DNS/Netbox

IP allocation is done via Netbox, that is our IPAM, and DNS records of IPs allocated in Netbox are automatically generated from Netbox data.

Infrastructure

IP allocation is done on Netbox.
Netbox data is exported via Netbox#DNS.
Netbox data is checked out on the authoritative DNS servers in /srv/git/netbox_dns_snippets.
When compiling the gdnsd final zones, the Netbox data is copied into /etc/gdnsd/zones/netbox for later inclusion.
In the actual zonefiles, within an $ORIGIN , the related snippet file auto-generated from Netbox data is included using the $INCLUDE directive.

IP Allocation

Netbox is currently the IPAM for the production infrastructure and IP allocation must be done in Netbox unless it's on some very specific subnets that are exceptions and not managed within Netbox for now. The frack hosts are exempt from this as their IPs are not tracked in Netbox.

As historical data, the management IP address allocation has been migrated to Netbox on Wed. June 24th 2020 and the primary IPs address allocation for production infrastructure devices has been migrated to Netbox on Sep. 14th 2020.

DNS records involved

The following records are managed by Netbox in all data centres (except the frack infrastructure):

Management forward (A) and reverse (PTR) records for both the hostname (foo.mgmt.eqiad.wmnet) and the asset tag (wmf1234.mgmt.eqiad.wmnet)
Primary IPv4 (A ) and IPv6 (AAAA ) and related reverse (PTR ) records for the hostname (foo.eqiad.wmnet or foo.wikimedia.org )
Any additional IP in Netbox that has the DNS Name property set.

Missing AAAA records for primary IPv6

Although we're aiming for all hosts to have A/AAAA DNS records for their primary IPs, some clusters do not yet fully support IPv6 and as such do not have the DNS Name set in Netbox for their primary IPv6 address. Although not ideal, that's still considered ok as long as there is a mid/long term plan to remove the technical debt and make the cluster IPv6 compatible. If you manage one of those clusters, make sure to always notify DCOps of the fact that the AAAA records should be skipped when provisioning new hosts to prevent their addition.

Mixed clusters

A separate, more dangerous case is when a given cluster is mixed, having some hosts with the AAAA records for their primary IPv6 and some without. This is an indication that the cluster was not fully supporting IPv6 in the past and some hosts now have their primary IPv6 with the AAAA records, potentially because DCOps was not notified to skip the AAAA assignment at provision time. In this case there are two possible options:

The hosts that have AAAA records are mis-configured, clients are trying to connect via IPv6, fail and have to retry on IPv4, and the AAAA records should be removed. In this case get in touch with SRE/Infrastructure Foundations/Contact for their removal.
The hosts that have AAAA records are functioning correctly, meaning that the cluster actually supports IPv6 and we could just add the missing AAAA records to the remaining hosts. In this case follow DNS/Netbox#Add_missing_DNS_name_to_the_primary_IPv6_address.

Operations

Update generated records

To update the dynamically generated records based on the current Netbox data and deploy them to all the authoritative DNS servers, the sre.dns.netbox cookbook must be run. The cookbook must be run anytime records are changed in Netbox. See also Cookbooks#Cookbook_Operations. For example:

 sudo cookbook sre.dns.netbox -t T12345 "Add newly racked cp hosts in eqiad"

There is an Icinga check if changes in Netbox are not committed after a while, see Monitoring/Netbox_DNS_uncommitted_changes for troubleshooting.

If when running the cookbook the presented diff show changes unrelated to your work, follow the instructions in Monitoring/Netbox_DNS_uncommitted_changes#What_to_do.

Force update generated records

It might happen that one or more authdns hosts fail to run authdns-update, leading to an inconsistent state. The sre.dns.netbox cookbook offers a --force option, that takes as input the SHA of the git commit that you want all authdns servers to be synced on. In order to find the SHA, just do:

ssh netbox.wikimedia.org
sudo -i
cd /srv/netbox-exports/dns.git
git log -1

Convert an hardcoded $ORIGIN to Netbox

This is an example patch to convert an hardcoded $ORIGIN to the dynamically generated data.

Check if any automatically generated zone file is not included

This is the way to check if any of the automatically generated zone files is not $INCLUDEed in the operations/dns repository. From a local checkout of the automatically generated repository (see Netbox#DNS), assuming that the operations/dns repository is checked out in ../dns/, run:

for file in $(ls *); do git -C ../dns/ grep "INCLUDE" | grep -q "netbox/${file}" || echo "missing ${file}"; done

Some outstanding files are expected at the moment, namely related to SVC, wikimediacloud, and frack records.

You can use this script to help generate reverse zonefile includes for IPv6 and submit a CR to the dns repo to add them.

Atomically deploy auto-generated records and a manual change

In case there is a change in the generated Netbox data that requires at the same time a change in the manual operations/dns repository too, this is the procedure to follow:

prepare the patch for the operations/dns repository, send it for review
- CI will fail if there is any $INCLUDE of files not yet existing in the generated data, that's expected
Modify the data in Netbox
Run the sre.dns.netbox cookbook as described above in DNS/Netbox#Update_generated_records adding the option --skip-authdns-update
Comment recheck in the CR sent for the operations/dns repository, CI should now pass
Merge and deploy the patch, once deployed it will include also the generated data that was pushed but not deployed by the cookbook, making the change atomic from the DNS point of view.

The above procedure should be run all together without let too much time pass between each step and it will be wise to ask in the various SRE channels to refrain during this operation from running authdns-update or any of the cookbooks that in turn run the sre.dns.netbox one (as of Oct. 2020 sre.hosts.decommission and sre.ganeti.makevm) or the sre.dns.netbox cookbook itself.

