Accept running as daemon

CHANGELOG.md

@@ -8,6 +8,7 @@ Ensure you use consistent title format.
 ## UNRELEASED
 
 - Fix register-instance CLI.
+- Fix running daemonized.
 - Raise proper error on unknown environment.
 - Improve alerting performance.
 
@@ -27,7 +28,7 @@ Ensure you use consistent title format.
 - Drop rhel7 and buster support, minimum versions are now 6.0.2 for tornado
 and 1.3.2 for sqlalchemy.
 - Drop plugin hotplug. Just restart temBoard UI or agent.
-- Drop daemonization. Use nohup or systemd.
+- Drop daemonization. Use systemd.
 
 **Other changes**
 

agent/temboardagent/web/service.py

@@ -15,12 +15,6 @@
 class HTTPDService(syncio.Service):
     name = "web"
 
-    def __init__(self, app):
-        self.app = app
-
-    def __str__(self):
-        return self.name
-
     # for services.run
     def setup(self, *_, **__):
         ServerHandler.server_software = "temBoard-agent/%s" % __version__

ui/temboardui/toolkit/services.py

@@ -173,6 +173,8 @@ def fork(self, service):
         if hasattr(service, "command"):
             execute(service)
         else:
+            # Enable parent pid check.
+            service.ppid = os.getppid()
             os._exit(run(service))
 
     def _read_pids(self):
@@ -263,7 +265,7 @@ def sigchld_handler(self, *a):
                 if pid == 0:  # Still alive
                     continue
             except ChildProcessError:
-                pass
+                pid = self.pids.get("name", "<unknown>")
 
             logger.warning(
                 "Background service dead. Restarting. service=%s pid=%s", name, pid

ui/temboardui/toolkit/syncio.py

@@ -7,30 +7,33 @@
 
 
 class Service:
-    # Mixin for scheduler and worker pool services
+    # Mixin for scheduler, worker pool and agent httpd services
     # Manages sync loop and signals.
     name = None
 
     def __init__(self, app):
         self.app = app
+        self.ppid = None
 
     def __str__(self):
         return self.name
 
     # interface for services.run
     def create_loop(self):
-        return Loop(self)
+        return Loop(self, self.ppid)
 
     # Interface for SignalMultiplexer
     def sighup_handler(self, *a):
         self.app.reload()
 
 
 class Loop:
-    def __init__(self, service):
+    def __init__(self, service, ppid=None):
         self.service = service
         self.signalmngr = SignalManager()
         self.running = False
+        # If not None, enable parent pid check.
+        self.ppid = ppid
 
     def __repr__(self):
         return "<Loop %s>" % self.service
@@ -49,10 +52,9 @@ def remove_signal_handler(self, sig):
     # Interface for services.run
     def start(self):
         with self.signalmngr:
-            ppid = os.getppid()
             self.running = True
             while self.running:
-                if self._stop_with_parent(ppid):
+                if self._stop_with_parent():
                     break
 
                 self.service.accept()
@@ -63,10 +65,13 @@ def stop(self):
     def close(self):
         self.running = False
 
-    def _stop_with_parent(self, ppid):
-        self.running = os.getppid() == ppid
+    def _stop_with_parent(self):
+        if self.ppid is None:  # Don't check parent.
+            return
+        # If ppid changed (parent exited), stop the loop.
+        self.running = os.getppid() == self.ppid
         if not self.running:
-            logger.info("Parent process exited. Exiting. ppid=%s", ppid)
+            logger.info("Parent process exited. Exiting. ppid=%s", self.ppid)
         return not self.running