Skip to content

Model & Compute Platform — workers, grains and sagas

Generated from the architecture model

This page is rendered from model/**/*.yaml by tools/docs_model during mkdocs build. Edit the YAML, not this page; hand-written design lives on the linked service, platform and plane pages.

Workers

Worker Service Trigger Output events Retry Idempotency key Concurrency
UtilizationCollector ComputePoolService schedule */1 * * * * — fixed ×2 poolId + minute 4 by Region
WarmKeeper ComputePoolService schedule */5 * * * * ModelWarmed, ModelEvicted exponential ×3 poolId + model + window 2 by PoolId
DeploymentHealthProbe ModelRouterService schedule */2 * * * * DeploymentStateChanged fixed ×2 deploymentId + probeWindow 4 by Provider
ReceiptReconciler ModelRouterService schedule 30 * * * * ModelInvoked exponential ×5 invocationId + statementRef 2 by TenantId
CapabilityEvidenceImporter ModelRouterService event QualityScoreService.EvaluationCompleted CapabilityEvidenceRecorded, DeploymentStateChanged exponential ×5 deploymentId + suiteVersion + evidenceRef 2 by TenantId
  • UtilizationCollector (ComputePoolService): Marks pools whose utilization report is older than the staleness window as not placeable and refreshes cold-start estimates from observed loads.
  • WarmKeeper (ComputePoolService): Keeps the models of profiles with warm-keep policy loaded on their pools and evicts models unused beyond the idle window.
  • DeploymentHealthProbe (ModelRouterService): Probes every Approved deployment with a minimal request and marks it Degraded or Approved; records latency.
  • ReceiptReconciler (ModelRouterService): Confirms receipts with usage Unknown or Estimated from provider statements and forwards the correction to Usage Accounting.
  • CapabilityEvidenceImporter (ModelRouterService): Imports evaluation records of the capability suite into deployment scores and moves Candidate deployments to Evaluated.

Grains (actors)

Grain Service Key Reminders / timers Reentrant Summary
PoolGrain ComputePoolService string = poolId staleness no Serialises placements and cache updates of one pool so that capacity is never double-booked.

Sagas

None modelled

No sagas are modelled for this item yet.