I can’t seem to figure out why architect flows are dropping calls when we ramp up concurrency in our load tests. we are using jmeter 5.6 in the us1 region to simulate inbound voice traffic. the setup is simple: 100 concurrent threads hitting a basic ivr flow with a single gather input block.
at 20 concurrent calls, everything works fine. but when we push to 50 or 100 concurrent calls, we start seeing immediate hangups. the jmeter sampler returns a 200 ok, but the call duration is near zero. checking the architect logs, we see the call enters the flow but disappears before reaching the next block. no error code in the architect trace, just a gap.
we are not hitting any obvious rate limits on the api side since this is rtcp/websocket traffic for the voice leg, not rest api calls. is there a hidden limit on active architect sessions per org or per flow? or is this a capacity issue with the media servers in us1? we have checked our license count and we are well within limits. any help would be appreciated.