如何关闭PyWinRM会话？使用PyWinRM是否会引发文件描述符泄漏？

Hey there! Let's break down your questions and give you practical, actionable solutions for both:

1. How to close a PyWinRM Session? #

You’re right that PyWinRM’s winrm.Session doesn’t come with a built-in close() method. That’s because it relies on the requests library’s connection pooling under the hood. Still, you can explicitly release underlying HTTP connections if needed:

• Manually shut down the underlying requests session: The PyWinRM Session has a transport attribute that holds the requests session instance. Call its close() method to terminate all connections in the pool:

  s = winrm.Session('10.10.10.10', auth=('administrator', 'password'))
  # Do your work with the session...
  s.transport.session.close()
  

• Let garbage collection handle cleanup: If you don’t explicitly close, Python’s garbage collector will eventually clean up the Session object and release connections once it’s no longer referenced. This works for most standard use cases, but explicit closure is better for high-volume or long-running scenarios.

2. Mitigating File Descriptor Leaks with PyWinRM Polling #

Your concern about leaks is totally valid, especially given Ansible’s past issues with PyWinRM. Here’s how you can stay ahead of the problem:

• Reuse Session objects instead of creating new ones: Spinning up a new Session for every poll is inefficient and raises the risk of unclosed connections. Reuse a single Session across multiple polling cycles—this leverages requests’ connection pooling to keep file descriptor usage stable.
• Explicitly close sessions if you must create new ones: If you can’t reuse sessions (e.g., per-poll authentication changes), make sure to call s.transport.session.close() right after finishing work with each session to free up file descriptors immediately.
• Use the latest PyWinRM version: The Ansible leaks you referenced were mostly tied to older PyWinRM versions where connection cleanup wasn’t handled properly. Upgrading to the latest stable version ensures you get bug fixes for connection management.
• Keep monitoring: You’re already doing this with lsof -p <myprocess> | wc -l—if the count stays stable over time, you’re in good shape. If you see a steady upward trend, double-check that you’re not forgetting to close sessions or creating unnecessary Session instances.
• Wrap Session in a custom context manager: For cleaner, automatic resource management, create a simple context manager to handle closing for you:

  from contextlib import contextmanager
  import winrm
  
  @contextmanager
  def winrm_session(host, auth):
      session = winrm.Session(host, auth=auth)
      try:
          yield session
      finally:
          session.transport.session.close()
  
  # Usage example:
  with winrm_session('10.10.10.10', ('administrator', 'password')) as s:
      # Run your commands or polling logic here
      pass
  

Since you’ve already confirmed your file descriptor count is stable, you’re likely not facing an active leak. Sticking to these practices will help keep it that way.

内容的提问来源于stack exchange，提问作者laycat