When executing a long-running REST API call (exceeding 15 minutes) via the Integration.REST agent, the job fails with the following error even if the responseTimeout is set to a higher value:
java.net.SocketTimeoutException: Read timed out
This behavior is common when the backend service (e.g., a Cloud Function or API gateway) takes several minutes to process data, during which time no packets are transmitted over the socket.
A stateful network device between the agent and the endpoint silently drops the idle TCP connection when its internal idle timeout is reached. The agent remains unaware the connection was closed and eventually triggers its own internal socket timeout, resulting in a Read timed out exception.
This issue is resolved by hardening the agent to use TCP Keep-Alive probes, which prevent network devices from marking the connection as idle via a new build.
This will be fixed in Integration.REST Agent 24.2.1 (Not released yet)