RWS Architecture article

ADR-015 Keycloak IAM (Identity and Access Management) Architecture

keycloak must provide resilient smart access across clusters. We need to choose a deployment model that meets availability, scale, and operational requirements while remaining mana

  1. Typeadr
  2. Statusdraft
  3. Domainplatform
On this page
  1. Context and Problem Statement
  2. Decision Drivers
  3. Considered Options
  4. Decision Outcome
  5. Next Decision Trigger
  6. Pros and Cons of the Options
  7. Option 1 – Replicated clusters with active/passive failover
  8. Option 2 – Independent clusters with failover automation

Context and Problem Statement

keycloak must provide resilient smart access across clusters. We need to choose a deployment model that meets availability, scale, and operational requirements while remaining manageable in automation workflows.

Scope boundary: this ADR covers IAM architecture (identity domains, integration posture, and operational model). Detailed SSO deployment topology and failover options remain in ADR-012.

Decision Drivers

  • Resilience to single-node and single-datacenter failures.
  • Capacity for peak load across all integrated applications.
  • Ability to scale Keycloak and cache layers independently when needed.

Considered Options

  1. Multiple Keycloak clusters with active/passive replication across datacenters.
  2. Multiple Keycloak clusters without cross-site replication, with failover automation.

Decision Outcome

Decision remains draft. Both models remain under evaluation for operational complexity, failover behavior, and scalability.

Next Decision Trigger

  • Decision owner: Platform Architecture with Security Architecture review.
  • Promote to accepted when identity domain design, failover behavior, and scaling test evidence are approved.
  • If neither model satisfies operational objectives, add a third option with explicit trade-offs before 2026-03-15.

Pros and Cons of the Options

Option 1 – Replicated clusters with active/passive failover

Pros

  • Predictable failover with shared state between datacenters.

Cons

  • Complex replication and cache configuration.

Option 2 – Independent clusters with failover automation

Pros

  • Simpler per-site operation and isolation.

Cons

  • Requires automated activation and potential user-impact during failover.