Skip to content

scheduling.md

Latest commit

 

History

History
188 lines (143 loc) · 5.04 KB

scheduling.md

File metadata and controls

188 lines (143 loc) · 5.04 KB
Raw

Bot Scheduling

How to auto-scale your bot instance on a time-based schedule using KEDA. This lets you run the bot only during business hours and save compute costs overnight and on weekends.

KEDA Cron Scaler

Scale your bot to 1 replica during a configured time window and 0 outside it. Uses KEDA (already installed on the AI cluster).

What it does

  • Scales the bot Deployment to desiredReplicas during the configured window
  • Outside the window, scales to minReplicaCount: 0 — the pod is completely stopped
  • No compute costs outside working hours
  • KEDA takes ownership of replica count — BOT_REPLICAS in the template becomes the initial value before KEDA kicks in

Adding to your deploy template

Add a ScaledObject resource to your deploy/template.yaml, after the NetworkPolicy:

# --- Cron Scaler ---
- apiVersion: keda.sh/v1alpha1
  kind: ScaledObject
  metadata:
    name: ${BOT_NAME}-cron-scaler
    labels:
      app.kubernetes.io/name: ${BOT_NAME}
      app.kubernetes.io/part-of: devbot
  spec:
    scaleTargetRef:
      apiVersion: apps/v1
      kind: Deployment
      name: ${BOT_NAME}
    minReplicaCount: 0
    maxReplicaCount: 1
    triggers:
    - type: cron
      metadata:
        timezone: "Europe/Prague"
        start: "0 9 * * 1-5"
        end: "0 23 * * 1-5"
        desiredReplicas: "1"

The triggers section is where you define when the bot runs. Each trigger has:

Field Description
timezone IANA timezone (e.g. Europe/Prague, America/New_York, US/Eastern)
start Cron expression — when to scale up
end Cron expression — when to scale down
desiredReplicas How many replicas during the window (always "1" for bot instances)

The cron format is standard 5-field: minute hour day-of-month month day-of-week where day-of-week is 0=Sunday through 6=Saturday.

Schedule examples

Business hours, Monday–Friday

The framework instance uses this — bot runs 9:00–23:00 Prague time on weekdays:

triggers:
- type: cron
  metadata:
    timezone: "Europe/Prague"
    start: "0 9 * * 1-5"
    end: "0 23 * * 1-5"
    desiredReplicas: "1"

US East Coast working hours

9am–6pm ET, weekdays only:

triggers:
- type: cron
  metadata:
    timezone: "America/New_York"
    start: "0 9 * * 1-5"
    end: "0 18 * * 1-5"
    desiredReplicas: "1"

24/7 weekdays, off on weekends

Run all day Monday through Friday, off Saturday and Sunday:

triggers:
- type: cron
  metadata:
    timezone: "UTC"
    start: "0 0 * * 1"
    end: "0 0 * * 6"
    desiredReplicas: "1"

Mondays only

Bot runs only on Mondays, 8am–5pm:

triggers:
- type: cron
  metadata:
    timezone: "America/New_York"
    start: "0 8 * * 1"
    end: "0 17 * * 1"
    desiredReplicas: "1"

Weekdays with reduced weekend hours

Full hours on weekdays, shorter window on weekends. Use multiple triggers — if any trigger's window is active, the bot scales up:

triggers:
- type: cron
  metadata:
    timezone: "America/New_York"
    start: "0 8 * * 1-5"
    end: "0 18 * * 1-5"
    desiredReplicas: "1"
- type: cron
  metadata:
    timezone: "America/New_York"
    start: "0 10 * * 0,6"
    end: "0 14 * * 0,6"
    desiredReplicas: "1"

Morning and afternoon blocks (skip lunch)

Two separate windows per day:

triggers:
- type: cron
  metadata:
    timezone: "Europe/Prague"
    start: "0 8 * * 1-5"
    end: "0 12 * * 1-5"
    desiredReplicas: "1"
- type: cron
  metadata:
    timezone: "Europe/Prague"
    start: "0 13 * * 1-5"
    end: "0 18 * * 1-5"
    desiredReplicas: "1"

How multiple triggers work

KEDA evaluates all triggers independently. If any trigger's window is active, the bot scales up. The bot only scales to 0 when no trigger is active. This means triggers are effectively OR'd together — you can combine them to build complex schedules.

Timezone and DST

KEDA cron triggers use a single IANA timezone per trigger. DST transitions are handled automatically within that timezone — if you set Europe/Prague, the schedule shifts with CET/CEST automatically.

If your team spans multiple timezones, pick the primary one and document the effective hours for others. Different triggers can use different timezones if needed.

App-interface requirements

Your SaaS file (deploy.yml) must include ScaledObject.keda.sh in managedResourceTypes:

managedResourceTypes:
- Deployment
- NetworkPolicy
- ScaledObject.keda.sh      # required for KEDA

Without this, app-interface will prune the ScaledObject on the next sync.

The namespace file (namespaces/*.yml) must also allow it — either managedResourceTypes: [] (allow all, which is the default) or explicitly list ScaledObject.keda.sh.

Reference