As an example the above procedure was used when a new prefix was created and as a result the generated data got moved from one file to another, see operations/dns/+/632953

Modify the generated data in an emergency

To modify the generated data in an emergency it's possible to just run the sre.dns.netbox cookbook as described above in DNS/Netbox#Update_generated_records adding the option --emergency-manual-edit.

After the generation of the data the cookbook will stop and prompt the user to make the modifications, something like:

Generated temporary files are available on netbox1001.wikimedia.org:/tmp/dns-c25pcHBldHM-iad8k5x_
SSH there, as root modify any file, git stage them and run "git commit --amend" to commit them
Then run "git log --pretty=oneline -1" and copy the new SHA1 of HEAD

N.B.: any subsequent run of the cookbook will try to revert the manual changes, make all SREs aware of the emergency situation.

Add missing DNS name to the primary IPv6 address

Adding a missing DNS name to a primary IPv6 address is simple, but if it needs to be done on many hosts, it can be done programmatically, see below.

Before proceeding make sure that the host does fully support IPv6, all services are listening also on the IPv6 address, any ACL/grant/ferm rule is also setup for the IPv6 space. Keep in mind that most clients will default to connect via IPv6 once the AAAA records are live.

Single host

Find the host's primary IPv6 address in Netbox, click edit and fill the DNS Name field with the FQDN of the host. To double check verify that the host primary IPv4 has the same DNS Name.
Run the sre.dns.netbox cookbook as described above in DNS/Netbox#Update_generated_records

Multiple hosts

When the addition has been tested on few hosts and you're ready to convert multiple hosts:

Get in touch with the SRE/Infrastructure Foundations/Contact so that the DNS names can be added programmatically in Netbox running something like this.
Run the sre.dns.netbox cookbook as described above in DNS/Netbox#Update_generated_records

Transition FAQ

Am I affected?

If your workflows will be affected by this change depends entirely on your interaction with the operations/dns repository:

I never read or contribute to this repository:
- you're not affected and nothing has changed for you. You can stop reading here.
I sometimes read or search for things in this repository:
- you're marginally affected as the manual records were moved from the operations/dns repository and replaced by the auto-generated files. You can search directly in Netbox. If you want to see the raw content of the generated files you can clone the auto-generated repository to read or search in it following the instructions in Netbox#DNS. You can optionally read the rest of the document.
I contribute to the repository:
- you're affected and should keep reading this FAQ section and the rest of the document.

What has changed

IP allocation is now done directly on Netbox , that is our IPAM tool. Only Fundraising-tech (frack) non-mgmt records were left out of this transition.
- All new host's primary IPv4/IPv6 are automatically assigned to them at provision time.
- Additional IPs require manual allocation in Netbox [see below] unless they are for Cassandra instances, that use case is already covered by the provisioning script.
The automatic DNS record generation (see above DNS/Netbox#Update_generated_records ) generates all of the records present in Netbox.

Why some IPv6 do not have a related AAAA/PTR DNS records?

Some clusters do not (yet?) support IPv6 on their infrastructure $hostname.$dc.$wmnet / $hostname.wikimedia.org addresses. For those, when the data was imported into Netbox, to keep the existing behaviour, the DNS Name field of the related IPv6 address in Netbox has been left empty. That means that at generation time only the IPv4 A/PTR records are generated and no AAAA/PTR record is generated for the IPv6.

Who can I ping for questions?

For questions, concerns or comments please get in touch with the SRE/Infrastructure Foundations/Contact.

What to do if the diff has spurious changes?

Follow the instructions in Monitoring/Netbox_DNS_uncommitted_changes#What_to_do.

How to allocate primary IPs for a server

Physical hosts

The management, primary IPv4 and primary IPv6 for any new physical host will be automatically assigned at provision time by DCOps running a Netbox script.

Virtual machines

For Ganeti virtual machines the sre.ganeti.makevm cookbook takes care of the new workflow automatically.

How to manually allocate a special purpose IP address in Netbox

This procedure is meant to be used only to create IPs in Netbox that are not attached to any device's interface because have special purposes like virtual IP addresses (VIPs, which are generally used for service addresses). Depending on real life use cases the following procedure might be automated into a Netbox script in the near future.

Go to the VLANs page in Netbox VLANs and Netbox Prefixes
Search for the correct VLAN based on datacenter, type, row (if applicable), etc.
Click on the desired prefix (v4 or v6) in the Prefixes column for that VLAN
- If you are setting up a new LVS service, use the prefix for LVS service IPs: codfw, eqiad
Click on the IP Addresses tab in the prefix page
Click on the Add an IP Address green button on the top-right, Netbox will automatically select the first available IP in that subnet
- Make sure to change the netmask to /32 in the address field for IPv4.
- To create an IPv6 that is a mapped version of an existing IPv4, modify the Address field at the top to override the automatically selected address, set the netmask to /128
- If this is a VIP, make sure you get the same last octect in both eqiad and codfw datacentres
Select the relevant Role (VIP for LVS, anycast, etc.)
Set the DNS Name field with the FQDN to assign to this IP
Select the Tenant if applicable (FR-Tech, RIPE)
Click on the Create blue button at the bottom
Once all Netbox changes are completed follow the instructions at DNS/Netbox#Update_generated_records
For VIPs that are in the *.svc.* zones, until T270071 will be solved, it is also required to perform a manual change in the operations/dns repository to add the IP manually. This is the one that will be used, the automatically generated records for *.svc.* zones are not currently used.