Expose celery queue metrics

flower/__init__.py

@@ -1,2 +1,2 @@
-VERSION = (2, 0, 0)
-__version__ = '.'.join(map(str, VERSION)) + '-dev'
+VERSION = (2, 0, 1)
+__version__ = '.'.join(map(str, VERSION))

flower/api/tasks.py

@@ -390,17 +390,7 @@ async def get(self):
 :statuscode 401: unauthorized request
 :statuscode 503: result backend is not configured
         """
-        app = self.application
-
-        http_api = None
-        if app.transport == 'amqp' and app.options.broker_api:
-            http_api = app.options.broker_api
-
-        broker = Broker(app.capp.connection().as_uri(include_password=True),
-                        http_api=http_api, broker_options=self.capp.conf.broker_transport_options,
-                        broker_use_ssl=self.capp.conf.broker_use_ssl)
-
-        queues = await broker.queues(self.get_active_queue_names())
+        queues = await self.get_active_queue_lengths()
         self.write({'active_queues': queues})
 
 

flower/app.py

@@ -8,16 +8,20 @@
 
 from tornado import ioloop
 from tornado.httpserver import HTTPServer
+from tornado.ioloop import PeriodicCallback, IOLoop
 from tornado.web import url
 
 from .urls import handlers as default_handlers
-from .events import Events
+from .events import Events, get_prometheus_metrics
 from .inspector import Inspector
 from .options import default_options
-
+from .utils.broker import get_active_queue_lengths
 
 logger = logging.getLogger(__name__)
-
+# TODO: does this need to be configuration from options?
+BROKER_METRICS_UPDATE_INTERVAL_SECONDS = 10
+# Main dashboard view is updated regardless of this, because it subscribes to live events from celery.
+WORKER_DETAILS_UPDATE_INTERVAL = 120
 
 if sys.version_info[0] == 3 and sys.version_info[1] >= 8 and sys.platform.startswith('win'):
     import asyncio
@@ -79,7 +83,9 @@ def start(self):
             server.add_socket(socket)
 
         self.started = True
-        self.update_workers()
+        self.io_loop.spawn_callback(self.update_broker_metrics)
+        # otherwise self.workers are only updated on UI events and metrics get outdated after some time
+        self.io_loop.spawn_callback(self.update_worker_details)
         self.io_loop.start()
 
     def stop(self):
@@ -101,3 +107,42 @@ def workers(self):
 
     def update_workers(self, workername=None):
         return self.inspector.inspect(workername)
+
+    async def update_broker_metrics(self):
+        logger.debug("Updating broker metrics.")
+
+        def is_worker_alive(worker_name):
+            worker = self.events.state.workers.data.get(worker_name)
+            if not worker:
+                return None
+            return worker.alive
+        while True:
+            next_call = tornado.gen.sleep(BROKER_METRICS_UPDATE_INTERVAL_SECONDS)
+            try:
+                active_queues = await get_active_queue_lengths(self)
+                metrics = get_prometheus_metrics()
+                # clear old data to not leave metrics for queues no longer active
+                metrics.queue_online_workers.clear()
+                metrics.queue_length.clear()
+                for queue_entry in active_queues:
+                    queue = queue_entry["name"]
+                    metrics.queue_length.labels(queue).set(queue_entry["messages"])
+                    nr_of_workers = sum(
+                        1 for name, data in self.workers.items() if
+                        is_worker_alive(name) and any(q["name"] == queue for q in data.get("active_queues", []))
+                    )
+                    metrics.queue_online_workers.labels(queue).set(nr_of_workers)
+            except Exception as e:
+                logger.warning("Updating broker metrics failed with %s", repr(e))
+            else:
+                logger.debug("Done updating metrics.")
+            await next_call
+
+    async def update_worker_details(self):
+        while True:
+            next_call = tornado.gen.sleep(WORKER_DETAILS_UPDATE_INTERVAL)
+            try:
+                self.update_workers()
+            except Exception as e:
+                logger.warning("Failed to update workers list from celery %s", repr(e))
+            await next_call

