[wulczer]

That sounds wrong. You need async database access so that while you're waiting for a query to finish, you can still process other requests.

Imagine an app that receives an HTTP request every second. One in five of these requests requires making a query that takes 3 seconds to complete. The rest is serviced immediately. You don't want the long request freezing your entire server and preventing the fast requests from being serviced.

[dchest]

From this thread https://groups.google.com/forum/#!topic/python-tornado/mgj18... (posting here because Google Groups sometimes require login):

  We experimented with different async DB approaches, but settled on 
  synchronous at FriendFeed because generally if our DB queries were 
  backlogging our requests, our backends couldn't scale to the load 
  anyway. Things that were slow enough were abstracted to separate 
  backend services which we fetched asynchronously via the async HTTP 
  module. 

  I may open source the async MySQL client I wrote, but I am still 
  skeptical of the long term value given the code complexity it 
  introduces. 

  Bret

[wulczer]

Erm, so if things are fast it doesn't matter and if things are slow you should write a whole new service, put a HTTP interface into it and use that? Not sure I buy that :)

Disclaimer: I wrote an async DB module for Twisted/PostgreSQL and it did not turn out to be all that complex.

[wisty]

There's async libraries here: https://github.com/facebook/tornado/wiki/Links - You can get just about any db except MySQL. There's ones for Couch, Mongo, HBase, a bunch of other No-SQLs and Psycopg (PostGres).

Also, Tornado is now introducing some kind of interface to Twisted's event loop, which does have everything ported.

Bret just doesn't seem to want async My-SQL in the main code base.

[spokengent]

Not sure friendfeed ever got enough users to know scaling issues...

Personally I would go with async DB, or have DB access in its own thread(s) and async communicate with that.

[wisty]

From Wikipedia: [FriendFeed] had on average one million monthly visitors ... [and] was bought for $15 million in cash, and $32.5 million in Facebook stock.

I figure after the first million users, your embarrassingly parellel app server layer is not your bottleneck, it's getting the database to scale.

[bdarnell]

The impact of an occasional slow query is mitigated by the fact that you're running multiple processes. You need multiple processes anyway because of the GIL, so just run a few more to make sure you're able to keep the CPU busy. An async DB interface would be nice, but as long as you keep your database performance under control it's not crucial, so it hasn't been written yet.