Autoscaling Controller  ·  View 08 of 37  ·  2 · Structure

Decision Plane

Sharded with a lease per shard, and what one shard does in one interval.

Editable source SVG draw.io All views
Sharding — decision throughput scales with target count Shard 0 · leader Decision Engine holds valid lease Shard 1 · leader Decision Engine holds valid lease Shard 2 · standby Decision Engine no lease, no write Lease authority Lease Store K8s Lease · 10 s What one shard does in one interval Inputs Qualified Signals with freshness Policy Version immutable Bounds and Overrides Propose Target Tracking hold a setpoint Step / Threshold capacity per band Scheduled Floor time window Forecast Floor may not undercut Constrain Combination Rule max out · min in Hysteresis + Sustain 15% · 2 / 6 Rate Limiter +100% / -10% / 60 s Hard Bounds min, max, ceiling Emit Desired Capacity durable write Journal Record acted on or not lease denied clamped last Decision Plane — Sharded, Leader per Shard Application we own Risk / gap Security / platform Data store failure / alternate synchronous Bounds are applied after every policy has spoken, by a component no policy can reach. A policy cannot breach a limit it also computes. v 1.0 · owner Data & AI Global Practice · date 2026-09

Decisions

  • Leader per shard, not leader per region: losing one replica affects its shard and nothing else.
  • A desired-state write without a valid lease is rejected by the store, so split brain is prevented at the write rather than detected afterwards.
  • The constraint stage runs after every policy has spoken, in a component no policy can reach.

Where the decision runs

  • Central regional controller, not a per-cluster agent or a sidecar.
  • Chosen because the constraint set is substantially global: fleet capacity ceilings, per-region spend ceilings and shared warm pools cannot be enforced from inside one target.
  • The cost is a shared blast radius, which is why the plane separation and fail-static behaviour matter more here than they would for a distributed design.

Assumptions

  • Lease term 10 s against a 15 s interval, so a lost leader costs at most one interval of decisions for its shard.
  • Shard assignment is by consistent hash of target id; rebalancing is rare and does not need to be fast.
◀  Signal Plane All views Actuation Plane  ▶
Solution Architecture v1.0 · Data & AI Global Practice · 2026-09 · 37 views · Azure and open source08-decision-plane