docs: artifact registry infrastructure ADR

Steve Xuerebrequested to merge
docs/ar-infra-adr into main

What

  • Add ADR-024 (Multi-Tenant SaaS Artifact Registry Infrastructure) describing the cell-based architecture, per-Cell data plane (GKE, PostgreSQL, Redis, GCS), edge layer (Cloudflare + Cloud CDN), and Cell lifecycle managed via Terraform, Argo CD, and Fairway.
  • Add the corresponding architecture diagram and link the new ADR from the Artifact Registry design document index.
  • Update ADR-022 (Namespace Decoupling) to describe routing in terms of the shared Anchor Router and Anchor Topology service introduced by ADR-024, replacing the generic "topology service" wording.
  • Update ADR-004 (Data and Application Limits) and ADR-006 (Technology Stack) to reference ADR-024 for edge rate limiting, per-Cell deployment, and per-Cell Cloud CDN configuration.
  • Add Omni and Yugabyte to the Vale spelling exceptions and correct an existing Enviroments typo.

Why

ADR-024 fills the gap between the application-level ADRs and the infrastructure that runs them. The earlier ADRs (003, 006, 022) referenced a Runway runtime and a generic "topology service" without defining how the service is actually deployed, scaled, isolated per tenant, or recovered. As Artifact Registry is one of the first stateful GitLab Modules heading to production, we need a single canonical document describing the Cell-based deployment model, ownership boundaries between Package and Production Engineering, and the path from MVP (single Cell) to multi-Cell with slug-based routing. The cross-references in ADR-004, ADR-006, and ADR-022 keep the rest of the design document consistent with that model.

Assisted-by: Claude:claude-opus-4-7

Edited by Steve Xuereb

Merge request reports