Revert "Remove watchdog failure counter and fail on the first try (#584…

…)" (#586)

This reverts commit 0fab7de.

bellows/config/__init__.py

@@ -22,6 +22,7 @@
 CONF_USE_THREAD = "use_thread"
 CONF_EZSP_CONFIG = "ezsp_config"
 CONF_EZSP_POLICIES = "ezsp_policies"
+CONF_PARAM_MAX_WATCHDOG_FAILURES = "max_watchdog_failures"
 CONF_FLOW_CONTROL = "flow_control"
 CONF_FLOW_CONTROL_DEFAULT = "software"
 
@@ -37,6 +38,7 @@
 CONFIG_SCHEMA = CONFIG_SCHEMA.extend(
     {
         vol.Required(CONF_DEVICE): SCHEMA_DEVICE,
+        vol.Optional(CONF_PARAM_MAX_WATCHDOG_FAILURES, default=4): int,
         vol.Optional(CONF_EZSP_CONFIG, default={}): dict,
         vol.Optional(CONF_EZSP_POLICIES, default={}): vol.Schema(
             {vol.Optional(str): int}

bellows/zigbee/application.py

@@ -27,6 +27,7 @@
 from bellows.config import (
     CONF_EZSP_CONFIG,
     CONF_EZSP_POLICIES,
+    CONF_PARAM_MAX_WATCHDOG_FAILURES,
     CONF_USE_THREAD,
     CONFIG_SCHEMA,
     SCHEMA_DEVICE,
@@ -921,9 +922,9 @@ def _handle_id_conflict(self, nwk: t.EmberNodeId) -> None:
     async def _watchdog(self):
         """Watchdog handler."""
         LOGGER.debug("Starting EZSP watchdog")
+        failures = 0
         read_counter = 0
         await asyncio.sleep(WATCHDOG_WAKE_PERIOD)
-
         while True:
             try:
                 async with asyncio_timeout(WATCHDOG_WAKE_PERIOD * 2):
@@ -954,9 +955,13 @@ async def _watchdog(self):
                         cnt._last_reset_value = 0
 
                     LOGGER.debug("%s", counters)
+
+                failures = 0
             except (asyncio.TimeoutError, EzspError) as exc:
                 LOGGER.warning("Watchdog heartbeat timeout: %s", repr(exc))
-                break
+                failures += 1
+                if failures > self.config[CONF_PARAM_MAX_WATCHDOG_FAILURES]:
+                    break
             except asyncio.CancelledError:
                 raise
             except Exception as exc:
@@ -968,7 +973,7 @@ async def _watchdog(self):
             await asyncio.sleep(WATCHDOG_WAKE_PERIOD)
 
         self.state.counters[COUNTERS_CTRL][COUNTER_WATCHDOG].increment()
-        self._handle_reset_request("Watchdog timeout")
+        self._handle_reset_request(f"Watchdog timeout. Heartbeat timeouts: {failures}")
 
     async def _get_free_buffers(self) -> int | None:
         status, value = await self._ezsp.getValue(

tests/test_application.py

@@ -1134,22 +1134,31 @@ async def test_watchdog(app, monkeypatch, ezsp_version):
 
     monkeypatch.setattr(application, "WATCHDOG_WAKE_PERIOD", 0.01)
     monkeypatch.setattr(application, "EZSP_COUNTERS_CLEAR_IN_WATCHDOG_PERIODS", 2)
+    nop_success = 7
     app._ezsp.ezsp_version = ezsp_version
 
-    app._ezsp.nop = AsyncMock(side_effect=EzspError())
-    app._ezsp.readCounters = AsyncMock(side_effect=EzspError())
-    app._ezsp.readAndClearCounters = AsyncMock(side_effect=EzspError())
+    async def nop_mock():
+        nonlocal nop_success
+        if nop_success:
+            nop_success -= 1
+            if nop_success % 3:
+                raise EzspError
+            else:
+                return ([0] * 10,)
+        raise asyncio.TimeoutError
+
+    app._ezsp.nop = AsyncMock(side_effect=nop_mock)
+    app._ezsp.readCounters = AsyncMock(side_effect=nop_mock)
+    app._ezsp.readAndClearCounters = AsyncMock(side_effect=nop_mock)
     app._handle_reset_request = MagicMock()
     app._ctrl_event.set()
 
     await app._watchdog()
 
     if ezsp_version == 4:
-        assert app._ezsp.nop.await_count == 1
-        assert app._ezsp.readCounters.await_count == 0
+        assert app._ezsp.nop.await_count > 4
     else:
-        assert app._ezsp.nop.await_count == 0
-        assert app._ezsp.readCounters.await_count == 1
+        assert app._ezsp.readCounters.await_count >= 4
 
     assert app._handle_reset_request.call_count == 1