Lesson 4 — The Four Metadata Architectures
The fastest way to distinguish the four patterns is to ask two questions:
- Where does Metadata persist?
- Can managed changes flow back to sources?
Centralized
Multiple source repositories are harvested into one persistent enterprise repository.
Strengths: integrated/global search, common presentation, standardization, manual enrichment, enterprise relationships.
Challenges: adapters, extraction, refresh schedules, synchronization, validation, keeping copied Metadata current.
Deciding clue: persistent central copy.
Distributed
A common portal/broker queries source repositories live at request time. There is no persistent central enterprise store.
Strengths: source-held Metadata can remain current; less duplication.
Challenges: source uptime/quality directly affects results; difficult to add enterprise-only manual Metadata; standardization and persistent cross-source analysis are harder.
Deciding clue: live retrieval + no persistent central store.
Hybrid
Selected enterprise-standard, manually enriched, or critical Metadata is persisted centrally, while other details remain at source and are retrieved when needed.
This can be useful when operational Metadata changes rapidly, source volume is large, or enterprise consistency is needed without copying every changing detail continuously.
Deciding clue: deliberate central persistence for some Metadata + source retrieval for other Metadata.
Bi-Directional
Metadata flows into the enterprise environment, but governed changes may also flow back to originating source repositories.
This creates additional synchronization, version, authority, timing, and conflict-resolution problems.
Deciding clue: controlled write-back / feedback to sources.
What does not prove Bi-Directional?
Frequent refresh does not. A centralized repository can refresh every minute and still remain one-way. Bi-Directional requires controlled return flow.
Architecture is a requirement decision
Do not buy a catalog first and force the requirements into its storage model. Consumer needs should determine persistence, latency, integration, manual enrichment, history, security, and delivery patterns.
Changed-fact drill
- One search interface over a persistent harvested copy → Centralized.
- Same interface, but every query goes to sources and nothing persists → Distributed.
- Approved definitions persist, runtime details stay source-held → Hybrid.
- Approved repository changes update source repositories → Bi-Directional.
Source anchor: pp. 408–412.