Open Infrastructure

h1. The ungleich DNS infrastructure

{{toc}}

h2. Status

This document is *IN PRODUCTION*.

h2. SEE ALSO

* [[The_ungleich_network_infrastructure]]

h2. Overview

|                   | *place4*             | *place5*                  | *place6*                          |

| *DNS64 prefix*    | -                    | 2a0a:e5c0:0:1::/96        | 2a0a:e5c0:2:10::/96               |

| *DNS resolver*    | -                    | 2a0a:e5c0::3 2a0a:e5c0::4 | 2a0a:e5c0:2:1::5 2a0a:e5c0:2:1::6 |

| *DNS64 resolvers* | -                    | -                         | unbound1.place6.ungleich.ch (2a0a:e5c0:2:12:0:f0ff:fea9:c451)

unbound2.place6.ungleich.ch (2a0a:e5c0:2:12:0:f0ff:fea9:c45d)|

| *DNS auth BIND*   | dns1.ungleich.ch     | dns2.ungleich.ch          | dns3.ungleich.ch                  |

|                   | 2a01:4f8:150:7092::2 | 2a0a:e5c0::1              | 2a0a:e5c0:2:1::7                  |

|                   | 176.9.50.202         | 185.203.112.1             | 185.203.114.1                     |

| *DNS auth KNOT*   | -                    | dns7.ungleich.ch          | dns6.ungleich.ch                  |

* Every place has 2 redundant caching nameservers.

* All zones have 3 authorative nameservers, located in 3 different places

* Important zones (like ungleich.ch) need to be resolvable, even if a place goes offline

** For this reason some authorative data needs to be on the caching name servers

** For this reason we stay with a bind9 based setup for the moment (might change in the future)

h2. Architecture

In total we are running 5 servers that are responsible for caching and authorative answers:

* Authorative

** 1x server in place4 (bind)

** 1x VRRP IP of routers in place5 (bind)

** 1x VRRP IP of routers in place6 (bind)

* Caching

** 2x server ip of router in place5 (bind)

** 2x server ip of router in place6 (bind)

h2. How to update the ungleich DNS servers

To update all servers, use:

<pre>

cdist config d{1..7}.ungleich.ch

</pre>

|    |                  | "virtual"        | Note       |

| d1 | router1.place5   | dns2.ungleich.ch | cache+auth |

| d2 | router2.place5   | dns2.ungleich.ch | cache+auth |

| d3 | router1.place6   | dns3.ungleich.ch | cache+auth |

| d4 | router2.place6   | dns3.ungleich.ch | cache+auth |

| d5 | server1.place4   | dns1.ungleich.ch | auth       |

| d6 | dns6.ungleich.ch | -                | auth+synth |

| d7 | dns7.ungleich.ch | -                | auth+synth |

h2. How to use the authorative DNS servers in zone files

Add the following to your zone file:

<pre>

    ; server1.place4

    IN NS dns1.ungleich.ch.

    ; vrrp active router @ place5

    IN NS dns2.ungleich.ch.

    ; vrrp active router @ place6

    IN NS dns3.ungleich.ch.

</pre>

h2. DNS64 at datacenterlight/ipv6onlyhosting

"NAT64":https://en.wikipedia.org/wiki/NAT64 allows ipv6-only nodes to reach the v4 world. The production infrastructure for DCL/V6OnlyHosting runs at place6 and networks are assigned as follow:

* IPv6Only VMs are assigned to the `place6-ipv6-nat64` OpenNebula network.

* Dual-stack VM are assigned to the `place6-ipv4` and `place6-ipv6`

The `place6-ipv6-nat64` networks *provides NAT64* but the `place6-ipv6` *does not*: we do not want ipv4-capable VMs to be NAT'ed behind NAT64. Due to *legacy reasons*, some ipv6only VMs are in `place6-ipv6` but have NAT64 due to hardcoded per-ip configuration our bind DNS server (see `type/__ungleich_dns_server` type in dot-cdist).

h3. place6-ipv6-with-ip-spoofing

This OpenNebula network is used to routes v6 prefixes (/64, /56, /48) to customer VMs and is shared by Ipv6-Only and Dual-Stack VMs: NAT64 is *disabled* on this network. IPv6-Only customers on this network *MUST* use unbound1.place6.ungleich.ch and unbound2.place6.ungleich.ch as name server. Their `/etc/resolve.conf` file should look like:

<pre>

nameserver 2a0a:e5c0:2:12:0:f0ff:fea9:c451

nameserver 2a0a:e5c0:2:12:0:f0ff:fea9:c45d

</pre>