NetFUNNEL Integration
Turns NetFUNNEL waiting-room segments ON when CPU, memory, or max replicas thresholds are exceeded, and OFF after a cooldown, so users wait in line instead of getting errors.
Pods reach maxReplicas at 10, and new users get errors instead of a place in line
kubernetes.io, HorizontalPodAutoscaler v2 API reference and Horizontal Pod Autoscaling; Ticketmaster, statement on the Eras Tour presale, Nov 19, 2022. The 10 replicas are illustrative, from an example in the Wave NetFUNNEL Integration docs.
HPA cannot go past maxReplicas: "maxReplicas is the upper limit for the number of replicas to which the autoscaler can scale up" (kubernetes.io). Once a Deployment runs 10 of 10 replicas, every extra request goes to the same 10 Pods.
Below that limit, HPA adds replicas at most every 15 seconds (kubernetes.io). By default, nothing holds the users who arrive in between.
At Ticketmaster's 2022 presale, traffic reached 4x the previous peak, and Ticketmaster said it "had to slow down queues to keep them stable".
maxReplicas as the upper limit: kubernetes.io HPA reference. Segments ON and OFF, Max Inflow, and the 60s default cooldown: Wave NetFUNNEL Integration docs. The 10 replicas are illustrative, from a docs example.
How Wave does it
- HPAadds replicas at most every 15s
- scalesDeploymentstops at maxReplicas, 10 of 10
- no queueIngress Gatewayforwards every request; limits default to 2^32-1
- all usersPodstake every request, and users get errors
- Wave Metrics Agentcollects CPU and memory per Deployment
- checksWaveevery set threshold exceeded, such as replicas at max
- turns ONNetFUNNEL segmentusers wait and see their position; up to Max Inflow get in
- while users waitAutopilotscales the Pods if it is on; segments OFF after the cooldown, 60s by default
Connect your NetFUNNEL account
Enter the tenant ID and organization ID in Settings > NetFUNNEL in the Wave web console, or apply a WaveNetfunnelConnection. NetFUNNEL SaaS is the only supported type today.
Map a Deployment to segments
Pick the NetFUNNEL segments for a Deployment, an Argo Rollout, or a DeploymentConfig, and set CPU, memory, or max replicas thresholds and a cooldown. Wave Metrics Agent collects the CPU and memory data.
Wave turns the queue ON and OFF
When every set threshold is exceeded, Wave turns all mapped segments ON. Each segment lets in up to its Max Inflow, and the other users wait and see their position. After the metrics drop and the cooldown ends (60 seconds by default), Wave turns the segments OFF and logs each change.