title: TPA-RFC-6: Naming Convention
costs: None
approval: TPA
affected users: TPA
deadline: 2020
status: standard
Summary: naming things is hard, but should at least be consistent. This policy documents how domain names are used, how to name machines, services, networks and might eventually document IP addressing as well.
Domain names
Tor uses two main domain names for things:
torproject.org
torproject.net
There might be other domains managed by us or registered in the DNS,
but they should eventually point to one of those, generally
torproject.org
.
All TPA-managed machines and services on those machines should be
under torproject.org
. The naming scheme of the individual machines
is detailed below. This is managed by TPA directly through
howto/dns.
External services and machines can be hosted under
torproject.net
. In that case, the only association is a CNAME
or
A
record pointing to the other machine. To get such a record,
contact TPA using the normal communication channels detailed in
support.
Machine names
There are multiple naming schemes in use:
- onion species
- role-based
- location-based
We are trying to phase out the onion-based names, in favor of more descriptive names. It kind of takes the soul out of the infrastructure, but it makes things much easier to figure out for newcomers. It also scales better.
Onion species
Note that this naming scheme is deprecated. Favor role-based names, see below.
Wikipedia list of onion species, preferably picking a first letter matching purpose (e.g. "m" for monitoring, "b" for backups, "p" for puppet) and ideally not overlapping with existing machines at debian.org in the first three letters or at least the short hostname part
Example: monticola.torproject.org was picked as a "monitoring" ("mon") server to run the experimental Prometheus server. no machine is named "monticola" at Debian.org and no machine has "mon" or smaller as its first three letters there either.
Roles
Another naming scheme is role-ID
, where:
-
role
is what the server is for, for examplegitlab
,mon
for monitoring,crm
, etc. try to keep it short and abbreviate to at most three letters if role is longer than five.role
might have a dash (-
) in it to describe the service better (crm-ext
vscrm-int
) -
ID
is a two-character number, padded with zero, starting from one, to distinguish between multiple instances of the same server (e.g.mon-01
,mon-02
)
Some machines do include a location name, when their location is
actually at least as important as their function. For example, the
Ganeti clusters are named like gnt-LOC
where LOC
is the location
(example, gnt-fsn
is in Falkenstein, Germany). Nodes inside the
cluster are named LOC-node-ID
(e.g. fsn-node-01
for the first
Ganeti node in the gnt-fsn
cluster).
Other servers may be named using that convention, for example,
dal-rescue-01
is a rescue box hosted near the gnt-dal
cluster.
Location
Note that this naming scheme is deprecated. Favor role-based names, see above.
Another naming scheme used for virtual machines is hoster-locN-ID
(example hetzner-hel1-01
), where:
-
hoster
: is the hosting provider (examplehetzner
) -
locN
: is the three-letter code of the city where the machine is located, followed by a digit in case there are multiple locations in the same city (e.g.hel1
) -
ID
: is an two-character number, padded with zero, starting from one, to distinguish multiple instances at the same location
This is used for virtual machines at Hetzner that are bound to a specific location.
Network names
Networks also have names. The network names are used in reverse DNS to designate network, gateway and broadcast addresses, but also in howto/ganeti, where networks are managed automatically for virtual machines.
Future networks should be named FUN-LOCNN-ID
(example
gnt-fsn13-02
) where:
-
FUN
is the function (e.g.gnt
for howto/ganeti) -
LOCNN
is the location (e.g.fsn13
for Falkenstein) -
ID
is a two-character number, padded with zero, starting from one, to distinguish multiple instances at the same function/location pair
The first network was named gnt-fsn
, for Ganeti in the Falkenstein datacenter
. That naming convention is considered a legacy exception
and should not be reused. It might be changed in the future.
Deadline
Considering this documentation has been present in the wiki for a while, it is already considered adopted. The change to deprecate the location and onions names was informally adopted some time in 2020.
References
- RFC1178: "Choosing a Name for Your Computer", August 1990
- RFC2100: "The Naming of Hosts", 1 April 1997
- Wikipedia: Computer network naming scheme
- https://namingschemes.com/
- Another naming scheme
- Location code names from the UN