Experimental non-blocking futures for Plone
This is a collective package for providing yet another way to do asynchronous (non-blocking) processing on Plone.
This time we speak in terms of promises and futures: promises are asynchronously run functions, which provide their results as requested futures for add-on your code.
A major differences for any other alternatives is that this does not require any additional services, but requires only Plone running on top of a Zope instance.
A major limitation is that the asynchronously executed code cannot access the database in any way (or you may face unexpected consequences). Also, this brings no benefits with HAProxy and fixed amount current requests per instance.
from Products.Five.browser import BrowserView from collective import futures def my_async_task(*args): # a lot of time consuming async processing return u'my asynchronously computed value' class MyView(BrowserView): def __call__(self, *args): try: return futures.result('my_unique_key') except futures.FutureNotSubmittedError: futures.submit('my_unique_key', my_async_task, *args) return u'just a placeholder value'
from Products.Five.browser import BrowserView from collective import futures def my_async_task(*args): # a lot of time consuming async processing return u'my asynchronously computed value' class MyView(BrowserView): def __call__(self, *args): return futures.resultOrSubmit( 'my_unique_key', u'placeholder value', my_async_task, *args)
This package uses approach, which kind of splits a single request into two separate passes:
Whenever some add-on code requires a value to be computed asynchronously, it tries to request for a named future result at first and only then submits a promise function to compute result for that future.
If any futures are submitted, the initial response is never published, but instead the current transaction is aborted and the submitted promise functions are executed in parallel threads separate from the default Zope threads (or even in parallel processes) and their return values are collected (see also the documentation of concurrent.futures in Python).
When all promise functions have been resolved, the original request is cloned, resolved values are set as futures and a new internal request is dispatched.
After this second pass, the add-on code can use the now available futures, not submit more futures, and finally, the response is published all the way to the browser.
For more background information: http://datakurre.pandala.org/2014/05/asynchronous-stream-iterators-and.html
- Fix issue where resolved futures were not passed for nested requests [datakurre]
- First release.