Fix #228: Replace hardcoded port 1108 with dynamic port allocation, s…

[Ashutosh Mishra (Ashut0sh-mishra)]

Summary

Fixes #228

Problem

The Carbon Sequestration notebook fails when run more
than once simultaneously because:

• HTTP_SERVER_PORT was hardcoded to 1108 — second run
  crashes with "port already in use"
• Global ngrok.kill() in finally block killed tunnels
  belonging to other running instances

Fix

• comet_server.py: Replace hardcoded port 1108 with
  find_free_port() that binds to port 0 to get an
  OS-assigned available port
• comet_server.py: Use self.port instance variable
  for per-server port isolation
• comet_requester.py: Remove global ngrok.kill() from
  finally block — server.shutdown() already calls
  ngrok.disconnect() for current tunnel only
• whatif_comet_local.py: Remove global ngrok.kill()
  from del and remove unused pyngrok import

Files Changed

• notebooks/carbon/comet_server.py
• notebooks/carbon/comet_requester.py
• notebooks/carbon/whatif_comet_local.py

Validation

• No hardcoded port 1108 references remain
• Each server instance now gets its own OS-assigned port
• ngrok cleanup is scoped to current tunnel only

Co-authors

Co-authored-by: nik464 nikhil18chaudhary@gmail.com

[Ashutosh Mishra (Ashut0sh-mishra)]

@microsoft-github-policy-service agree

[Renato L. de F. Cunha (renatolfc)]

Thanks for addressing the hard-coded webhook port. We cannot merge this as-is: the ngrok tunnel and agent lifecycle are not cleaned up, pyngrok still uses shared mutable configuration across runs, and choosing a free port before binding it retains a TOCTOU race. Please rebase onto current main, use per-run scoped configuration, bind the port atomically, and guarantee cleanup on success and failure. Note that #199 is a separate COMET API v2.5 compatibility issue.