Real-Time Analytics Platform  ·  View 10 of 21  ·  Data

Schema Contract Governance

How an event schema changes without a redeployment and without breaking a consumer — the lane-by-lane path from proposal to retirement.

Editable source SVG draw.io All views
Propose
Propose
Validate
Validate
Register
Register
Publish
Publish
Retire
Retire
Producer team
Producer team
Schema pull request
Schema pull request
Local compat check
Local compat check
SDK regenerated
SDK regenerated
Cut over to v(n)
Cut over to v(n)
Platform CI
Platform CI
Lint and policy
Azure DevOps
Lint and policy...
Backward-compat gate
Backward-compat gate
Push to registry
Push to registry
Contract tests
Contract tests
Sunset job
90 d window
Sunset job...
Schema Registry
Schema Registry
Diff against v(n-1)
Diff against v(n-1)
Avro schema v(n)
Avro schema v(n)
Schema ID in header
Schema ID in header
Marked deprecated
Marked deprecated
Consumers
Consumers
Impact report
Impact report
Consumer contract test
Consumer contract test
Resolve by schema ID
Resolve by schema ID
Migration confirmed
Migration confirmed
breaking change rejected
breaking change rejected
Schema Contract Governance
Schema Contract Governance
Application we own
Application we own
Decision point
Decision point
Security / platform
Security / platform
failure / alternate
failure / alternate
Only backward-compatible changes reach the registry; a breaking change requires a new event type.
Only backward-compatible changes reach the registry; a breaking change requires a new event type.
v 1.0 · owner Data Governance · date 2026-08
v 1.0 · owner Data Governance · date 2026-08Text is not SVG - cannot display

Compatibility rule

  • BACKWARD compatibility is enforced in CI; a breaking change requires a new event type
  • Consumers resolve payloads by schema ID from the header, never by assumed shape
  • Deprecation runs a 90-day window with per-consumer usage reporting before sunset

Why the registry, not code

  • Schema changes ship without redeploying producers, processors or the hot store
  • The compatibility gate is machine-checked, so review is about semantics not syntax
  • Contract tests give the producer team evidence before the change reaches production

Risk

  • A semantically breaking but structurally compatible change passes the gate
  • Mitigation is the consumer impact report plus mandatory data-owner review, not automation
  • Registry availability is on the ingestion critical path; schema IDs are cached locally
◀  Analytics Data Model All views Stream Processing Pipeline  ▶
Solution Architecture v1.0 · Azure · Data & AI Architecture · 2026-0810-schema-contract-governance