Continuous memory increase in Sanic web server after calling `loop.start_tls(...)` in `connect_utils._create_ssl_connection`

• asyncpg version: 0.2.7
• PostgreSQL version: PostgreSQL 13.3 (Debian 13.3-1.pgdg110+1) on x86_64-pc-linux-gnu, compiled by gcc (Debian 10.2.1-6) 10.2.1 20210110, 64-bit
• Do you use a PostgreSQL SaaS? If so, which? Can you reproduce
  the issue with a local PostgreSQL install?: I use PostgreSQL installed in docker image kartoza/postgis:pr-333-13-3.1
• Python version: 3.8.10
• Platform: 20.04.1-Ubuntu x86_64
• Do you use pgbouncer?: Not sure, I use it indirectly. (behind sqlalchemy)
• Did you install asyncpg with pip?: Yes
• If you built asyncpg locally, which version of Cython did you use?: /
• Can the issue be reproduced under both asyncio and
  uvloop?: I see the code below is based on uvloop.

Background: I built a simple web server using sqlalchemy and sanic, then created a request handler function that can query my local PostgreSQL server running in docker. However, I saw continuous memory increase while benching my handler function.

After running tracemalloc for help, I got the result below:

Top 10 lines
#1: /home/tusimple/Workspace/projects/map-workdata-manager/venv/lib/python3.8/site-packages/asyncpg/connect_utils.py:700: 5670.1 KiB
    new_tr = await loop.start_tls(

The code for creating the async_engine is:

from sqlalchemy.ext.asyncio import create_async_engine
engine = create_async_engine("postgresql+asyncpg://{}:{}@{}/{}".format(...))

At first, I tried adding ?ssl=disable after the previous connect string in order to prevent connect_utils._create_ssl_connection calls, and it works.
Then, I tried to make some stupid changes to the source code of asyncpg, because I think this SSL transport is not closed after creating new transport. (I'm not sure is that correct)

# line 697 
    if hasattr(loop, 'start_tls'):
        if do_ssl_upgrade:
            try:
                new_tr = await loop.start_tls(
                    tr, pr, ssl_context, server_hostname=host)        # <--- here
            except (Exception, asyncio.CancelledError):
                tr.close()
                raise
        else:
            new_tr = tr
# line 706

Finally, I modified the code starts from line 702 to:

            except (Exception, asyncio.CancelledError):
                raise
            finally:
                tr.close()

After this modification, my web server keeps working fine, but I don't know whether this is a proper modification.

If you have any questions or ideas, please let me know, thank you so much!