Stuck on Architect Script Node failing with 500 error during high concurrency JMeter test

General Discussion
1

Background

We are running a load test scenario using JMeter to validate the performance of our Genesys Cloud Architect flow. The specific flow involves an incoming call being routed to an agent queue, but before the agent accepts, the system must fetch a dynamic script from an external HTTP API. We are using the “Fetch Document” node in the Architect. The goal is to simulate 500 concurrent calls arriving within a 10-second window to see if the script retrieval causes bottlenecks or failures.

The environment is our staging org. We have increased the API rate limits for the specific integration as much as allowed for our tier. The external API is hosted on AWS and can handle the load easily, returning 200 OK with JSON payload in under 50ms under normal conditions.

Issue

When the concurrent load exceeds 100 calls, the Architect flow starts failing at the “Fetch Document” node. The call does not drop, but it falls through to the error handling path. The error message in the Architect trace log shows:

Error fetching document: HTTP 500 Internal Server Error. Timeout after 3000ms.

This is strange because the external API is fast. It seems like Genesys Cloud’s internal service handling the HTTP request is hitting a resource limit or a queue depth issue, causing a timeout even though the downstream service is responsive. We see this happening randomly, not consistently for all calls above 100.

Troubleshooting

  1. Verified the external API logs. No errors there. All requests from Genesys IP range are successful.
  2. Checked Genesys Cloud Admin > System Settings. No obvious restrictions on outbound HTTP requests.
  3. Tried increasing the timeout in the “Fetch Document” node configuration to 5000ms and 10000ms. The error persists, just with a longer wait time before failing.
  4. The JMeter script uses a constant throughput timer to control the load. We verified the IP rotation is not causing issues.

Has anyone seen this specific 500 error on the Fetch Document node during load tests? Is there a hidden limit on how many concurrent HTTP fetches an Architect flow can execute per second? We are just trying to understand if this is a platform limitation or a configuration error on our side.