This is the DNS server powering the NTP Pool system and other similar services. It supersedes the pgeodns server.
If you already have go installed, just run go get
to install the Go dependencies.
You will also need the GeoIP C library, on RedHat derived systems
that's yum install geoip-devel
.
If you don't have Go installed the easiest way to build geodns from source is to
download Go from http://code.google.com/p/go/downloads/list and untar'ing it in
/usr/local/go
and then run the following from a regular user account:
export PATH=$PATH:/usr/local/go/bin
export GOPATH=~/go
go get github.com/abh/geodns
cd ~/go/src/github.com/abh/geodns
go test
go build
./geodns -h
There's a sample configuration file in dns/example.com.json
. This is currently
derived from the test.example.com
data used for unit tests and not an example
of a "best practices" configuration.
For testing there's also a bigger test file at:
mkdir -p dns
curl -o dns/test.ntppool.org.json http://tmp.askask.com/2012/08/dns/ntppool.org.json.big
After building the server you can run it with:
./geodns -log -interface 127.1 -port 5053
To test the responses run
dig -t a test.example.com @127.1 -p 5053
The binary can be moved to /usr/local/bin, /opt/geodns/ or wherever you find appropriate.
Notable command line parameters (and their defaults)
- -config="./dns/"
Directory of zone files (and configuration named geodns.conf
).
- -checkconfig=false
Check configuration file, parse zone files and exit
- -interface="*"
Comma separated IPs to listen on for DNS requests.
- -port="53"
Port number for DNS requests (UDP and TCP)
- -http=":8053"
Listen address for HTTP interface. Specify as 127.0.0.1:8053
to only listen on
localhost.
- -identifier=""
Identifier for this instance (hostname, pop name or similar).
It can also be a comma separated list of identifiers where the first is the "server id" and subsequent ones are "group names", for example region of the server, name of anycast cluster the server is part of, etc. This is used in (future) reporting/statistics features.
- -log=false
Enable to get lots of extra logging, only useful for testing and debugging. Absolutely not recommended in production unless you get very few queries (less than 1-200/second).
- -cpus=1
Maximum number of CPUs to use. Set to 0 to match the number of CPUs available on the system. Only "1" (the default) has been extensively tested.
geodns runs a WebSocket server on port 8053 that outputs various performance
metrics. The WebSocket URL is /monitor
. There's a "companion program" that can
use this across a cluster to show aggregate statistics, email for more information.
There's a page with various runtime information (queries per second, queries and
most frequently requested labels per zone, etc) at /status
.
GeoDNS can post runtime data to StatHat. (Documentation)
Most records can have a 'weight' assigned. If any records of a particular type
for a particular name have a weight, the system will return max_hosts
records
(default 2).
If the weight for all records is 0, all matching records will be returned. The weight for a label can be any integer as long as the weights for a label and record type is less than 2 billion.
As an example, if you configure
10.0.0.1, weight 10
10.0.0.2, weight 20
10.0.0.3, weight 30
10.0.0.4, weight 40
with max_hosts
2 then .4 will be returned about 4 times more often than .1.
In the configuration file the whole zone is a big hash (associative array). At the top level you can (optionally) set some options with the keys serial, ttl and max_hosts.
The actual zone data (dns records) is in a hash under the key "data". The keys in the hash are hostnames and the value for each hostname is yet another hash where the keys are record types (lowercase) and the values an array of records.
For example to setup an MX record at the zone apex and then have a different A record for users in Europe than anywhere else, use:
{
"serial": 1,
"data": {
"": {
"ns": [ "ns.example.net", "ns2.example.net" ],
"txt": "Example zone",
"spf": [ { "spf": "v=spf1 ~all", "weight": 1 } ],
"mx": { "mx": "mail.example.com", "preference": 10 }
},
"mail": { "a": [ ["192.168.0.1", 100], ["192.168.10.1", 50] ] },
"mail.europe": { "a": [ ["192.168.255.1", 0] ] },
"smtp": { "alias": "mail" }
}
}
The configuration files are automatically reloaded when they're updated. If a file can't be read (invalid JSON, for example) the previous configuration for that zone will be kept.
Each label has a hash (object/associative array) of record data, the keys are the type. The supported types and their options are listed below.
Adding support for more record types is relatively straight forward, please open a ticket in the issue tracker with what you are missing.
Each record has the format of a short array with the first element being the IP address and the second the weight.
[ [ "192.168.0.1", 10], ["192.168.2.1", 5] ]
See above for how the weights work.
Same format as A records (except the record type is "aaaa").
Internally resolved cname, of sorts. Only works internally in a zone.
"foo"
"target.example.com."
"www"
The target will have the current zone name appended if it's not a FQDN (since v2.2.0).
MX records support a weight
similar to A records to indicate how often the particular
record should be returned.
The preference
is the MX record preference returned to the client.
{ "mx": "foo.example.com" }
{ "mx": "foo.example.com", "weight": 100 }
{ "mx": "foo.example.com", "weight": 100, "preference": 10 }
weight
and preference
are optional.
NS records for the label, use it on the top level empty label (""
) to specify
the nameservers for the domain.
[ "ns1.example.com", "ns2.example.com" ]
There's an alternate legacy syntax that has space for glue records (IPv4 addresses), but in GeoDNS the values in the object are ignored so the list syntax above is recommended.
{ "ns1.example.net.": null, "ns2.example.net.": null }
Simple syntax
"Some text"
Or with weights
{ "txt": "Some text", "weight": 10 }
An SPF record is semantically identical to a TXT record with the exception that the label is set to 'spf'. An example of an spf record with weights:
{ "spf": "v=spf1 ~all]", "weight": 1 }
An spf record is typically at the root of a zone, and a label can have an array of SPF records, e.g
"spf": [ { "spf": "v=spf1 ~all", "weight": 1 } , "spf": "v=spf1 10.0.0.1", "weight": 100]
This software is Copyright 2012-2013 Ask Bjørn Hansen. For licensing information please see the file called LICENSE.