signup: Avoid bloated Stream objects for default streams.

Basically, I eliminate the use of select_all() in a query
that still makes a single round trip.  We have good test
enforcement that Django never needs to lazily fetch
objects off the Stream object. (It used to be common
to fetch stream.realm a while back, but we upgraded
bulk_add_subscription, in particular, a while back.)

zerver/actions/create_user.py

@@ -18,7 +18,7 @@ from zerver.actions.user_groups import do_send_user_group_members_update_event
 from zerver.actions.users import change_user_is_active, get_service_dicts_for_bot
 from zerver.lib.avatar import avatar_url
 from zerver.lib.create_user import create_user
-from zerver.lib.default_streams import get_default_streams_for_realm
+from zerver.lib.default_streams import get_slim_realm_default_streams
 from zerver.lib.email_notifications import enqueue_welcome_emails
 from zerver.lib.mention import silent_mention_syntax_for_user
 from zerver.lib.send_email import clear_scheduled_invitation_emails
@@ -137,10 +137,13 @@ def set_up_streams_for_new_human_user(
         prereg_user.referred_by is not None or prereg_user.multiuse_invite is not None
     )
 
-    # If the Preregistration object didn't explicitly list some streams (it happens when user
-    # directly signs up without any invitation), we add the default streams
+    # If the Preregistration object didn't explicitly list some streams (it
+    # happens when user directly signs up without any invitation), we add the
+    # default streams for the realm. Note that we are fine with "slim" Stream
+    # objects for calling bulk_add_subscriptions and add_new_user_history,
+    # which we verify in StreamSetupTest tests that check query counts.
     if len(streams) == 0 and not user_was_invited:
-        streams = get_default_subs(user_profile)
+        streams = get_slim_realm_default_streams(realm.id)
 
     for default_stream_group in default_stream_groups:
         default_stream_group_streams = default_stream_group.streams.all()
@@ -610,9 +613,3 @@ def do_reactivate_user(user_profile: UserProfile, *, acting_user: Optional[UserP
         stream_dict=stream_dict,
         private_peer_dict=subscriber_peer_info.private_peer_dict,
     )
-
-
-def get_default_subs(user_profile: UserProfile) -> List[Stream]:
-    # Right now default streams are realm-wide.  This wrapper gives us flexibility
-    # to some day further customize how we set up default streams for new users.
-    return get_default_streams_for_realm(user_profile.realm_id)

zerver/lib/default_streams.py

@@ -4,14 +4,18 @@ from zerver.lib.types import APIStreamDict
 from zerver.models import DefaultStream, Stream
 
 
-def get_default_streams_for_realm(realm_id: int) -> List[Stream]:
-    # TODO: Deprecate this extremely expensive query. We can't immediately
-    #       just improve it by removing select_related(), because then you
-    #       may get the opposite problem of multiple round trips.
-    return [
-        default.stream
-        for default in DefaultStream.objects.select_related().filter(realm_id=realm_id)
-    ]
+def get_slim_realm_default_streams(realm_id: int) -> List[Stream]:
+    # We really want this query to be simple and just get "thin" Stream objects
+    # in one round trip.
+    #
+    # The above is enforced by at least three tests that verify query counts,
+    # and test_query_count in test_subs.py makes sure that the query itself is
+    # not like 11000 bytes, which is what we had in a prior version that used
+    # select_related() with not arguments (and thus joined to too many tables).
+    #
+    # Please be careful about modifying this code, as it has had a history
+    # of performance problems.
+    return list(Stream.objects.filter(defaultstream__realm_id=realm_id))
 
 
 def get_default_stream_ids_for_realm(realm_id: int) -> Set[int]:
@@ -23,11 +27,6 @@ def get_default_streams_for_realm_as_dicts(realm_id: int) -> List[APIStreamDict]
     Return all the default streams for a realm using a list of dictionaries sorted
     by stream name.
     """
-
-    # This slightly convoluted construction makes it so that the Django ORM gets
-    # all the data it needs to serialize default streams as a list of dictionaries
-    # using a single query without using select_related().
-    # This is enforced by test_query_count in test_subs.py.
-    stream_ids = DefaultStream.objects.filter(realm_id=realm_id).values_list("stream_id")
-    streams = Stream.objects.filter(id__in=stream_ids)
-    return sorted((stream.to_dict() for stream in streams), key=lambda elt: elt["name"])
+    streams = get_slim_realm_default_streams(realm_id)
+    stream_dicts = [stream.to_dict() for stream in streams]
+    return sorted(stream_dicts, key=lambda stream: stream["name"])