mirror of
https://github.com/zulip/zulip.git
synced 2025-11-04 05:53:43 +00:00
47 lines
1.3 KiB
Python
47 lines
1.3 KiB
Python
from typing import List
|
|
|
|
from django.db import connection
|
|
from psycopg2.extras import execute_values
|
|
from psycopg2.sql import SQL
|
|
|
|
from zerver.models import UserMessage
|
|
|
|
|
|
class UserMessageLite:
|
|
"""
|
|
The Django ORM is too slow for bulk operations. This class
|
|
is optimized for the simple use case of inserting a bunch of
|
|
rows into zerver_usermessage.
|
|
"""
|
|
|
|
def __init__(self, user_profile_id: int, message_id: int, flags: int) -> None:
|
|
self.user_profile_id = user_profile_id
|
|
self.message_id = message_id
|
|
self.flags = flags
|
|
|
|
def flags_list(self) -> List[str]:
|
|
return UserMessage.flags_list_for_flags(self.flags)
|
|
|
|
|
|
def bulk_insert_ums(ums: List[UserMessageLite]) -> None:
|
|
"""
|
|
Doing bulk inserts this way is much faster than using Django,
|
|
since we don't have any ORM overhead. Profiling with 1000
|
|
users shows a speedup of 0.436 -> 0.027 seconds, so we're
|
|
talking about a 15x speedup.
|
|
"""
|
|
if not ums:
|
|
return
|
|
|
|
vals = [(um.user_profile_id, um.message_id, um.flags) for um in ums]
|
|
query = SQL(
|
|
"""
|
|
INSERT into
|
|
zerver_usermessage (user_profile_id, message_id, flags)
|
|
VALUES %s
|
|
"""
|
|
)
|
|
|
|
with connection.cursor() as cursor:
|
|
execute_values(cursor.cursor, query, vals)
|