This is invaluable, thanks for sharing. We're running into some really interesting problems building Djangy.com - this slideshow is a goldmine.
How do you deal with schema changes? Do they force you to switch into a read-only mode?
The talk was recorded, and the video will be made available (once it's ready).
For such web services, what is the average/peak load on the network? Is it a bottleneck at all? Are the networks over provisioned?
The article says peak is 17k req/sec ... (I'm impressed)
Yes, I presume that's incoming 17k req/sec. Each incoming request can then lead to many database queries/memcached queries/etc. It would be more insightful if request/response sizes were also known.
How many HAProxy server is being used? Just a main one with heartbeat for failover?
nice starcraft2 pitch at the end ;)
I see someone playing Starcraft most of the time while I pass by their office (ours is right above theirs).
It's true :/
Strategical Thinking Practice.
Probably dcramer
Why is emailing an exception bad (Slide 48)? Can a small site get away with it ?
It's fine for small sites. You'll know when your site is too big for that feature from the pain in your inbox.
A possible disaster scenario: connection issues with the SMTP server are causing the initial exception, resulting in webserver connections that are open for excessively long periods of time (unless you're sending the emails asynchronously), since the process is blocked first on the failed SMTP connection, then again on the exception email attempt (and you're still not getting any notification that something is broken, unless you're doing some other kind of monitoring!)