Autopilot Scheduler
Applies an Autopilot preset on a cron schedule, so replicas are ready before a known peak and return to the default after it.
Logins peak at 9:00: HPA takes 30+ seconds to go from 10 to 80 Pods
kubernetes.io, Horizontal Pod Autoscaling; Google Cloud and The Harris Poll, 2019. The 9:00 peak and Pod counts are illustrative.
HPA scales on the load it sees now. When everyone logs in at 9:00, it starts adding Pods only after the load arrives.
With default settings, HPA can at most double the Pods every 15 seconds. From 10 Pods, reaching 80 takes three increases, the last one 30 seconds after the first.
Known peaks still cause outages: 10% of retail executives said their website went down during Black Friday or Cyber Monday the year before (Google Cloud and The Harris Poll, 2019).
Times and Pod counts are illustrative, and the timing is not to scale. HPA scale-up default (100% or 4 Pods per 15s, whichever is more): kubernetes.io. Presets and pre-execution: Autopilot Scheduler docs.
How Wave does it
- Load arriveseveryone logs in at 9:00
- every 15sHPA controllersees the load only after it arrives
- up to 2x / 15sDeployment10 Pods, then 20, 40, 80
- Pod startupPods readyafter the peak has started
- Schedulecron in UTC: 50 8 * * 1-5
- starts earlyWave Coreapplies the Autopilot preset 10 minutes early (pre-execution)
- min 80Kubernetes APIraises the replicas to 80
- before 9:00Deployment80 Pods ready; back to the default after the duration
Choose a preset
An Autopilot preset sets the strategy, thresholds, and min/max replicas, for example min 80 for the morning peak.
Set a cron schedule
Pick the Deployments and a cron expression in UTC. Add a year for a one-time event such as a campaign launch.
Start early, then return
Pre-execution starts the preset minutes before the peak. When the duration ends, the Deployments return to their default config.