feat: allocate a network's fabric identity - #412
Closed
scotwells wants to merge 1 commit into
Closed
Conversation
A network's identity on the fabric is chosen at random, per location, by whatever places the network there, and checked against nothing. Two locations of one network get unrelated values, so the Route Target derived from each imports neither the other's routes, and the edge VRF device is named differently in each. Two networks that collide do not fail closed: they merge, and tenant traffic crosses between them. The platform now allocates one identity per network, once, and projects it into every location the network reaches. It is drawn from an identifier space that is never routed, so uniqueness, exhaustion accounting, quota and audit come from an allocator that already solves them. It is 32 bits wide because that is what survives into the Route Target; anything wider would be uniqueness the platform believes it has and the fabric does not. A network that holds no identity behaves exactly as it does today, so nothing already running changes. Giving the networks that exist a platform-wide identity renames live VRF devices and moves Route Targets on traffic in flight, and needs its own proposal. Key changes: - Add Network.status.fabricIdentity, immutable once allocated, with an Allocated condition, and project it into NetworkContext.spec beside the address families and MTU already carried there - Allocate a /64 from a platform-scoped identifier pool and read the identity out of the block's index within it; refuse the zero block, since zero is what an unallocated network reads as - Add a platform tenancy to the IPAM config and a ClientForPlatform seam on the client factory, so an identity is never drawn from a consumer's own space or gated on them enabling the address service - Refuse at startup a deployment naming an identifier space with nowhere platform-owned to allocate it from
This was referenced Aug 27, 2026
Contributor
Author
|
Fabric identity moved to datum-cloud/cloud#12, now merged. It is a fabric concept, and network-services-operator would have allocated, stored and versioned a value it never reads and cannot interpret. The allocation logic, retention rules and enumerability analysis carried over intact; only the owning service changed. network-services-operator ends up carrying none of it. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A network that spans two locations is two networks to the fabric. Whatever places it there picks an identity at random per location and checks it against nothing, so the Route Target derived from each imports neither the other's routes and the edge VRF device is named differently in each; two networks that happen to collide silently merge and pass traffic between each other. The platform now allocates one identity per network, once, and projects it into every location the network reaches, so a network is one network wherever it is placed.
The identity is drawn from an identifier space that is never routed, which is what gives it uniqueness, exhaustion accounting, quota and an audit trail without a second allocator. Nothing already running changes: a network holding no identity behaves exactly as it does today.
Scope
Scaffolding only. Networks that exist today keep their per-location identities, because giving them a platform-wide one renames live VRF devices and moves Route Targets on traffic in flight. That migration needs its own proposal, and this does not attempt it.
Allocation is off unless a deployment names an identifier space, and none does yet. The design and what it rests on are in
docs/enhancements/vpc-fabric-identity.md.Assumed
The address service has no platform tenancy of its own. Until it does, every network's identity is allocated in one project control plane the platform owns, which is one pool and one allocator for the whole platform. The seam is a single method on the client factory, so nothing above it changes when a real platform scope arrives.
Related