Does anyone understand why the bot architect flow is dropping connections when i ramp up concurrent users? we are running a load test using jmeter to simulate 200 concurrent digital channel sessions hitting our ai agent. the setup is pretty standard. we have a simple intent classification step followed by a data retrieval step. the issue happens right after the intent is matched. the system returns a 504 gateway timeout. this is not a network issue on our side. the websocket connection stays open but no response comes back from the bot engine.
i checked the api logs and see that the initial handshake works fine. but once the payload is sent for processing, the timeout occurs. we are using the latest version of the architect tool. the environment is us-east-1. i noticed that if i reduce the concurrency to 50 users, the flow completes successfully without any errors. but at 100+ users, the error rate spikes to nearly 40%.
is there a known limit on concurrent bot executions per organization? i read something about rate limits for api calls but this seems like a processing limit inside the bot engine itself. we are not doing any heavy external api calls in the flow. just simple variable assignments. the latency between the websocket open and the first message is about 200ms. but after the intent match, it hangs for 30 seconds then fails.
i tried increasing the timeout in jmeter but that did not help. the 504 comes from the genesys cloud side. i also checked the websocket connection limits. we have plenty of capacity there. the problem seems specific to the bot processing queue. can anyone share their jmeter configs for testing bot flows? maybe i am missing a header or a keep-alive setting. any help would be great. we need to validate the system can handle our peak volume before going live next week.