reduce repeat queries in GET /api/users

add eager loading of several relationships that are ~always used when the given objects are requested
add specific eager loading of spawners to the users query

- roles, groups (always needed to resolve permissions)
- APIToken.user, service

jupyterhub/apihandlers/users.py

@@ -9,6 +9,7 @@ from datetime import datetime, timedelta, timezone
 from async_generator import aclosing
 from dateutil.parser import parse as parse_date
 from sqlalchemy import func, or_
+from sqlalchemy.orm import raiseload, selectinload
 from tornado import web
 from tornado.iostream import StreamClosedError
 
@@ -90,6 +91,10 @@ class UserListAPIHandler(APIHandler):
         # post_filter
         post_filter = None
 
+        # starting query
+        # fetch users and groups, which will be used for filters
+        query = self.db.query(orm.User).outerjoin(orm.Group, orm.User.groups)
+
         if state_filter in {"active", "ready"}:
             # only get users with active servers
             # an 'active' Spawner has a server record in the database
@@ -97,9 +102,9 @@ class UserListAPIHandler(APIHandler):
             # it may still be in a pending start/stop state.
             # join filters out users with no Spawners
             query = (
-                self.db.query(orm.User)
+                query
                 # join filters out any Users with no Spawners
-                .join(orm.Spawner)
+                .join(orm.Spawner, orm.User._orm_spawners)
                 # this implicitly gets Users with *any* active server
                 .filter(orm.Spawner.server != None)
             )
@@ -114,9 +119,8 @@ class UserListAPIHandler(APIHandler):
             # this is the complement to the above query.
             # how expensive is this with lots of servers?
             query = (
-                self.db.query(orm.User)
+                query.outerjoin(orm.Spawner, orm.User._orm_spawners)
-                .outerjoin(orm.Spawner)
+                .outerjoin(orm.Server, orm.Spawner.server)
-                .outerjoin(orm.Server)
                 .group_by(orm.User.id)
                 .having(func.count(orm.Server.id) == 0)
             )
@@ -124,7 +128,19 @@ class UserListAPIHandler(APIHandler):
             raise web.HTTPError(400, "Unrecognized state filter: %r" % state_filter)
         else:
             # no filter, return all users
-            query = self.db.query(orm.User)
+            query = query.outerjoin(orm.Spawner, orm.User._orm_spawners).outerjoin(
+                orm.Server, orm.Spawner.server
+            )
+
+        # apply joinedload options
+        query = query.options(
+            selectinload(orm.User.groups),
+            selectinload(orm.User._orm_spawners),
+        )
+        # if testing, add raiseload to prevent lazy loading of anything we didn't ask for
+        if True:
+            # FIXME: detect tests
+            query = query.options(raiseload("*"))
 
         sub_scope = self.parsed_scopes['list:users']
         if sub_scope != scopes.Scope.ALL:

jupyterhub/orm.py

@@ -263,7 +263,10 @@ class User(Base):
     name = Column(Unicode(255), unique=True)
 
     roles = relationship(
-        'Role', secondary='user_role_map', back_populates='users', lazy="selectin"
+        'Role',
+        secondary='user_role_map',
+        back_populates='users',
+        lazy="selectin",
     )
 
     _orm_spawners = relationship(
@@ -285,6 +288,7 @@ class User(Base):
         "Group",
         secondary='user_group_map',
         back_populates="users",
+        lazy="selectin",
     )
     oauth_codes = relationship(
         "OAuthCode", back_populates="user", cascade="all, delete-orphan"