flower/events.py

@@ -52,6 +52,8 @@ def __init__(self):
             "Number of tasks currently executing at a worker",
             ['worker']
         )
+        self.queue_length = Gauge('flower_broker_queue_length', "Broker queue length", ['queue'])
+        self.queue_online_workers = Gauge('flower_broker_queue_online_workers', "Workers online per queue", ['queue'])
 
 
 class EventsState(State):

flower/utils/broker.py

@@ -89,8 +89,9 @@ class RedisBase(BrokerBase):
     DEFAULT_SEP = '\x06\x16'
     DEFAULT_PRIORITY_STEPS = [0, 3, 6, 9]
 
-    def __init__(self, broker_url, *_, **kwargs):
+    def __init__(self, broker_url, io_loop=None, *args, **kwargs):
         super().__init__(broker_url)
+        self.io_loop = io_loop or ioloop.IOLoop.instance()
         self.redis = None
 
         if not redis:
@@ -109,6 +110,11 @@ def _q_for_pri(self, queue, pri):
         return '{0}{1}{2}'.format(*((queue, self.sep, pri) if pri else (queue, '', '')))
 
     async def queues(self, names):
+        # TODO: use redis.asyncio instead of synchronous client with ThreadPoolExecutor
+        queue_sizes = await self.io_loop.run_in_executor(None, self._queues_synchronous, names)
+        return queue_sizes
+
+    def _queues_synchronous(self, names):
         queue_stats = []
         for name in names:
             priority_names = [self.broker_prefix + self._q_for_pri(
@@ -244,6 +250,40 @@ async def queues(self, names):
         raise NotImplementedError
 
 
+def get_active_queue_names(application):
+    queues = set([])
+    for _, info in application.workers.items():
+        for q in info.get('active_queues', []):
+            queues.add(q['name'])
+    return queues
+
+
+async def get_active_queue_lengths(application):
+    app = application
+    capp = application.capp
+    broker_options = capp.conf.BROKER_TRANSPORT_OPTIONS
+
+    http_api = None
+    if app.transport == 'amqp' and app.options.broker_api:
+        http_api = app.options.broker_api
+
+    broker_use_ssl = None
+    if capp.conf.BROKER_USE_SSL:
+        broker_use_ssl = capp.conf.BROKER_USE_SSL
+
+    broker = Broker(app.capp.connection().as_uri(include_password=True),
+                    http_api=http_api, broker_options=broker_options, broker_use_ssl=broker_use_ssl)
+
+    queue_names = get_active_queue_names(application)
+
+    if not queue_names:
+        queue_names = set([capp.conf.CELERY_DEFAULT_QUEUE]) | \
+                      set([q.name for q in capp.conf.CELERY_QUEUES or [] if q.name])
+
+    queues = await broker.queues(sorted(queue_names))
+    return queues
+
+
 async def main():
     broker_url = sys.argv[1] if len(sys.argv) > 1 else 'amqp://'
     queue_name = sys.argv[2] if len(sys.argv) > 2 else 'celery'

flower/views/__init__.py

@@ -10,6 +10,7 @@
 import tornado
 
 from ..utils import template, bugreport, strtobool
+from ..utils.broker import get_active_queue_names, get_active_queue_lengths
 
 logger = logging.getLogger(__name__)
 
@@ -125,12 +126,9 @@ def format_task(self, task):
         return task
 
     def get_active_queue_names(self):
-        queues = set([])
-        for _, info in self.application.workers.items():
-            for queue in info.get('active_queues', []):
-                queues.add(queue['name'])
-
-        if not queues:
-            queues = set([self.capp.conf.task_default_queue]) |\
-                {q.name for q in self.capp.conf.task_queues or [] if q.name}
-        return sorted(queues)
+        return get_active_queue_names(self.application)
+
+
+    async def get_active_queue_lengths(self):
+        queues = await get_active_queue_lengths(self.application)
+        return queues