This page documents the naming conventions of servers, routers, and data center sites.

Our servers currently fall in broadly two categories:

• Clustered servers: These use numeral sequences with a descriptive prefix (see #Networking and #Servers). For example: db1001.
• Miscellaneous servers: These used unique hostnames (see #Miscellaneous servers). For example: helium. This naming convention is deprecated and not used for new hosts, but some older miscellaneous-named hosts still exist.

Name reuse

Historically, we did not reuse names of past servers for new servers. For example, after db1001 is decommissioned, no other server will be named db1001. Ganeti VMs sometimes reuse hostnames, but bare metal typically will not.

The notable exception is networking gear, which are deterministically specified by rack. For example the access switch in Eqiad rack A8 is named asw-a8-eqiad. If it is replaced, the new switch will take the same name.

All hardware in the datacenter space is tracked in Netbox, which can be used to check for existing hostnames for both hardware and ganeti instances.

Data centers

Data centers were traditionally named as vendor initials (at time of lease signing) followed by the IATA code for a nearby major airport. For example, for Eqiad data center the vendor is Equinix, and IAD the large nearby airport. This convention was used from 2003 upto 2023. Because vendors go through acquisitions, and original initials no longer apply after some time, starting with Magru in 2023, only the airport code is used along with a freely chosen prefix.^[1]

Networking

Naming for network equipment is based on role and location.

This also applies to: power distribution units, serial console servers, and other networking infrastructure.

OpenStack deployments

[Datacenter Site][numeric identifier](optional dev suffix to indicate non-external non-customer facing deployments) - [r (if region)][letter for AZ]

• Current Eqiad/Codfw deployments will not fully meet these standards until rebuilt: [eqiad0 (deployment), eqiad (region), nova (AZ)]

Disks

• Arrays must use the Storage array device role in Netbox.
• Naming follows two conventions:

• Array is attached to a single host:

• hostname_of_host_system-arrayN

• Example: ms2001-array1, ms2001-array2

• all arrays get a number, even if there is only a single array.

• Example: dataset1001-array1

• Array is attached to multiple hosts

• Labs uses this for labstore, each shelf connects to two different hosts. As such, the older single host naming scheme fails.
• servicehostgroup-arrayN-site

• Example: labstore-array1-codfw, labstore-array2-codfw

Kubernetes

Any cluster that is not the main wikikube cluster should use a consistent identifier and follow these conventions:

• Control plane service name: <identifier>-ctrl
• Ingress service name: <identifier>-ingress [-ro|-rw] for active/active or active/passive
• Hostnames for control plane : <identifier>-ctrlXXXX.$site.wmnet
• Hostnames for kubelets : <identifier>-workerXXXX.$site.wmnet

Servers

sudo cumin --no-color 'A:all' 2>/dev/null | nodeset -S '\n' -e | sed 's/\..*//g' | sed 's/[0-9]\{4\}//g' | sort | uniq