Merge pull request #1290 from yuvipanda/concurrent_spawn_limit

Add support for limiting the number of concurrent spawns

jupyterhub/app.py

@@ -534,11 +534,11 @@ class JupyterHub(Application):
     @default('authenticator')
     def _authenticator_default(self):
         return self.authenticator_class(parent=self, db=self.db)
-    
+
     allow_named_servers = Bool(False,
         help="Allow named single-user servers per user"
     ).tag(config=True)
-    
+
     # class for spawning single-user servers
     spawner_class = Type(LocalProcessSpawner, Spawner,
         help="""The class to use for spawning single-user servers.
@@ -547,6 +547,18 @@ class JupyterHub(Application):
         """
     ).tag(config=True)
 
+    concurrent_spawn_limit = Integer(
+        0,
+        help="""
+        Maximum number of concurrent users that can be spawning at a time.
+
+        If more than this many users attempt to spawn at a time, their
+        request is rejected with a 429 error asking them to try again.
+
+        If set to 0, no concurrent_spawn_limit is enforced.
+        """
+    ).tag(config=True)
+
     db_url = Unicode('sqlite:///jupyterhub.sqlite',
         help="url for the database. e.g. `sqlite:///jupyterhub.sqlite`"
     ).tag(config=True)
@@ -1257,6 +1269,7 @@ class JupyterHub(Application):
             statsd=self.statsd,
             allow_named_servers=self.allow_named_servers,
             oauth_provider=self.oauth_provider,
+            concurrent_spawn_limit=self.concurrent_spawn_limit,
         )
         # allow configured settings to have priority
         settings.update(self.tornado_settings)

jupyterhub/handlers/base.py

@@ -367,6 +367,19 @@ class BaseHandler(RequestHandler):
     def spawn_single_user(self, user, server_name='', options=None):
         if server_name in user.spawners and user.spawners[server_name]._spawn_pending:
             raise RuntimeError("Spawn already pending for: %s" % user.name)
+
+        concurrent_spawn_limit = self.settings['concurrent_spawn_limit']
+        if concurrent_spawn_limit and self.settings.get('_spawn_pending_count', 0) > concurrent_spawn_limit:
+            self.log.info(
+                'More than %s pending spawns, throttling',
+                self.settings['concurrent_spawn_limit']
+            )
+            raise web.HTTPError(
+                429,
+                "User startup rate limit exceeded. Try to start again in a few minutes.")
+
+        # FIXME: Move this out of settings, since this isn't really a setting
+        self.settings['_spawn_pending_count'] = self.settings.get('_spawn_pending_count', 0) + 1
         tic = IOLoop.current().time()
         user_server_name = user.name
         if server_name:
@@ -403,6 +416,7 @@ class BaseHandler(RequestHandler):
                 spawner.add_poll_callback(self.user_stopped, user)
             finally:
                 spawner._proxy_pending = False
+                self.settings['_spawn_pending_count'] -= 1
 
         try:
             yield gen.with_timeout(timedelta(seconds=self.slow_spawn_timeout), f)