auto-scaling

Threshold-Based Auto-Scaling with Verification

Your platform runs multiple services at varying load levels. When CPU crosses 80%, you need to scale up; when it drops below 30%, you should scale down to save cost. But scaling without verification risks silent failures -- pods crashing, load not actually decreasing. This workflow analyzes metrics, plans the action, executes it, and confirms the result.

Workflow

service, metric, threshold
         |
         v
  +-------------+     +-----------+     +-------------+     +-------------+
  | as_analyze  | --> | as_plan   | --> | as_execute  | --> | as_verify   |
  +-------------+     +-----------+     +-------------+     +-------------+
    currentLoad         action:          newInstanceCount     verified=true
    per service         scale-up/down    5 (up), 2 (down)    newLoad checked

Workers

Analyze -- Takes service and metric (defaults to "cpu"). Returns deterministic currentLoad values by service name: "api-server" = 85%, "web-frontend" = 45%, "batch-worker" = 92%, "cache-service" = 30%, default = 60%. Also computes avgLoad15m (currentLoad - 5), peakLoad1h (currentLoad + 8), and currentInstances = 3.

Plan -- Reads currentLoad as an integer. If >= 80 the action is "scale-up" (from 3 to 5 instances). If <= 30 the action is "scale-down" (from 3 to max(1, 2)). Otherwise "no-change". Outputs a human-readable reason string.

Execute -- Receives the action string. For "scale-up", sets newInstanceCount to 5. For "scale-down", sets it to 2. For anything else, keeps 3 and marks scaled: false. Stamps an executedAt ISO timestamp.

Verify -- Confirms the scaling worked. After "scale-up", reports CPU dropped to 52%. After "scale-down", reports load stable at 45%. No-change reports 50%. Always sets verified: true.

Tests

37 unit tests cover metric analysis, planning thresholds, execution paths, and verification.

Running

See ../../RUNNING.md for setup and execution instructions.