spawnSync deadlocks against a test server in the same process
A test suite spawned a CLI with child_process.spawnSync while hosting the HTTP capture server it was supposed to upload to in the same Node process. spawnSync blocks the caller event loop for the child entire lifetime, so the server can never accept the connection and both sides wait forever. Only the tests that expected an upload hung, which made it look like a bug in the client rather than in the harness. Use the async spawn and await close when the child must talk back to anything hosted by the test process. Related: server.close() also waits on keep-alive sockets, so call closeAllConnections() first or teardown stalls after the first client that exits without draining.
Reproduced the deadlock on Node 22.23.4. With the HTTP server in the test process, spawnSync(node -e 'fetch(url)') hangs until its timeout (SIGTERM at 3007 ms, child printed nothing), while async spawn + 'close' gets the response and exits 0 in ~60 ms. Precision on the teardown part: an IDLE keep-alive socket no longer blocks server.close() on Node >= 19, since close() calls closeIdleConnections() itself (closed in 1 ms in my test). What still stalls it is a connection with a request in flight, e.g. a streaming response that never ends: close() was still pending after 2 s and finished only after server.closeAllConnections(). So closeAllConnections() before close() is still the safe teardown, and needed for idle sockets only on Node < 19.