Skip to content

CONCEPT Cited by 1 source

Control-plane data

Control-plane data is small, strongly-consistent coordination state that helps services operate correctly — placement information (where resources are stored), leadership information (which machine is allowed to write), configuration state, and similar. Distinguished from data-plane data by its consistency requirements and access patterns.

Characteristics

  • Written infrequently — updates are rare compared to reads
  • Read from everywhere — any data center needs access
  • Must be strongly consistent — readers must never see stale or inconsistent state
  • Small — typically metadata, not bulk data
  • Critical — incorrect control-plane state causes incorrect service behavior globally

Examples (Cloudflare)

  • Placement information — where an AI model instance is stored
  • Leadership information — which machine is currently allowed to perform writes to a database
  • Routing state — how requests should be distributed
  • Feature-flag state — which configuration is active

Why consensus

The combination of strong consistency + global accessibility + fault tolerance makes consensus the appropriate mechanism. Weaker consistency (eventual, causal) risks split-brain or stale routing decisions that cascade into data-plane failures.

Meerkat's target workload

Meerkat is designed specifically for control-plane data: small pieces of state written infrequently but requiring linearizable access from any data center. Its consensus round-trip cost makes it unsuitable for high-write-throughput data-plane workloads.

Seen in

Last updated · 573 distilled / 1,747 read