abort(): register live socket at urllib3 request() so Force Stop wakes recv() even while waiting for response headers (was only after headers via active_response)

agentmain.py

@@ -116,12 +116,10 @@ class GenericAgent:
         if self.handler is not None: self.handler.code_stop_signal.append(1)
         for sess in getattr(self.llmclient.backend, '_sessions', [self.llmclient.backend]):
             sess.should_stop = lambda: self.stop_sig  # live read; cleared by run()'s finally
-            try:  # wake a recv() blocked in another thread. Verified on Windows: shutdown()/close() do NOT
-                  # wake it (makefile refcount defers real closesocket); _real_close() does -> ChunkedEncodingError
+            try:  # wake a recv() blocked in another thread (waiting headers OR reading stream). Verified on Windows:
+                  # shutdown()/close() do NOT wake it (makefile refcount defers closesocket); _real_close() does
                 import socket as _socket
-                raw = sess.active_response.raw
-                fp = getattr(getattr(raw, '_fp', None), 'fp', None)  # http.client response -> buffered socket file
-                sock = fp.raw._sock if fp else raw.connection.sock   # SocketIO._sock (SSL-wrapped OK); fallback urllib3 conn
+                sock = sys.modules['llmcore']._INFLIGHT[sess._tid]  # socket registered at urllib3 request() time, before headers
                 try: sock.shutdown(_socket.SHUT_RDWR)  # for non-Windows semantics
                 except OSError: pass
                 try: sock._real_close()  # CPython internal; bypasses refcount -> actual closesocket

llmcore.py

@@ -1,6 +1,11 @@
 import os, json, re, time, requests, sys, threading, urllib3, base64, importlib, uuid, pathlib, copy
 from datetime import datetime
 urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
+_INFLIGHT = {}  # thread ident -> live socket; lets abort() close it even before response headers arrive
+_orig_conn_request = urllib3.connection.HTTPConnection.request
+def _conn_request_hook(self, *a, **k):  # after request() the socket is connected+sent; conn.sock may later be None'd by http.client
+    r = _orig_conn_request(self, *a, **k); _INFLIGHT[threading.get_ident()] = self.sock; return r
+urllib3.connection.HTTPConnection.request = _conn_request_hook
 _RESP_CACHE_KEY = str(uuid.uuid4()); _RESP_CODEX_KEY = str(uuid.uuid4())
 _ROOT = os.path.dirname(os.path.abspath(__file__))
 if _ROOT not in sys.path: sys.path.append(_ROOT)
@@ -459,6 +464,7 @@ def _stream_with_retry(sess, url, headers, payload, parse_fn):
         STATS.update(t_start=time.time(), t_ttft=None)
         if not sess.stream: STATS['t_ttft'] = STATS['t_start']
         try:
+            sess._tid = threading.get_ident()  # abort() looks up _INFLIGHT[_tid]
             with requests.post(url, headers=headers, json=payload, stream=sess.stream, 
                                timeout=(sess.connect_timeout, sess.read_timeout), proxies=sess.proxies, verify=sess.verify) as r:
                 sess.active_response = r