A single authoritative DNS server is a single point of failure for every player trying to resolve the game lobby hostname. If the node’s resolver process dies or the machine is rebooting after a kernel update, clients get SERVFAIL and the launcher shows a connection error that looks like an outage. Running two nameservers is not optional for anything public, and NSD is a good fit because it is small, memory-safe and answers purely from zone files.
The interesting part is keeping the secondary in sync. NSD does not do AXFR by default for a primary it serves, and the transfer must be gated on the right address, authenticated where possible, and monitored so the secondary’s serial expires before anyone notices. This is the whole configuration.
Installing and defining the primary zone
Install NSD from the distribution package, then create a zone file under /etc/nsd/zones/. The only fields that must be exactly right for transfer are the SOA serial, the MNAME, and the NS records for both servers. Keep the serial in a numeric form you increment deliberately rather than a date-based value you might repeat.
$TTL 3600
@ IN SOA ns1.example.net. hostmaster.example.net. (
2026110401 ; serial
7200 ; refresh
900 ; retry
1209600 ; expire
300 ) ; minimum
@ IN NS ns1.example.net.
@ IN NS ns2.example.net.
@ IN A 203.0.113.10
ns1 IN A 203.0.113.10
ns2 IN A 198.51.100.20
lobby IN A 203.0.113.10
world IN A 203.0.113.11
api IN A 203.0.113.12
nsd.conf for the primary
The primary needs to allow transfers from the secondary’s exact address only. Use a network prefix if you run several secondaries in the same range, and never open transfers to the world — an open AXFR is a free zone dump for anyone who asks. The provide-xfr option takes a source prefix and an optional TSIG key name.
server:
ip-address: 203.0.113.10
port: 53
hide-version: yes
verbosity: 1
remote-control:
control-enable: yes
key:
name: xfr-key
algorithm: hmac-sha256
secret: "BASE64SECRETREPLACEME"
zone:
name: example.net
zonefile: "/etc/nsd/zones/example.net.zone"
provide-xfr: 198.51.100.20 xfr-key
notify: 198.51.100.20 xfr-key
Configuring the secondary
The secondary does not need a zone file at all; NSD writes its own copy, stores it in the configured database directory and treats it as read-only. The critical settings are the primary address and the request-xfr line, which must name the same key. If the key name mismatches, the secondary silently retries every retry interval and logs a reset.
zone:
name: example.net
zonefile: "/var/lib/nsd/example.net.zone"
allow-notify: 203.0.113.10 xfr-key
request-xfr: AXFR 203.0.113.10 xfr-key
acl:
name: allow-transfer
action: deny
nokey: xfr-key
Reload with nsd-control reload and check nsd-checkzone output first. A syntax error loads on the primary and fails on the secondary, leaving two servers with different answers — the worst state during an incident.
Timers that decide how long a stale zone survives
The SOA timer set is the real availability design. Refresh is how often the secondary polls; retry is how soon it tries again after a failure; expire is how long it keeps serving data after it can no longer reach the primary. A zone with a 12-hour expire will serve stale records for half a day after the primary dies, which is usually too long for a game service with rotating endpoints.
| Timer | Small zone (game lobby) | Large or change-heavy zone | What it controls |
|---|---|---|---|
| Refresh | 600 s | 1800 s | Poll frequency |
| Retry | 120 s | 300 s | Backoff after failure |
| Expire | 86400 s | 604800 s | Stale serving window |
| Minimum | 300 s | 3600 s | Negative caching |
Verifying that transfer actually happened
Do not assume. Ask the secondary directly, bypassing the recursive resolver on your workstation, and compare the serial it answers with the primary. Both should return the same SOA serial for the zone, and the answer must come from the secondary’s own address.
dig @203.0.113.10 example.net SOA +short
dig @198.51.100.20 example.net SOA +short
dig @198.51.100.20 lobby.example.net A +short
nsd-control zone-status
A mismatch means the secondary has not transferred. Check nsd-control zone-status for the last transfer time, then verify the key material and the source address seen by the primary. Address translation in front of the secondary is the most common cause of a silent failure — the primary sees a NAT address and rejects the transfer.
Publishing both servers at the registrar
Both NS records must be in the delegation at the registrar, and both addresses should be reachable from the public internet on UDP and TCP 53. A secondary in the same rack as the primary survives a process crash but not a power event, so place it in a different facility. If you keep everything on one provider, at minimum use a different subnet and a different power feed.
Serve the game infrastructure on dedicated nodes in separate racks and the DNS secondaries follow naturally from that topology. One node answering DNS in-region and one out-of-region also gives you a crude latency benefit for large player populations.
Monitoring, NOTIFY and rollback
Alert when the serial difference between primary and secondary exceeds zero for more than 15 minutes, and alert when nsd-control zone-status reports an expired zone. NOTIFY makes propagation near-instant, but it is UDP and unacknowledged, so the refresh timer is your safety net. Test NOTIFY by bumping the serial and watching the secondary log the transfer within the refresh window.
Keep the previous zone file version on both hosts and know the rollback command. If a bad change publishes, restoring the prior file and incrementing the serial is faster than debugging with players offline. See the secondaries in a separate facility for the topology that survives a rack failure.

Leave a Reply
You must be logged in to post a comment.