Skip to main content

Resolve devices with DNS

Out of the box you address simulated devices by their 10.42.x.x management IPs. With the CoreDNS sidecar you can resolve them by name instead, with forward <device-name>.nl6.local records and reverse PTRs. The records update automatically as devices come and go.

nl6 runs an authoritative DNS server as a hidden primary; a stock CoreDNS runs as a secondary, transferring the zones via AXFR and refreshing on NOTIFY. No custom CoreDNS plugin.

 dig @localhost  ──►  CoreDNS :53 (secondary)  ──AXFR/NOTIFY──►  nl6 :5353 (primary)
                                                                  derived from the
                                                                  live device set
  • Forward <device-name>.nl6.local → device management IP
  • Reverse <ip> (PTR) → ip4.mgmt.<device-name>.nl6.local, which resolves forward to the same IP (round-trips)

Prerequisites​

  • Docker with Compose v2 (docker compose). The sidecar stack runs nl6 in a container, so you don't need host TUN/netns setup. See Docker.
  • dig (or any DNS client) and curl on the host to verify.

Run the stack​

The repo ships a ready-to-run stack at examples/coredns-sidecar/:

git clone https://github.com/labmonkeys-space/nl6
cd nl6/examples/coredns-sidecar
docker compose up -d

It auto-starts a 10-device demo fleet (10.42.0.1 to 10.42.0.10) with the DNS subsystem enabled. On every device create/delete nl6 bumps the zone serial (debounced ~1s) and NOTIFYs CoreDNS, which re-transfers.

The nl6 service runs with these DNS flags (see the CLI flags for the full set):

command:
  - -dns-enable
  - -dns-domain=nl6.local
  - -dns-reverse-zone=42.10.in-addr.arpa
  - -dns-notify=coredns:53

…and the CoreDNS Corefile is a stock secondary block per zone:

nl6.local:53 {
    secondary {
        transfer from nl6:5353
    }
    log
    errors
}
42.10.in-addr.arpa:53 {
    secondary {
        transfer from nl6:5353
    }
    log
    errors
}

Verify name resolution​

# List the running fleet to pick a real name/IP (names are random per run):
curl -s localhost:8080/api/v1/devices | jq -r '.data[].ip'
dig @localhost -p 5353 nl6.local AXFR | grep ' A '   # the forward names + IPs

# Forward + reverse against CoreDNS (host :53):
dig @localhost core-rtr-01.nl6.local +short          # device name -> 10.42.0.x
dig @localhost -x 10.42.0.5 +short                   # 10.42.0.5  -> ip4.mgmt.<name>.nl6.local.

# Publish + NOTIFY counters:
curl -s localhost:8080/api/v1/dns/status | jq

Names are synthesised. nl6 generates each device's sysName randomly, so the exact forward names change per run. Read them from the AXFR (dig @localhost -p 5353 nl6.local AXFR) or from /api/v1/devices. Duplicate names are disambiguated deterministically (lowest IP keeps the bare label).

How it stays current​

Creating or deleting a device marks the zones dirty. A debounced worker waits for a quiescence window (-dns-debounce, default 1s), then bumps the SOA serial once and NOTIFYs each secondary. A large batch therefore coalesces into a single transfer rather than one per device. Try it:

# Create a device, then watch the serial advance and CoreDNS pick it up.
curl -s -XPOST localhost:8080/api/v1/devices \
  -H 'content-type: application/json' \
  -d '{"device_count":1,"start_ip":"10.42.0.50"}'
curl -s localhost:8080/api/v1/dns/status | jq '.zones, .zone_bumps'

Adjust the configuration​

  • Different domain / subnets. Change -dns-domain and -dns-reverse-zone on the nl6 service, and add/rename the matching secondary blocks in the Corefile. A CoreDNS secondary needs one block per zone. Device IPs outside every configured reverse zone resolve forward-only (no PTR).
  • No NOTIFY. Drop -dns-notify; CoreDNS still picks up changes on its SOA refresh interval, but less promptly.

Next​