Re: WSGI: creating a new tracker instance for each request?

On Wed, 5 Feb 2020 12:28:32 +0100, Thomas Arendsen Hein wrote:
> * Christof Meerwald <[email protected]> [20200205 00:08]:
>> just noticed that in wsgi_handler.py we create a new tracker instance
>> for each request:
>> 
>>   tracker = roundup.instance.open(self.home, not self.debug)
>> 
>> is that really intentional/needed? It seems like a significant
>> performance issue to me.
> It cost a bit of performance.
>
> Some quick measurements:
> echo -e 'help'|roundup-admin -i . -> 0.280s
> echo -e 'display user1'|roundup-admin -i . -> 0.502s
> echo -e 'display user1\ndisplay user2'|roundup-admin -i . -> 0.505s
>
> So I'd say about 0.2 seconds on our roundup server, which is a VM on
> a slightly old hardware.

Yes, 0.2 seconds for each request (even if it's just a static .css
file), seems significant to me.

Anyway, I think at the moment I would only allow easier customization,
something like

https://bitbucket.org/cmeerw/roundup/commits/4f6d7744edd03c4abbcc7b8e890761a1fd25b4d8

so the behaviour can easily be overridden, probably with something
like this to reuse tracker instances:

    @contextmanager
    def get_tracker(self):
        try:
            tracker = self._pool.pop()
        except IndexError:
            tracker = roundup.instance.open(self.home, not self.debug)

        yield tracker

        self._pool.append(tracker)


> The only problem that I see is that with database backends that
> allow concurrent access without a global lock, commits of a tracker
> state may happen too early.

I am confused - is that related to the above?


Christof

-- 

http://cmeerw.org                              sip:cmeerw at cmeerw.org
mailto:cmeerw at cmeerw.org                   xmpp:cmeerw at cmeerw.org