fix(telegram): set explicit timeouts on HTTPXRequest with fallback transport

gateway/platforms/telegram.py

@@ -8,6 +8,7 @@
 """
 
 import asyncio
+import httpx
 import json
 import logging
 import os
@@ -178,6 +179,27 @@ def _looks_like_network_error(error: Exception) -> bool:
             pass
         return isinstance(error, OSError)
 
+    @staticmethod
+    def _is_retryable_send_network_error(error: Exception) -> bool:
+        """Retry only when the request clearly never reached Telegram.
+
+        PTB wraps underlying httpx failures in higher-level telegram errors.
+        Connect/pool failures are safe to retry because no request was
+        delivered. Read/write timeouts are ambiguous and can duplicate sends.
+        """
+        if "Request was *not* sent to Telegram" in str(error):
+            return True
+
+        current: BaseException | None = error
+        seen: set[int] = set()
+        while current is not None and id(current) not in seen:
+            seen.add(id(current))
+            if isinstance(current, (httpx.ConnectError, httpx.ConnectTimeout, httpx.PoolTimeout)):
+                return True
+            current = current.__cause__ or current.__context__
+
+        return False
+
     async def _handle_polling_network_error(self, error: Exception) -> None:
         """Reconnect polling after a transient network interruption.
 
@@ -520,8 +542,28 @@ async def connect(self) -> bool:
                     ", ".join(fallback_ips),
                 )
                 transport = TelegramFallbackTransport(fallback_ips)
-                request = HTTPXRequest(httpx_kwargs={"transport": transport})
-                get_updates_request = HTTPXRequest(httpx_kwargs={"transport": transport})
+                # Explicit timeouts are required here because passing a
+                # pre-built HTTPXRequest to ApplicationBuilder bypasses
+                # the builder's default timeout configuration. The
+                # fallback transport adds latency (DoH discovery + IP
+                # rotation), so these values must be generous enough to
+                # absorb that overhead. Without them the library's 5s
+                # defaults cause frequent TimedOut errors which the
+                # retry logic then compounds into duplicate deliveries.
+                request = HTTPXRequest(
+                    read_timeout=20.0,
+                    write_timeout=20.0,
+                    connect_timeout=10.0,
+                    pool_timeout=5.0,
+                    httpx_kwargs={"transport": transport},
+                )
+                get_updates_request = HTTPXRequest(
+                    read_timeout=30.0,
+                    write_timeout=5.0,
+                    connect_timeout=10.0,
+                    pool_timeout=5.0,
+                    httpx_kwargs={"transport": transport},
+                )
                 builder = builder.request(request).get_updates_request(get_updates_request)
             self._app = builder.build()
             self._bot = self._app.bot
@@ -820,6 +862,8 @@ async def send(
                                 continue
                             # Other BadRequest errors are permanent — don't retry
                             raise
+                        if not self._is_retryable_send_network_error(send_err):
+                            raise
                         if _send_attempt < 2:
                             wait = 2 ** _send_attempt
                             logger.warning("[%s] Network error on send (attempt %d/3), retrying in %ds: %s",

tests/gateway/test_telegram_thread_fallback.py

@@ -14,6 +14,7 @@
 import types
 from types import SimpleNamespace
 
+import httpx
 import pytest
 
 from gateway.config import PlatformConfig, Platform
@@ -33,11 +34,16 @@ class FakeBadRequest(FakeNetworkError):
     pass
 
 
+class FakeTimedOut(FakeNetworkError):
+    pass
+
+
 # Build a fake telegram module tree so the adapter's internal imports work
 _fake_telegram = types.ModuleType("telegram")
 _fake_telegram_error = types.ModuleType("telegram.error")
 _fake_telegram_error.NetworkError = FakeNetworkError
 _fake_telegram_error.BadRequest = FakeBadRequest
+_fake_telegram_error.TimedOut = FakeTimedOut
 _fake_telegram.error = _fake_telegram_error
 _fake_telegram_constants = types.ModuleType("telegram.constants")
 _fake_telegram_constants.ParseMode = SimpleNamespace(MARKDOWN_V2="MarkdownV2")
@@ -146,15 +152,17 @@ async def mock_send_message(**kwargs):
 
 @pytest.mark.asyncio
 async def test_send_retries_network_errors_normally():
-    """Real transient network errors (not BadRequest) should still be retried."""
+    """Connect-stage network errors should still be retried."""
     adapter = _make_adapter()
 
     attempt = [0]
 
     async def mock_send_message(**kwargs):
         attempt[0] += 1
         if attempt[0] < 3:
-            raise FakeNetworkError("Connection reset")
+            raise FakeNetworkError("httpx.ConnectError: Connection reset") from httpx.ConnectError(
+                "Connection reset"
+            )
         return SimpleNamespace(message_id=200)
 
     adapter._bot = SimpleNamespace(send_message=mock_send_message)
@@ -168,6 +176,31 @@ async def mock_send_message(**kwargs):
     assert attempt[0] == 3  # Two retries then success
 
 
+@pytest.mark.asyncio
+async def test_send_does_not_retry_ambiguous_timeouts():
+    """Read/write timeouts may happen after delivery starts, so don't resend."""
+    adapter = _make_adapter()
+
+    attempt = [0]
+
+    async def mock_send_message(**kwargs):
+        attempt[0] += 1
+        raise FakeTimedOut("Timed out waiting for Telegram response") from httpx.ReadTimeout(
+            "read timeout"
+        )
+
+    adapter._bot = SimpleNamespace(send_message=mock_send_message)
+
+    result = await adapter.send(
+        chat_id="123",
+        content="test message",
+    )
+
+    assert result.success is False
+    assert "Timed out waiting for Telegram response" in result.error
+    assert attempt[0] == 1
+
+
 @pytest.mark.asyncio
 async def test_thread_fallback_only_fires_once():
     """After clearing thread_id, subsequent chunks should also use None."""