snarky-puppy / node-socket-hang-up-bug

Reproduction of the node bug "socket hang up"

Bug Summary: Axios Connection Pooling Misconception

Issue

Axios relies on nodes HTTP/HTTPS agent for connection pooling rather than implementing its own. Consequently, Axios "instances" are not truly isolated; they serve primarily to set default configurations for HTTP requests. This misunderstanding leads to unexpected behavior when different timeouts are set for requests on what is assumed to be isolated instances.

Trace Analysis

The TCP trace shows that with HTTP keep-alive, requests GET /1 and GET /wait are transmitted over a single TCP connection. The initial request is configured with a 1-second timeout, followed by a request with no timeout. The re-use of the connection for both requests causes the second request to be prematurely terminated at the 1-second mark, despite the server still processing the request.

Recommendation

Utilise a consistent Axios instance for all requests to a specific server. This instance should be defined once, preferably at the module level, to avoid the repeated instantiation of Axios instances that can lead to unexpected timeouts due to shared underlying connections.