Open luz-arreola opened 1 year ago
Yes ! Embedding Python into C++ doesn't mean to run a subprocess. Instead, you can run a Python interpreter within the same process and share your application state with the Python runtime. That is described in http://boostorg.github.io/python/doc/html/tutorial/tutorial/embedding.html
Concurrency is still an issue, but mostly only due to the GIL (https://wiki.python.org/moin/GlobalInterpreterLock), i.e. the problem is due to Python's internal design and entirely independent of Boost.Python.
In the past, I have used fastcgi on both IIS and Apache web servers to avoid the expense of creating a process for each request. What happens is that fastcgi creates a pool and any process that is not busy is automatically used. This works great for php applications as well as C++ executables.
Of course, what I was thinking of doing would be very different, e.g. my app would used an unused python process from the pool thus eliminating all of the process construction/deconstruction. Of course, I have no idea how I would go about communicating with the pool of python processes from my C++ application because the documentation I remember was very sparse and not too helpful. I will look a little into your link on GIL, but this does seem a bit frightening to be quite honest.
Thank you for your time and I will eagerly read any suggestion and or comment that you may have. Best regards!
I will be looking at this: https://fastcgi-archives.github.io/FastCGI_Developers_Kit_FastCGI.html
Like I said, I have used fastcgi on c++ cgi executables as well as php websites. It makes a huuuuge difference because each request is handled by an available (reusable) process. Like I said, on my PC, it takes 60ms to create a new process to handle a request.
I am able to use Boost.Process to successfully execute a Python script from within a C++ application. The C++ app can obtain the data returned by Python and this is very trivial.
The problem is performance. Creating a new process needed to execute the Python script is very expensive, e.g. 60ms on a mid-level desktop Windows PC. I have not tested on Linux yet but I'm sure that there will still be a performance hit for each process.
Things that have come to mind:
Can Boost.Python help me with either of these ideas? I've seen older posts online and people seem to give up with the concurrency method. Please advise, thank you.