From 190894010c3e7b2b7c8377ecd48fccc4fa407cc6 Mon Sep 17 00:00:00 2001
From: Stefan Agner <stefan@agner.ch>
Date: Wed, 31 Jan 2024 17:41:21 +0100
Subject: [PATCH] Reset failed API call counter on successful API call (#4862)

* Reset failed API call counter on successful API call

Make sure to reset the failed API call counter after a successful
API call. While at it also update the log messages a bit to make it
clearer what the problem is exactly.

* Address pytest changes
---
 supervisor/misc/tasks.py | 23 ++++++++++++++---------
 tests/misc/test_tasks.py | 18 ++++++++++++------
 2 files changed, 26 insertions(+), 15 deletions(-)

diff --git a/supervisor/misc/tasks.py b/supervisor/misc/tasks.py
index f2a17204a..af8558b75 100644
--- a/supervisor/misc/tasks.py
+++ b/supervisor/misc/tasks.py
@@ -16,8 +16,9 @@ from ..utils.sentry import capture_exception
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
-HASS_WATCHDOG_API = "HASS_WATCHDOG_API"
+HASS_WATCHDOG_API_FAILURES = "HASS_WATCHDOG_API_FAILURES"
 HASS_WATCHDOG_REANIMATE_FAILURES = "HASS_WATCHDOG_REANIMATE_FAILURES"
+HASS_WATCHDOG_MAX_API_ATTEMPTS = 2
 HASS_WATCHDOG_MAX_REANIMATE_ATTEMPTS = 5
 
 RUN_UPDATE_SUPERVISOR = 29100
@@ -169,6 +170,7 @@ class Tasks(CoreSysAttributes):
         if await self.sys_homeassistant.api.check_api_state():
             # Home Assistant is running properly
             self._cache[HASS_WATCHDOG_REANIMATE_FAILURES] = 0
+            self._cache[HASS_WATCHDOG_API_FAILURES] = 0
             return
 
         # Give up after 5 reanimation failures in a row. Supervisor cannot fix this issue.
@@ -176,23 +178,26 @@ class Tasks(CoreSysAttributes):
         if reanimate_fails >= HASS_WATCHDOG_MAX_REANIMATE_ATTEMPTS:
             if reanimate_fails == HASS_WATCHDOG_MAX_REANIMATE_ATTEMPTS:
                 _LOGGER.critical(
-                    "Watchdog cannot reanimate Home Assistant, failed all %s attempts.",
+                    "Watchdog cannot reanimate Home Assistant Core, failed all %s attempts.",
                     reanimate_fails,
                 )
                 self._cache[HASS_WATCHDOG_REANIMATE_FAILURES] += 1
             return
 
         # Init cache data
-        retry_scan = self._cache.get(HASS_WATCHDOG_API, 0)
+        api_fails = self._cache.get(HASS_WATCHDOG_API_FAILURES, 0)
 
         # Look like we run into a problem
-        retry_scan += 1
-        if retry_scan == 1:
-            self._cache[HASS_WATCHDOG_API] = retry_scan
-            _LOGGER.warning("Watchdog miss API response from Home Assistant")
+        api_fails += 1
+        if api_fails < HASS_WATCHDOG_MAX_API_ATTEMPTS:
+            self._cache[HASS_WATCHDOG_API_FAILURES] = api_fails
+            _LOGGER.warning("Watchdog missed an Home Assistant Core API response.")
             return
 
-        _LOGGER.error("Watchdog found a problem with Home Assistant API!")
+        _LOGGER.error(
+            "Watchdog missed %s Home Assistant Core API responses in a row. Restarting Home Assistant Core API!",
+            HASS_WATCHDOG_MAX_API_ATTEMPTS,
+        )
         try:
             await self.sys_homeassistant.core.restart()
         except HomeAssistantError as err:
@@ -203,7 +208,7 @@ class Tasks(CoreSysAttributes):
         else:
             self._cache[HASS_WATCHDOG_REANIMATE_FAILURES] = 0
         finally:
-            self._cache[HASS_WATCHDOG_API] = 0
+            self._cache[HASS_WATCHDOG_API_FAILURES] = 0
 
     @Job(name="tasks_update_cli", conditions=PLUGIN_AUTO_UPDATE_CONDITIONS)
     async def _update_cli(self):
diff --git a/tests/misc/test_tasks.py b/tests/misc/test_tasks.py
index feff83403..c29d2e617 100644
--- a/tests/misc/test_tasks.py
+++ b/tests/misc/test_tasks.py
@@ -34,15 +34,21 @@ async def test_watchdog_homeassistant_api(
         await tasks._watchdog_homeassistant_api()
 
         restart.assert_not_called()
-        assert "Watchdog miss API response from Home Assistant" in caplog.text
-        assert "Watchdog found a problem with Home Assistant API!" not in caplog.text
+        assert "Watchdog missed an Home Assistant Core API response." in caplog.text
+        assert (
+            "Watchdog missed 2 Home Assistant Core API responses in a row. Restarting Home Assistant Core API!"
+            not in caplog.text
+        )
 
         caplog.clear()
         await tasks._watchdog_homeassistant_api()
 
         restart.assert_called_once()
-        assert "Watchdog miss API response from Home Assistant" not in caplog.text
-        assert "Watchdog found a problem with Home Assistant API!" in caplog.text
+        assert "Watchdog missed an Home Assistant Core API response." not in caplog.text
+        assert (
+            "Watchdog missed 2 Home Assistant Core API responses in a row. Restarting Home Assistant Core API!"
+            in caplog.text
+        )
 
 
 async def test_watchdog_homeassistant_api_off(tasks: Tasks, coresys: CoreSys):
@@ -120,10 +126,10 @@ async def test_watchdog_homeassistant_api_reanimation_limit(
         await tasks._watchdog_homeassistant_api()
 
         restart.assert_not_called()
-        assert "Watchdog miss API response from Home Assistant" not in caplog.text
+        assert "Watchdog missed an Home Assistant Core API response." not in caplog.text
         assert "Watchdog found a problem with Home Assistant API!" not in caplog.text
         assert (
-            "Watchdog cannot reanimate Home Assistant, failed all 5 attempts."
+            "Watchdog cannot reanimate Home Assistant Core, failed all 5 attempts."
             in caplog.text
         )