Description
Describe your environment
OS: Linux
Python version: 3.13.2
SDK version: 1.28.2
API version: 1.28.2
What happened?
I've encountered an issue with the process.command and related attributes when running Python applications as modules using the -m flag. For example, in a Docker context with:
ENTRYPOINT ["python", "-m", "myapp"]
The collected value for process.command ends up being -m, rather than the actual module name or full command line. This affects observability, as the telemetry data doesn't reflect the real process invocation.
Steps to Reproduce
Create a minimal Python module:
mkdir myapp
echo 'print("Hello")' > myapp/__main__.py
Run it using the -m flag:
python -m myapp
Observe that sys.argv[0] is -m, which causes OpenTelemetry to set:
process.command = "-m"
instead of the expected executable path
Expected Result
If you run:
python -m myapp
Then:
Expected process.command:
/usr/bin/python3 (or similar full path to the Python executable)
Expected process.command_line:
python -m myapp (the full command line as invoked)
Expected process.command_args:
['-m', 'myapp'] (everything after the command)
Actual Result
-m for all three attributes
Additional context
From what I can tell, this happens because the resource detector relies on sys.argv, and when Python is invoked with -m, sys.argv[0] is set to -m (or sometimes the path to main.py), which doesn't give the full picture.
Possible Workaround
I've been experimenting with the idea of directly reading from /proc/self/cmdline on Linux to extract the true command line. Here’s a snippet that demonstrates this approach:
if sys.argv and sys.argv[0] == "-m":
try:
with open("/proc/self/cmdline", "rb") as command_file:
raw_cmdline = command_file.read()
if raw_cmdline:
command_line = raw_cmdline.decode(errors="ignore").split('\0')
if command_line and command_line[0]:
attributes[ResourceAttributes.PROCESS_COMMAND] = command_line[0]
attributes[ResourceAttributes.PROCESS_COMMAND_LINE] = " ".join(command_line)
Suggestions
Would it make sense to incorporate logic like this into the default resource detector?
Alternatively, is there a better approach to improve detection for -m module usage?
Happy to contribute a PR if this seems like a reasonable enhancement.
Would you like to implement a fix?
Yes