I am working on a web app with CherryPy that needs to access a few applications via COM.
Right now I create a new instance of the application with each request, which means each request waits 3 seconds for the application to start and 0.01 for the actual job.
I would like to start each COM application once and keep it alive and reuse it for a few seconds on the following requests because most of the time it is used by a burst of 5-10 ajax requests, then nothing for hours.
Is it possible to share a COM abject across all the threads of a CherryPy application?
Here is the summary of a few experiments that show how it is working now on each request and how it does not work across threads.
The following code successfully starts and stops Excel:
>>> import pythoncom, win32com.client
>>> def start():
global xl
xl = win32com.client.Dispatch('Excel.Application')
>>> def stop():
global xl
xl.quit()
xl = None
>>> start()
>>> stop()
But the following code starts Excel and closes it after 3 seconds.
>>> import pythoncom, win32com.client, threading, time
>>> def start():
global xl
pythoncom.CoInitialize()
xl = win32com.client.Dispatch('Excel.Application')
time.sleep(3)
>>> threading.Thread(target=start).start()
I added the call to CoInitialize()
otherwise the xl
object would not work (see this post).
And I added the 3 second pause, so I could see on the task manager that the EXCEL.EXE process starts and is alive for 3 seconds.
Why does it die after the thread that started it ends?
I checked the documentation of CoInitialize()
, but I couldn't understand if it is possible to get it to work in multithreaded environment.