Resource pool configuration parameter format
The resource pool configuration parameter accepts multiple resource pool IDs. Separate the IDs with commas.
The following examples show how this works.
Assume you have the following resource pools:
A subscription resource pool with the resource configuration ID `a`
A dedicated resource pool with the resource configuration ID `b`
A shared resource pool with the configuration ID `default`
Example 1: Use only the shared resource pool
If you do not specify this parameter, the value defaults to `default`. Image generation requests are routed only to the shared resource pool.
Example 2: Use only subscription resource pool `a` and dedicated resource pool `b`
Pass a,b. This configuration uses the subscription and dedicated resource pools. Requests are routed to the resource pool with the lowest utilization. If a maximum utilization per instance is configured, an error is returned if both resource pools exceed this limit.
Scenario 3: Subscription resource pool A and dedicated resource pool B have priority. If image generation tasks become severely stacked and exceed the configured maximum utilization limit per instance, a shared resource pool acts as a fallback.
Pass a,b,default. This configuration uses the subscription, dedicated, and shared resource pools, where the shared pool serves as a fallback. Requests are first routed to the private resource pool (subscription or dedicated) with the lowest utilization. If a maximum utilization per instance is configured and both private resource pools exceed this limit, requests fall back to the shared resource pool. Routing to the shared resource pool incurs additional charges based on the inference duration.