Fix stat widgets freezing permanently after a transient sensor error

library/scheduler.py

@@ -28,9 +28,15 @@
 
 import library.config as config
 import library.stats as stats
+from library.log import logger
 
 STOPPING = False
 
+# Some jobs run every millisecond, so a job failing at every run would flood the log file:
+# only log the same failing job once per this delay (in seconds).
+ERROR_LOG_MIN_INTERVAL = 1.0
+_last_error_log = {}
+
 
 def async_job(threadname=None):
     """ wrapper to handle asynchronous threads """
@@ -62,7 +68,16 @@ def periodic(scheduler, periodic_interval, action, actionargs=()):
                 # If the program is not stopping: re-schedule the task for future execution
                 scheduler.enter(periodic_interval, 1, periodic,
                                 (scheduler, periodic_interval, action, actionargs))
-            action(*actionargs)
+            # A failing sensor read must not kill this stat's thread: the next run is already
+            # scheduled above, so log the error and keep going.
+            try:
+                action(*actionargs)
+            except Exception:
+                name = getattr(action, "__name__", str(action))
+                now = time.monotonic()
+                if now - _last_error_log.get(name, 0.0) >= ERROR_LOG_MIN_INTERVAL:
+                    _last_error_log[name] = now
+                    logger.exception("Scheduled job '%s' failed, skipping this cycle", name)
 
         @wraps(func)
         def wrap(

library/stats.py

@@ -28,7 +28,7 @@
 import os
 import platform
 import sys
-from typing import List
+from typing import List, Optional
 
 import babel.dates
 import requests
@@ -121,6 +121,11 @@ def display_themed_value(theme_data, value, min_size=0, unit=''):
 
 
 def display_themed_percent_value(theme_data, value):
+    # A sensor can transiently return nan: int(nan) raises ValueError, which would kill this
+    # stat's thread. Note this happens before the SHOW check below, i.e. even when hidden.
+    if value is None or math.isnan(value):
+        return
+
     display_themed_value(
         theme_data=theme_data,
         value=int(value),
@@ -130,6 +135,9 @@ def display_themed_percent_value(theme_data, value):
 
 
 def display_themed_temperature_value(theme_data, value):
+    if value is None or math.isnan(value):
+        return
+
     display_themed_value(
         theme_data=theme_data,
         value=int(value),
@@ -142,6 +150,9 @@ def display_themed_progress_bar(theme_data, value):
     if not theme_data.get("SHOW", False):
         return
 
+    if value is None or math.isnan(value):
+        return
+
     display.lcd.DisplayProgressBar(
         x=theme_data.get("X", 0),
         y=theme_data.get("Y", 0),
@@ -201,6 +212,9 @@ def display_themed_radial_bar(theme_data, value, min_size=0, unit='', custom_tex
 
 
 def display_themed_percent_radial_bar(theme_data, value):
+    if value is None or math.isnan(value):
+        return
+
     display_themed_radial_bar(
         theme_data=theme_data,
         value=int(value),
@@ -210,6 +224,9 @@ def display_themed_percent_radial_bar(theme_data, value):
 
 
 def display_themed_temperature_radial_bar(theme_data, value):
+    if value is None or math.isnan(value):
+        return
+
     display_themed_radial_bar(
         theme_data=theme_data,
         value=int(value),
@@ -244,10 +261,14 @@ def display_themed_line_graph(theme_data, values):
     )
 
 
-def save_last_value(value: float, last_values: List[float], history_size: int):
+def save_last_value(value: Optional[float], last_values: List[float], history_size: int):
     # Initialize last values list the first time with given size
     if len(last_values) != history_size:
         last_values[:] = last_values_list(size=history_size)
+    # A sensor may transiently return None: store it as the nan "no data" marker, else
+    # math.isnan() in DisplayLineGraph raises TypeError on every later redraw of this graph.
+    if value is None:
+        value = math.nan
     # Store the value to the list that can then be used for line graph
     last_values.append(value)
     # Also remove the oldest value from list
@@ -925,22 +946,26 @@ def stats(cls):
         theme_data = config.THEME_DATA['STATS']['PING']
 
         delay = ping(dest_addr=PING_DEST, unit="ms")
+        if delay is None or delay is False:
+            # ping3 returns None on timeout and False on unknown host: no delay to display
+            delay = math.nan
 
         save_last_value(delay, cls.last_values_ping,
                         theme_data['LINE_GRAPH'].get("HISTORY_SIZE", DEFAULT_HISTORY_SIZE))
         # logger.debug(f"Ping delay: {delay}ms")
 
-        display_themed_progress_bar(theme_data['GRAPH'], delay)
-        display_themed_radial_bar(
-            theme_data=theme_data['RADIAL'],
-            value=int(delay),
-            unit="ms",
-            min_size=6
-        )
-        display_themed_value(
-            theme_data=theme_data['TEXT'],
-            value=int(delay),
-            unit="ms",
-            min_size=6
-        )
+        if not math.isnan(delay):
+            display_themed_progress_bar(theme_data['GRAPH'], delay)
+            display_themed_radial_bar(
+                theme_data=theme_data['RADIAL'],
+                value=int(delay),
+                unit="ms",
+                min_size=6
+            )
+            display_themed_value(
+                theme_data=theme_data['TEXT'],
+                value=int(delay),
+                unit="ms",
+                min_size=6
+            )
         display_themed_line_graph(theme_data['LINE_GRAPH'], cls.last_values_ping)