[Zope-dev] Re: SQLAlchemy integration experiment

[Martijn Faassen]

Hey,


[replying to myself]

Martijn Faassen wrote:
> Laurence Rowe wrote:
> [snip]
>> I'm not sure connection pooling is really useful in a threaded 
>> environment with recycled sessions. You want n threads = n 
>> connections. If we started creating new sessions each request then 
>> things would be different.
> 
> Mike Bayer says the following:
> 
>  > you should reuse a single engine.  It contains a pool of connections,
>  > so recreating engines means new connections are constantly rebuilt and
>  > it defeats the purpose of the pool.  Other than that the creation of
>  > an engine is not very expensive but even a small expense is needless
>  > here.
> 
> I explained to him we'd only end up creating a single engine per thread 
> (and app), and we'll see what he has to say about that.

Mike responded like this:

 > yeah that also kind of defeats the connection pool's purpose as it's
 > intended to handle all connections for a particular database across
 > threads.  Having a single point of pooling for a certain database has
 > the advantage that you can tune the total number of connections at a
 > single configuration point, and also that a database restart can be
 > detected just once, resulting in a bounce of the entire pool (this is
 > a distinct advantage of using a pool over a non-pooled approach).

So, it looks like we need to make sure we maintain a single engine per 
database.

Regards,

Martijn