RunPod
Uses the runpod SDK where it covers the call and REST otherwise. Creates a template (image, env, ports) and then a load-balancing endpoint.
- provider: runpod-eu type: runpod priority: 3 runpod: {dataCenters: [EU-RO-1, EU-SE-1]}Credentials: RUNPOD_API_KEY.
Translation
Section titled “Translation”| Spec field | RunPod |
|---|---|
container.image |
template imageName |
container.command, port |
template dockerStartCmd, LB endpoint port |
resources.gpu |
gpuIds list |
scaling.replicas |
workersMin, workersMax |
scaling.concurrency |
LB endpoint, scalerType: REQUEST_COUNT |
container.health |
/ping semantics on PORT_HEALTH: 204 warming, 200 ready |
container.secrets |
template env |
| endpoint | https://api.runpod.ai/v2/<id>/ |
The router treats a 204 probe response as warming, not down. Ref stored in state: template id plus endpoint id. Fallbacks stay warm because fallbackScaleToZero defaults to false, so workersMin is never 0 unless you opt in.