Auto-Scaling

How It Works

RaidFrame monitors your service metrics and automatically adjusts the number of running instances. New instances spin up in under 10 seconds. Scale-down happens after a configurable cooldown to prevent flapping.

Scaling Triggers

CPU-Based (Default)

Scale when average CPU across instances exceeds a threshold:

services:
  api:
    scaling:
      min: 2
      max: 20
      target_cpu: 70

Memory-Based

scaling:
  target_memory: 80

Request-Based

Scale based on requests per second per instance:

scaling:
  target_rps: 500

Queue-Based (Workers)

Scale workers based on queue depth:

services:
  processor:
    type: worker
    scaling:
      min: 1
      max: 50
      target_queue_depth: 100

When the queue has 500 items and target_queue_depth is 100, RaidFrame scales to 5 instances.

Custom Metrics

Push your own metrics and scale on them:

rf metrics push api.active_connections 1500

scaling:
  custom_metric: api.active_connections
  target_value: 200

Scheduled Scaling

Pre-scale for known traffic patterns:

scaling:
  min: 2
  max: 20
  schedule:
    - cron: "0 9 * * MON-FRI"
      min: 10
      comment: "Business hours"
    - cron: "0 18 * * MON-FRI"
      min: 2
      comment: "After hours"
    - cron: "0 0 25 12 *"
      min: 30
      comment: "Christmas sale"

Scale to Zero

Idle services can scale to zero to save costs. The first request triggers a cold start:

scaling:
  min: 0
  max: 10
  scale_to_zero:
    idle_timeout: 15m
    cold_start_target: 3s

RaidFrame keeps a warm standby image to minimize cold start time.

CLI Commands

# View current scaling status
rf services info api

# Manually scale
rf services scale api --min 5 --max 30

# Temporarily override auto-scaling
rf services scale api --fixed 10 --duration 2h

# View scaling events
rf metrics --service api --type scaling

SCALING EVENTS (last 24h)
─────────────────────────────
14:23  api  2 → 4 instances  (cpu: 82%)
14:45  api  4 → 6 instances  (cpu: 76%)
15:30  api  6 → 4 instances  (cooldown, cpu: 35%)
18:00  api  4 → 2 instances  (scheduled: after hours)

Scaling Configuration