Collecting trace for python async

[Artem Shcherbak]

Hello, @derekbruening

I've met this situation: When I collect the trace in the python server through drmemtrace, everything works well before the start of async functions and after. And when I collect a trace with a custom collector, then at the beginning, too, everything is assembled, and inside async functions it is not possible to make attach (the application crashes). Do you know what the specifics of python async can affect this?

[Abhinav Sharma]

Hi,

We'd be happy to help, but more information is required:

- What is the Python server? What are the "async functions" you mentioned? Are these RPC handlers or something? Do they launch any child sub-processes?

- How exactly are you using drmemtrace? By launching the Python server under DR (using something like "drrun -- <cmd for python server>"), or by attaching to an existing process (using something like "drrun -attach <pid>"), or using the start-stop drmemtrace APIs?

- What do you mean by "custom collector"? Do you mean a custom DynamoRIO client?

- Do you have a stack trace for the crash? Is the crash in the application or DynamoRIO code?

- Are you using PyPy or other JIT features?

Abhinav