sigterm and sigint both shutdown request shutdown, which stops consuming urls and waits for active browsers to finish; a second sigint/sigterm immediately shuts down active browsers

bin/umbra

@@ -9,6 +9,8 @@ import sys
 import signal
 import os
 import umbra
+import signal
+import threading
 
 if __name__=="__main__":
     import faulthandler
@@ -43,14 +45,34 @@ if __name__=="__main__":
             exchange_name=args.amqp_exchange, queue_name=args.amqp_queue,
             routing_key=args.amqp_routing_key)
 
+    class ShutdownRequested(Exception):
+        pass
+
+    def sigterm(signum, frame):
+        raise ShutdownRequested('shutdown requested (caught SIGTERM)')
+    def sigint(signum, frame):
+        raise ShutdownRequested('shutdown requested (caught SIGINT)')
+
+    signal.signal(signal.SIGTERM, sigterm)
+    signal.signal(signal.SIGINT, sigint)
+
     umbra.start()
 
     try:
         while True:
             time.sleep(0.5)
-    except:
+    except ShutdownRequested as e:
-        pass
+        logging.info(e)
+    except BaseException as e:
+        logging.fatal(e)
     finally:
+        try:
             umbra.shutdown()
+            for th in threading.enumerate():
+                if th != threading.current_thread():
+                    th.join()
+        except BaseException as e:
+            logging.warn("caught {}".format(e))
+            umbra.shutdown_now()
 
 

umbra/browser.py

@@ -20,6 +20,7 @@ class BrowserPool:
 
     def __init__(self, size=3, chrome_exe='chromium-browser', chrome_wait=60):
         self._available = set()
+        self._in_use = set()
 
         for i in range(0, size):
             port_holder = self._grab_random_port()
@@ -47,12 +48,19 @@ class BrowserPool:
         with self._lock:
             (browser, port_holder) = self._available.pop()
             port_holder.close()
+            self._in_use.add(browser)
             return browser
 
     def release(self, browser):
         with self._lock:
             port_holder = self._hold_port(browser.chrome_port)
             self._available.add((browser, port_holder))
+            self._in_use.remove(browser)
+
+    def shutdown_now(self):
+        for browser in self._in_use:
+            browser.shutdown_now()
+
 
 class Browser:
     """Runs chrome/chromium to synchronously browse one page at a time using
@@ -72,6 +80,10 @@ class Browser:
         self.chrome_wait = chrome_wait
         self._behavior = None
         self.websock = None
+        self._shutdown_now = False
+
+    def shutdown_now(self):
+        self._shutdown_now = True
 
     def browse_page(self, url, on_request=None):
         """Synchronously browses a page and runs behaviors. First blocks to
@@ -103,6 +115,9 @@ class Browser:
                         elif self._behavior != None and self._behavior.is_finished():
                             self.logger.info("finished browsing page according to behavior url={}".format(self.url))
                             break
+                        elif self._shutdown_now:
+                            self.logger.warn("immediate shutdown requested")
+                            break
 
                     try:
                         self.websock.close()

umbra/controller.py

@@ -73,9 +73,12 @@ class AmqpBrowserController:
         self.logger.info("shutting down amqp consumer {}".format(self.amqp_url))
         self._amqp_stop.set()
         self._amqp_thread.join()
-        with self._producer_lock:
+        # with self._producer_lock:
-            self._producer_conn.close()
+        #     self._producer_conn.close()
-            self._producer_conn = None
+        #     self._producer_conn = None
+
+    def shutdown_now(self):
+        self._browser_pool.shutdown_now()
 
     def _consume_amqp(self):
         # XXX https://webarchive.jira.com/browse/ARI-3811
@@ -86,11 +89,11 @@ class AmqpBrowserController:
         # reopen the connection every 15 minutes
         RECONNECT_AFTER_SECONDS = 15 * 60
 
-        browser = None
+        url_queue = kombu.Queue(self.queue_name, routing_key=self.routing_key,
+                exchange=self._exchange)
 
         while not self._amqp_stop.is_set():
             try:
-                url_queue = kombu.Queue(self.queue_name, routing_key=self.routing_key, exchange=self._exchange)
                 self.logger.info("connecting to amqp exchange={} at {}".format(self._exchange.name, self.amqp_url))
                 with kombu.Connection(self.amqp_url) as conn:
                     conn_opened = time.time()
@@ -99,13 +102,13 @@ class AmqpBrowserController:
                         while (not self._amqp_stop.is_set() and time.time() - conn_opened < RECONNECT_AFTER_SECONDS):
                             import socket
                             try:
-                                browser = self._browser_pool.acquire()
+                                browser = self._browser_pool.acquire() # raises KeyError if none available
                                 consumer.callbacks = [self._make_callback(browser)]
                                 conn.drain_events(timeout=0.5)
                                 consumer.callbacks = None
                             except KeyError:
                                 # no browsers available
-                                pass
+                                time.sleep(0.5)
                             except socket.timeout:
                                 # no urls in the queue
                                 self._browser_pool.release(browser)
@@ -122,9 +125,6 @@ class AmqpBrowserController:
         return callback
 
     def _browse_page(self, browser, client_id, url, parent_url_metadata):
-        """Kombu Consumer callback. Provisions a Browser and
-        asynchronously asks it to browse the requested url."""
-
         def on_request(chrome_msg):
             payload = chrome_msg['params']['request']
             payload['parentUrl'] = url
@@ -136,7 +136,9 @@ class AmqpBrowserController:
 
         def browse_page_async():
             self.logger.info('browser={} client_id={} url={}'.format(browser, client_id, url))
+            try:
                 browser.browse_page(url, on_request=on_request)
+            finally:
                 self._browser_pool.release(browser)
 
         import random