Keep gateways alive through transient UDP policy changes

CHANGELOG.md

@@ -1,3 +1,17 @@
+## PyTAK 7.5.1
+
+- Treats a transient `EPERM` from a network send as a recoverable transport
+  outage. Linux can return this while firewall, route, or network namespace
+  state is being replaced during an appliance update. PyTAK now tears down the
+  affected client cleanly and reconnects with bounded backoff instead of
+  terminating the gateway process.
+
+## PyTAK 7.5.0
+
+- Added a common, additive runtime health contract to `StatusWriter`: `health`,
+  `input`, and `output` blocks describe process state, receiver activity, and
+  egress connection health without removing gateway-specific counters.
+
 ## PyTAK 7.4.3
 
 - PyTAK-backed gateways now survive transient TAK transport failures in the
@@ -230,8 +244,3 @@ New Features:
 
 Bug & Performance Fixes:
 - Added async sleeps to each TX & RX loops iteration to fix broken async regiment in PYTAK.
-# 7.5.0
-
-* Add a common, additive runtime health contract to `StatusWriter`: `health`,
-  `input`, and `output` blocks describe process state, receiver activity, and
-  egress connection health without removing gateway-specific counters.

src/pytak/VERSION

@@ -1 +1 @@
-7.5.0
+7.5.1

src/pytak/client_functions.py

@@ -1048,6 +1048,11 @@ def _retryable_transport_error(exc: Exception) -> bool:
             errno.EHOSTUNREACH,
             errno.ENETDOWN,
             errno.ENETUNREACH,
+            # Linux can report EPERM from UDP sendto() while nftables,
+            # routes, or network namespaces are being replaced. Treat it as
+            # a transport outage so an appliance update cannot terminate a
+            # high-rate feeder during that short transition.
+            errno.EPERM,
             errno.ETIMEDOUT,
         }
         if hasattr(errno, "EHOSTDOWN"):

tests/test_client_functions.py

@@ -20,6 +20,7 @@
 
 
 import asyncio
+import errno
 import os
 
 from configparser import ConfigParser, SectionProxy
@@ -564,6 +565,42 @@ async def test_run_with_reconnect_keeps_local_os_errors_fatal():
     fake_sleep.assert_not_called()
 
 
+def test_reconnect_keeps_local_access_denied_fatal():
+    """Ordinary local file access failures remain configuration errors."""
+    error = PermissionError(errno.EACCES, "Permission denied")
+    assert not pytak.client_functions._retryable_transport_error(error)
+
+
+@pytest.mark.asyncio
+async def test_run_with_reconnect_retries_transient_udp_permission_error():
+    """A firewall transition must not terminate a UDP-backed gateway."""
+    config_p = ConfigParser(
+        {
+            "PYTAK_RECONNECT_INITIAL": "1",
+            "PYTAK_RECONNECT_MAX": "2",
+            "PYTAK_RECONNECT_FACTOR": "2",
+            "PYTAK_RECONNECT_JITTER": "0",
+            "PYTAK_RECONNECT_RESET": "300",
+        }
+    )
+    config_p.add_section("fakeapp")
+    config = config_p["fakeapp"]
+    fake_main = AsyncMock(
+        side_effect=[PermissionError(errno.EPERM, "Operation not permitted"), None]
+    )
+    fake_sleep = AsyncMock()
+
+    with mock.patch(
+        "pytak.client_functions.main", new=fake_main
+    ), mock.patch("pytak.client_functions.asyncio.sleep", new=fake_sleep):
+        await pytak.client_functions.run_with_reconnect(
+            "fakeapp", config, config_p
+        )
+
+    assert fake_main.call_count == 2
+    fake_sleep.assert_called_once_with(1)
+
+
 @pytest.mark.asyncio
 async def test_run_with_reconnect_retries_tak_enrollment_resolution():
     """A transient enrollment failure retries from the original deep link."""