fix: validate atomic configuration updates and expire stale monitoring snapshots

app/plugin_api.py

@@ -4,6 +4,7 @@ from contextlib import closing
 import http.client
 import json
 import socket
+import time
 import threading
 import uuid
 from pathlib import Path
@@ -65,14 +66,15 @@ class Publisher:
     """One replaceable snapshot, not an unbounded queue of animation commands."""
     def __init__(self,plugins,source,log):
         self.plugins=plugins;self.source=source;self.log=log
-        self.session=str(uuid.uuid4());self.revision=0;self.snapshot=None
+        self.session=str(uuid.uuid4());self.revision=0;self.snapshot=None;self.published_at=None
         self.lock=threading.Lock();self.stop_event=threading.Event();self.status={}
         self.threads=[threading.Thread(target=self.run,args=(key,path),name='plugin-'+key,daemon=True)
                       for key,path in plugins.items()]
         for thread in self.threads:thread.start()
 
     def publish(self,state):
-        with self.lock:self.snapshot=copy.deepcopy(state)
+        with self.lock:
+            self.snapshot=copy.deepcopy(state);self.published_at=time.monotonic()
 
     def current_status(self):
         with self.lock:return copy.deepcopy(self.status)
@@ -80,8 +82,11 @@ class Publisher:
     def run(self,key,path):
         previous_error=None;revision=0;descriptor=None
         while not self.stop_event.is_set():
-            with self.lock:state=self.snapshot
-            if state is not None:
+            with self.lock:
+                state=self.snapshot
+                fresh=self.published_at is not None and time.monotonic()-self.published_at<3
+            # Never renew a hardware lease from a stalled monitoring loop.
+            if state is not None and fresh:
                 revision+=1
                 payload={'api_version':1,'source':self.source,'session':self.session,'revision':revision,
                          'ttl_seconds':10,'system':state['system'],'network':state['network'],'alerts':state['display']['alerts']}