Merge pull request 'misc-stability-improvements' (#56) from misc-stability-improvements into main

Reviewed-on: #56
Reviewed-by: Stefan Kratochwil <kratochwil-la@gmx.de>

software/src/main.py

@@ -53,6 +53,13 @@ def setup_wifi():
     print(f"ifconfig: {wlan.ifconfig()}")
 
 
+async def wdt_task(wdt):
+    # TODO: more checking of app health
+    # Right now this only protects against the asyncio executor crashing completely
+    while True:
+        await asyncio.sleep_ms(500)
+        wdt.feed()
+
 DB_PATH = '/sd/tonberry.db'
 
 config = Configuration()
@@ -97,8 +104,10 @@ def run():
 
             start_webserver(config, the_app)
             # Start
+            wdt = machine.WDT(timeout=1000)
             asyncio.create_task(aiorepl.task({'timer_manager': TimerManager(),
                                               'app': the_app}))
+            asyncio.create_task(wdt_task(wdt))
             asyncio.get_event_loop().run_forever()
 
 

software/src/nfc/nfc.py

@@ -6,6 +6,7 @@ Copyright (c) 2025 Matthias Blankertz <matthias@blankertz.org>
 
 import asyncio
 import time
+from utils import safe_callback
 
 from mfrc522 import MFRC522
 try:
@@ -74,7 +75,7 @@ class Nfc:
                 self.last_uid = uid
                 self.last_uid_timestamp = time.ticks_us()
             if self.cb is not None and last_callback_uid != uid:
-                self.cb.onTagChange(uid)
+                safe_callback(lambda: self.cb.onTagChange(uid), "nfc tag change")
                 last_callback_uid = uid
 
             await asyncio.sleep_ms(poll_interval_ms)

software/src/utils/__init__.py

@@ -1,6 +1,7 @@
 # SPDX-License-Identifier: MIT
 # Copyright (c) 2025 Matthias Blankertz <matthias@blankertz.org>
 
+from utils.helpers import safe_callback
 from utils.buttons import Buttons
 from utils.config import Configuration
 from utils.leds import LedManager
@@ -11,4 +12,4 @@ from utils.sdcontext import SDContext
 from utils.timer import TimerManager
 
 __all__ = ["BTreeDB", "BTreeFileManager", "Buttons", "Configuration", "get_pin_index", "LedManager", "MBRPartition",
-           "SDContext", "TimerManager"]
+           "safe_callback", "SDContext", "TimerManager"]

software/src/utils/buttons.py

@@ -5,6 +5,7 @@ import asyncio
 import machine
 import micropython
 import time
+from utils import safe_callback
 try:
     from typing import TYPE_CHECKING  # type: ignore
 except ImportError:
@@ -74,4 +75,4 @@ class Buttons:
             await self.int_flag.wait()
             while len(self.pressed) > 0:
                 what = self.pressed.pop()
-                self.cb.onButtonPressed(what)
+                safe_callback(lambda: self.cb.onButtonPressed(what), "button callback")

software/src/utils/helpers.py

@@ -0,0 +1,12 @@
+# SPDX-License-Identifier: MIT
+# Copyright (c) 2025 Matthias Blankertz <matthias@blankertz.org>
+
+import sys
+
+
+def safe_callback(func, name="callback"):
+    try:
+        func()
+    except Exception as ex:
+        print(f"Uncaught exception in {name}")
+        sys.print_exception(ex)

software/src/utils/timer.py

@@ -4,6 +4,7 @@
 import asyncio
 import heapq
 import time
+from utils import safe_callback
 
 TIMER_DEBUG = True
 
@@ -49,28 +50,36 @@ class TimerManager(object):
         heapq.heapify(self.timers)
         return i
 
+    def _next_timeout(self):
+        if len(self.timers) == 0:
+            if self.timer_debug:
+                print("timer: worker: queue empty")
+            return None
+        cur_nearest = self.timers[0][0]
+        next_timeout = cur_nearest - time.ticks_ms()
+        if self.timer_debug:
+            if next_timeout > 0:
+                print(f"timer: worker: next is {self.timers[0]}, sleep {next_timeout} ms")
+            else:
+                print(f"timer: worker: {self.timers[0]} elapsed @{cur_nearest}, delay {-next_timeout} ms")
+        return next_timeout
+
+    async def _wait(self, timeout):
+        try:
+            await asyncio.wait_for_ms(self.worker_event.wait(), timeout)
+            if self.timer_debug:
+                print("timer: worker: event")
+            # got woken up due to event
+            self.worker_event.clear()
+            return True
+        except asyncio.TimeoutError:
+            return False
+
     async def _timer_worker(self):
         while True:
-            if len(self.timers) == 0:
+            next_timeout = self._next_timeout()
-                # Nothing to do
+            if next_timeout is None or next_timeout > 0:
-                await self.worker_event.wait()
+                await self._wait(next_timeout)
-                if self.timer_debug:
+            else:
-                    print("_timer_worker: event 0")
+                _, callback = heapq.heappop(self.timers)
-                self.worker_event.clear()
+                safe_callback(callback, "timer callback")
-                continue
-            cur_nearest = self.timers[0][0]
-            wait_time = cur_nearest - time.ticks_ms()
-            if wait_time > 0:
-                if self.timer_debug:
-                    print(f"_timer_worker: next is {self.timers[0]}, sleep {wait_time} ms")
-                try:
-                    await asyncio.wait_for_ms(self.worker_event.wait(), wait_time)
-                    if self.timer_debug:
-                        print("_timer_worker: event 1")
-                    # got woken up due to event
-                    self.worker_event.clear()
-                    continue
-                except asyncio.TimeoutError:
-                    pass
-            _, callback = heapq.heappop(self.timers)
-            callback()