Issue with Multiprocessing and Camera Connection Using Basler a2A4504-18ucBAS

[diezm]

Issue with Multiprocessing and Camera Connection Using Basler a2A4504-18ucBAS

[diezm]

Thanks for your reply. I

Either create different InstantCameras in each process ( although only one have to have the camera "open" at any time )

I tried this, but it also did not work.

The problem described above only occurs when the program structure is like this:

Main Process
.... do some stuff ....
open camera
read params
close camera
       start subprocess
       open camera

I found a workaround by starting a subprocess just to read out the camera params now, this works

Main Process
.... do some stuff ....

      start subprocess_1
      open camera
      read params
      close cam
      kill subprocess

      start subprocess_2
      open camera
      start streaming 

[thiesmoeller]

your code will try to transfer the InstantCamera between processes.

 def __new__(cls):
        if cls._instance is None:
            cls._instance = super(Basler, cls).__new__(cls)
            cls._instance.camera = pylon.InstantCamera(pylon.TlFactory.GetInstance().CreateFirstDevice())
            cls._instance.is_streaming = Value('b', False)
        return cls._instance

The communication in python multiprocessing is via pickling/unpickling of objects between the processes. This doesn't work and would be extremely inefficient for the InstantCamera.

Either create different InstantCameras in each process ( although only one have to have the camera "open" at any time )
Or by adding additional methods to set/get parameters or a list of parameters from the other process.
The objects transferred have to be pure python.

In #513 (comment) I gave an example how to run pypylon in one process and transfer video data to other process via shared memory to be efficient.

[jonahpearl]

We had the same issue -- your fix resolved it for us. Would still like to understand the root cause so we can avoid it!

EDIT: it appears to be an issue with multiprocessing using spawn instead of fork on Linux. The problem wasn't present on Windows, and we solved it equally well with:

current_mp_start_method = mp.get_start_method()
if current_mp_start_method != "spawn":
    mp.set_start_method("spawn", force=True)

[thiesmoeller]

The root cause is described in #714 (comment).

You have to centralize the camera access ( InstantCamera) in one process.
The object can't move between processes.

Best practice is linked in the above comment. Only process the data in sub processes.
This can be optimized by using shared memory for the image data

[jonahpearl]

Ok — I think I'm experiencing a slightly distinct problem that is better described in #659 — going to take this over there.