt with in a similar fashion, see the manual 4.11 for more examples and information. To sum it up, modules could be a way to accommodate for the "should this be a in a nameserver?" sort of things. If you don't like it, just don't put it in the configuration and you're good. As of now the module API is open (no "how to" though), and so are we for ideas. Phew, this was longer than I expected and I barely scratched the surface. As you see, this version is going to be a lot about tidying up and opening up the possibilities. So what's your take on this? Best, Marek -- Marek Vavrusa, Knot DNS CZ.NIC Labs http://www.knot-dns.cz ------------------------------------------- Americká 23, 120 00 Praha 2, Czech Republic WWW: http://labs.nic.cz http://www.nic.cz