Trying to understand why Architect flows are failing under specific load conditions. The setup involves simulating high-volume inbound calls via the Telephony API. We are using JMeter 5.6 for this test. The target environment is Genesys Cloud US1.
The goal is to validate call capacity planning. We want to see how the system handles concurrent WebSocket connections. The test ramps up to 500 concurrent users. Each user initiates a call and enters a basic IVR menu. The menu has three simple options. No complex logic is involved. No external API calls are made.
At around 300 concurrent sessions, errors start appearing. The error code is 504 Gateway Timeout. This happens during the initial connection phase. The WebSocket handshake completes successfully. The first prompt plays correctly. The failure occurs when sending DTMF inputs. The input is not registered by the flow.
We checked the Architect flow logs. The flow instance is created. The state transitions are recorded. However, the input event is missing from the log. This suggests a network issue. Or perhaps a rate limiting issue on the WebSocket channel.
We have reviewed the API rate limits documentation. The limits seem adequate for this volume. The WebSocket connection limit is not reached. The throughput is well within expected bounds.
The JMeter script uses constant throughput timer. The delay between inputs is 2 seconds. This should be sufficient. The issue persists across multiple test runs.
We are looking for guidance. Is there a known limitation with DTMF processing at scale? Are there specific headers required for high concurrency? We need to ensure the IVR remains stable. Any insights would be appreciated. We want to avoid false positives in our performance audit. The current results are inconclusive. We need to isolate the root cause. Is this a configuration error? Or a platform constraint?