feat(cloudflare): custom hostname SNI overrides and migration support

[mrozentsvayg]

What does it do ?

Three improvements to Cloudflare custom hostname management:

1. Origin SNI overrides

Extends the cloudflare-custom-hostname annotation to support Origin SNI
overrides using the format <customHostname>=<customOriginSNI>.

• No = suffix: SNI defaults to the origin server (existing behaviour, unchanged)
• Trailing = with no value: SNI is set to the request Host header
• =<value>: SNI is set to the specified hostname

The custom_origin_sni field is only sent to the Cloudflare API when
explicitly overridden (i.e. when SNI differs from the origin server),
since the field requires a Cloudflare account entitlement.

2. Skip self-referential custom hostnames

When a custom hostname resolves to a DNS name in the same managed zone,
creating a CH would be self-referential and is skipped with a warning.
This prevents Cloudflare API errors when the custom hostname and origin
are in the same zone.

3. Skip CH lifecycle with annotation value "-"

Setting the annotation to - tells external-dns to skip custom hostname
management for that record entirely — no create, update, or delete.
Existing custom hostnames in Cloudflare are left untouched.

This enables migrating CH ownership to another controller (e.g. a
Kubernetes operator) without triggering SSL re-validation from
delete+create cycles.

Motivation

Cloudflare for SaaS supports routing custom hostname traffic to different
backends based on SNI. Without SNI control, all custom hostnames must share
the same TLS routing, which is too restrictive for multi-tenant setups where
different customers need traffic routed to different origins at the TLS layer.

For example, when using Envoy Gateway in Merged Gateways deployment mode,
each custom hostname needs its own SNI to correctly route TLS traffic to the
right backend — something that wasn't possible before without manual
Cloudflare configuration.

The - sentinel value solves a critical migration problem: removing the
CH annotation causes external-dns to delete existing custom hostnames,
which triggers SSL re-validation (DCV). Cloudflare rate-limits DCV
attempts, and a delete+create loop can cause SSL outages that persist
until the rate limit expires (hours).

More

• Yes, this PR title follows Conventional Commits
• Yes, I added unit tests
• Yes, I updated end user documentation accordingly

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by:
Once this PR has been reviewed and has the lgtm label, please assign mloiseleur for approval. For more information see the Code Review Process.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[k8s-ci-robot]

Hi @mrozentsvayg. Thanks for your PR.

I'm waiting for a kubernetes-sigs member to verify that this patch is reasonable to test. If it is, they should reply with /ok-to-test on its own line. Until that is done, I will not automatically test new commits in this PR, but the usual testing commands by org members will still work. Regular contributors should join the org to skip this step.

Once the patch is verified, the new status will be reflected by the ok-to-test label.

I understand the commands that are listed here.

Details

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository.

[ivankatliarchuk]

This PR is more about traffic routing than DNS management, so very high chances it may be out of scope. It would be worth double-checking with the owners on Slack to see what they think.

[ivankatliarchuk]

Similar one #6085

[AndrewCharlesHay]

Here is a RFC for a repo that can house this kind of stuff

[mrozentsvayg]

@mloiseleur
this was discussed in #external-dns Slack; no blockers raised. With origin SNI support and coexistence with external controllers, this PR helps make the transition of non-DNS CF features out of external-dns smooth for existing users.

What do you think?

[mloiseleur]

🤔 I'm not sure I follow. Why someone who wants to migrate custom hostname management away from external-dns would take the risk to allow external-dns to delete those entries?

If I were in this case, I would probably disable this feature entirely in external-dns as my first step.

And if, for whatever reason, I need to do it progressively, I would most likely:

1. Set external dns replicas to 0
2. Remove CH annotations on all relevant endpoints
3. Restore external dns replicas to 1

We can add a doc / guide about this, but I'm not sure if I understand the value in this skip feature that does not really skip and may delete

[mrozentsvayg]

The use case is progressive migration in a production environment where external-dns manages both DNS records and custom hostnames for 170+ endpoints across multiple deployments.

You can't scale external-dns to 0 -- it manages A records for all endpoints, not just custom hostnames. And you can't remove CH annotations from all endpoints at once when migrating to a new CH manager (in our case a dedicated Kubernetes operator). You need to verify the new manager works per-endpoint before moving to the next.

The - sentinel enables per-endpoint opt-out:

1. Set - on one endpoint → new manager takes over its CH → verify SSL active
2. Repeat for remaining endpoints progressively
3. After all endpoints migrated: --cloudflare-custom-hostnames=false (the global disable you're suggesting)

Step 3 is safe precisely because steps 1-2 already verified each endpoint. Without the sentinel, there's no way to do steps 1-2 -- you'd have to go from "external-dns manages all CHs" to "external-dns manages no CHs" in one shot, with no rollback if the new manager has issues.

You're right that the sentinel doesn't protect against deletion during A record lifecycle changes -- that's documented in the PR as a known limitation. In practice, A record changes are infrequent during migration (you avoid infrastructure changes during the coexistence window). We ran this in production across 3 clusters with zero incidents during the migration phase.