[nsd-users] FRiCKLE Labs pres. NSD with dynamic zones

[Piotr Sikora]

Hello guys!
For the last couple of weeks I've been working on "dynamic zones" for NSD,
which at the moment means MySQL backend with simple cache (or "zones
dictionary", if you prefer) to avoid "zone matching" via database queries.

--- HOW IT WORKS? ---

Upon receiving query (class IN or ANY) new code checks in cache if there is
matching zone in MySQL database. If it finds one then all records for that
zone are imported from MySQL database, parsed, checked and then whole zone
is injected into regular NSD workflow instead of NSD's database. After reply
is sent, zone is discarded from memory.

"Soft-fail" approach is used for parsing and checking, which means that
error in single record doesn't discard whole zone and as long as SOA record
is imported properly, zone will be injected into NSD.

Cache is updated periodically (at the same time when BIND8 stats are
generated or every 60s if NSD was compiled without this feature).

--- PERFORMANCE ---

Test machine:
Intel Atom 1.6Ghz running OpenBSD 4.5 and MySQL 5.0.77.

--------------------------------
|  daemon   | queries / second |
|  version  | 1child | 2childs |
--------------------------------
| nsd-3.2.1 |  16326 |   3501  |
| nsd-dz    |   1029 |   1148  |
| nsd-dz*   |   3175 |   4066  |
--------------------------------
*MySQL with query cache enabled.

This can be easily improved with local cache (with short expire time) of
already generated zones, but at this time I'm not really sure if it's even
needed.

--- HOW TO INSTALL? ---

$ wget http://www.nlnetlabs.nl/downloads/nsd/nsd-3.2.1.tar.gz
$ wget http://labs.frickle.com/files/nsd-dz-0.2.patch.gz
$ tar zxf nsd-3.2.1.tar.gz
$ gunzip nsd-dz-0.2.patch.gz
$ patch -d nsd-3.2.1 < nsd-dz-0.2.patch
$ ./configure --enable-checking --enable-dz
$ make && make install

--- DATABASE CONFIGURATION ---

CREATE DATABASE nsd;
USE nsd;

CREATE TABLE zones (
  zone_id     INT UNSIGNED AUTO_INCREMENT,
  origin      VARCHAR(255) NOT NULL,
  PRIMARY KEY (zone_id),
  UNIQUE (zone)
) ENGINE = InnoDB;

CREATE TABLE records (
  record_id   INT UNSIGNED AUTO_INCREMENT,
  zone_id     INT UNSIGNED NOT NULL,
  owner       VARCHAR(255) NOT NULL,
  ttl         INT UNSIGNED NOT NULL,
  type        VARCHAR(10) NOT NULL,
  rdata       VARCHAR(255) NOT NULL,
  PRIMARY KEY (record_id),
  FOREIGN KEY (zone_id) REFERENCES zones(zone_id) ON DELETE CASCADE
) ENGINE = InnoDB;

GRANT SELECT ON nsd.* TO 'nsd'@'localhost' IDENTIFIED BY 'nsd';

If you want to use different database / username / password, then please
edit MYSQL_* in dz.h before compilation.

--- THE END ---

At the moment this patch completely ignores zone transfers (but I believe
that in such setup they should be done via MySQL replication anyway) and it
supports only few most popular RR types (A, AAAA, CNAME, DNAME, MX, NS, SOA,
SRV, TXT). This is still work in progress, so I'd really appreciate any
feedback.

Best regards,
Piotr Sikora < piotr.sikora at frickle.com >