mlco2 / mlco2/codecarbon

RAM's default pid is evaluated at import time, so forked workers track the parent process

Open Beginner friendly
#1,318 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
1.9k
Forks
323
Avg merge
1d 12h
Merged PRs (30d)
12

Description

Problem

RAM.__init__ uses pid: int = psutil.Process().pid as a default argument. Python evaluates default arguments once, when the module is first imported, so the default pid is frozen to the importing process for the life of the interpreter — including across fork(). A child process that constructs RAM() therefore records the parent's pid.

Reproduction

import os, multiprocessing as mp
from codecarbon.external.ram import RAM   # imported in the parent

def worker():
    print(os.getpid(), RAM(tracking_mode="process")._pid)  # these differ

mp.set_start_method("fork")
p = mp.Process(target=worker); p.start(); p.join()

Linux/fork only — with spawn the child re-imports and the default is recomputed.

Root cause

  • codecarbon/external/ram.py:37pid: int = psutil.Process().pid (evaluated at import).
  • codecarbon/external/ram.py:58self._pid = pid.
  • Neither real construction site passes a pid: codecarbon/core/resource_tracker.py:37 and codecarbon/core/hardware_cache.py:164, so the stale default is always the one in use.

By comparison CPU resolves it correctly inside __init__ (codecarbon/external/hardware.py:216).

Expected vs actual

Expected: with tracking_mode="process", RAM usage is measured for the process that created the tracker.

Actual, in a forked child:

  • If the parent is alive, process_memory_GB (ram.py:285-296) measures the parent's RSS plus all its descendants — i.e. every sibling worker — so N workers each report roughly the whole pool.
  • If the parent has exited, psutil.Process(pid) raises NoSuchProcess; total_power swallows it (ram.py:337-339) and RAM power silently reads 0 W for the rest of the run.

Machine tracking mode is unaffected (it never reads _pid).

Fix

Default pid to None and resolve os.getpid() in __init__.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start in codecarbon/external/ram.py at RAM.init and compare its pid handling with CPU in codecarbon/external/hardware.py:216. Check the construction paths in codecarbon/core/resource_tracker.py:37 and codecarbon/core/hardware_cache.py:164, then verify process-mode RAM in ram.py:285-296 under fork. Done means a forked child records and measures its own pid while machine mode remains unaffected.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
backend
Issue type
Bug
Difficulty
2/5
Estimated time
1-3 hours
Activity status
Quiet
Clarity
Clearly specified
Newbie friendliness
78/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.