Python Container exits without logs in certain cases

[ucheNkadiCode]

When a user does the repro steps below, by default, the container starts and exits immediately without writing any meaningful error logs in the Debug Console.

Repro Steps:

• During Add Docker Files to Workspace command, choose Python: Django or Python: Flask and select Port 1000.
• After the lines

WORKDIR /app
ADD . /app

Add in

RUN useradd appuser && chown -R appuser /app
USER appuser

• Hit F5 to run the container
• Container will start and stop

Current solution to diagnose this issue:
Although it's possible to find out what's really happening in the container, it is a pain and the user cannot be expected to know/do these steps mentioned below.

• Open and modify your launch.json file
• Add "removeContainerAfterDebug": false, above "python" line
• Run your container again
• After it fails, navigate to docker explorer and click view logs
  

In a Django app, you will see the error:

Error: You don't have permission to access that port.

In a Flask app, you may see the error:

self.socket.bind(self.server_address)
PermissionError: [Errno 13] Permission denied

This likely means you are exposing a system-port (ports 0 - 1023) while attempting to run as a non-root user. This incompatible configuration is demonstrated in the image above, but there must be an easier way to find this issue.

[bwateratmsft]

IMO this is kinda a self-inflicted issue. Our scaffolding will do non-root if all ports are >1023, and root otherwise, so what we scaffold works--of course, it's impossible to stop users from creating something that doesn't work.

In general, this falls into the class of "startup errors". Because Python (and Node) do attach instead of launch (like .NET does), this means startup errors may happen before we have a chance to attach...which means they are missed and the container stops.

We still need to clean up containers in general because we cannot run a new one until the old one is gone, but there may be ways to change the timing of that, or possibly detect containers that shut down immediately before attach.

[haniamr]

I think the problem here is that when the exception is thrown (due to the permission issue), it doesn't get written in the debug console. You can still debug and set breakpoints fine, so I don't think it is a startup issue since ptvsd doesn't execute the app process until something attaches to it.

[bwateratmsft]

In that case would this be a ptvsd issue? If anything, exceptions outta make it to the debug console.

[haniamr]

Might be, I opened an issue over there: microsoft/ptvsd#2117

[bwateratmsft]

From that issue, it sounds like data is being written to stderr but Django is terminating itself before stderr is flushed.