supabase realtime listener stops receiving events after sometime

[ekorman]

Bug report

• [ x ] I confirm this is a bug with Supabase, not with my own application.
• [ x ] I confirm I have searched the Docs, GitHub Discussions, and Discord.

Describe the bug

My supabase realtime listener stops receiving events after a while, without giving an errors.

To Reproduce

Here is the code snippet I'm using

async def realtime_insert_table_listener(table: str, cb: callable):
    while True:
        try:
            await async_supabase.realtime.connect()

            while not async_supabase.realtime.is_connected:
                await asyncio.sleep(0.2)

            channel = async_supabase.channel(table)
            channel.on_postgres_changes(
                event="INSERT",
                schema="public",
                table=table,
                callback=cb,
            )
            

            await channel.subscribe(
                lambda status, err: status
                == RealtimeSubscribeStates.SUBSCRIBED
                and logger.info("Successfully subscribed")
            )
            while True:
                await asyncio.sleep(10)
        except ConnectionClosedError as exc:
            logger.warning(
                "Connection closed unexpectedly: %s. Will retry...", exc
            )
        except Exception as exc:
            logger.exception(
                "Error in realtime listener: %s. Will retry...", exc
            )
        finally:
            try:
                if async_supabase.realtime.is_connected:
                    logger.info("Closing realtime connection...")
                    await async_supabase.realtime.close()
            except Exception:
                logger.exception("Error while closing realtime connection")

        # Backoff or small delay before attempting to reconnect
        logger.info("Will attempt to reconnect in 5 seconds...")
        await asyncio.sleep(5)

System information

• OS: ubuntu

package information:

>>> supabase.__version__
'2.15.2'
>>> realtime.__version__
'2.4.3'
>>> websockets.__version__
'14.2'

[Marzhal]

Please assign me this issue :)

[Rasheek16]

Can i take up this issue?

[silentworks]

@ekorman how long before it stops receiving events? what environment are you running the code above in and how are you running it?

[ekorman]

@ekorman how long before it stops receiving events? what environment are you running the code above in and how are you running it?

@silentworks i run it like this

asyncio.run(realtime_insert_table_listener(table=table, cb=cb))

inside a docker container (based on python:3.11-slim ) on an ubuntu VM.

hard to say exactly how long before it stops receiving events but seems like it works for between 12-36 hours before no longer receiving messages (or erroring) and never recovering.

[skusez]

Experiencing the same, 3.12-slim, supabase 2.16.0 - ubuntu amd64 vps inside docker

I setup a heartbeat that sends a message to the giveaway-events channel using realtime.send from database, it runs every 1 minute. I simply log the message in python

It emits logs exactly 60 times before silently stopping (60 mins)

...
 self.supabase_client = await create_async_client(REALTIME_URL, API_KEY)
        channel = self.supabase_client.channel("giveaway-events")
        channel.on_broadcast("heartbeat", self.handle_heartbeat)
        await channel.subscribe()
...
def handle_heartbeat(self, payload):
        """Handle when a heartbeat is received"""
        asyncio.create_task(self._handle_heartbeat(payload))

    async def _handle_heartbeat(self, payload):
        """Handle when a heartbeat is received"""
        await self.logger.info(f"Received heartbeat: {payload}")
...

Ran the same test in supabase studios realtime inspector and logs keep coming through with no hiccups (self hosted, running within the same VPS

)

[skusez]

Experiencing the same, 3.12-slim, supabase 2.16.0 - ubuntu amd64 vps inside docker

I setup a heartbeat that sends a message to the giveaway-events channel using realtime.send from database, it runs every 1 minute. I simply log the message in python

It emits logs exactly 60 times before silently stopping (60 mins)

...
self.supabase_client = await create_async_client(REALTIME_URL, API_KEY)
channel = self.supabase_client.channel("giveaway-events")
channel.on_broadcast("heartbeat", self.handle_heartbeat)
await channel.subscribe()
...
def handle_heartbeat(self, payload):
"""Handle when a heartbeat is received"""
asyncio.create_task(self._handle_heartbeat(payload))

async def _handle_heartbeat(self, payload):
    """Handle when a heartbeat is received"""
    await self.logger.info(f"Received heartbeat: {payload}")

...

Ran the same test in supabase studios realtime inspector and logs keep coming through with no hiccups (self hosted, running within the same VPS

)

The fact it drops out after exactly 1 hour leads me to believe its linked with the refresh token issue that has a PR open here #1171

[silentworks]

@skusez provide a full reproducible example please. This should include the Dockerfile setup along with the Python code. Also have you tested it with the PR you stated above to see if it fixes the issue?

[silentworks]

I've setup a test locally that has been running for more than an hour and it's still receiving messages. This test was without the docker container, I'm going to try this now with the docker container to see if it makes a difference.

[skusez]

I tested with the PR and still was silently failing around the hour mark, I did have to patch the package as there was a top level await causing a runtime error

My Dockerfile:

FROM python:3.12-slim

WORKDIR /app

RUN apt-get update && apt-get install -y --no-install-recommends \
    git \
    gcc \
    libpq-dev \
    ca-certificates \
    build-essential \
    && rm -rf /var/lib/apt/lists/*


COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
    # Remove top-level await bug in supabase async client
    RUN sed -i "s/await self.realtime.set_auth(access_token)/self.realtime.set_auth(access_token)/" /usr/local/lib/python3.12/site-packages/supabase/_async/client.py

COPY . .

WORKDIR /app
CMD ["python", "main.py"]

[silentworks]

@skusez if you had to remove the top level await then there is something wrong about the package version you are using, we have no top level await in the library code.

[silentworks]

I've tested this inside of a docker container and still not getting the issue you are having. You can check the code in the repo here https://github.com/silentworks/sb-python-issues/tree/main/supabase_py_1134

[Kabya002]

Hi @skusez — appreciate you testing the PR!

Just to clarify: the main goal of PR #1171 was to fix an Authorization header desync after token refreshes — specifically affecting service clients like auth.admin.list_users() returning 403s due to stale tokens.

The issue you're seeing sounds different:

• Realtime connections silently dropping after ~60 minutes
• No reconnect or error raised
• Possibly related to WebSocket expiry, heartbeat loss, or container idle behavior

That said, you're absolutely right to consider whether the realtime client is staying updated after token refreshes — it’s something we now handle via set_auth(...) in the patch.

Re: the await self.realtime.set_auth(...) change — that line is inside an async method, so if you're seeing a top-level await error, you might be testing an older commit. The current version should not raise that.

If you end up isolating the 60-minute drop to auth-related behavior, I’d be happy to collaborate on a repro or patch!

[silentworks]

@Kabya002 you should state the above on the PR that you opened as there should never be a scenario where auth.admin.list_users() has a stale token as it should only ever be called with the service_role key. Add the comments above on the PR so we can discuss this further.

[Kabya002]

@silentworks I've posted the clarification regarding auth.admin.list_users() and service_role usage in PR #1171 as you suggested.