Stuck on a weird latency issue. The queue wait time jumps to 45 seconds when load hits 500 agents, but API response time stays under 200ms. Is there a hidden capacity limit for predictive routing calculations in the free tier?
If I remember correctly, predictive routing latency at that scale is rarely about API response times. It is usually the internal calculation engine struggling with the state changes. The 45-second wait suggests the system is recalculating scores for every agent status change.
Check your queue configuration. If you have dynamic skills or complex wrap-up codes, the engine works harder. Try simplifying the routing strategy for the test. Also, ensure your JMeter script is not sending excessive status updates.
Here is a minimal payload to check your current queue settings:
{
"name": "Test_Queue",
"routingType": "PREDICTIVE",
"skills": ["general_support"],
"overflowQueue": null,
"utilizationThreshold": 0.85
}
If the latency persists, look at the agent availability transitions. A rapid flip between available and unavailable can trigger re-routes. The documentation suggests keeping the utilization threshold stable during load tests. This usually happens because of the state machine processing, not the API itself.