Database We Can Believe In: Stories from the Front Lines and Server Rooms of Barack Obama's Online Presidential Campaign
summary
Campaign systems scale from a few hundred hits to peaks of 4,300 hits/sec, house a 5 TB database, and raise over half a billion dollars from about 4 million online donors. MySQL tuning and architecture choices include moving MyISAM tables to InnoDB, using snapshotting on a block‑level filesystem (then reverting when long reporting jobs lock tables), sharding with merge tables, negative-table logic to cut lists, parallel mailers hitting ~2 million msgs/hour and a near‑real‑time Houdini voter pipeline that aggregates early‑vote feeds to inform field ops.
AI-written summary · about
chapters (20)
- 0:00Intro
- 0:40Selected clients
- 0:54The BSD Control Panel
- 2:28(Failure at) Capacity Planning
- 5:37How much data?
- 10:47Constituent Search
- 13:43Query Analyzer - our version
- 17:59Mail Recipient Generation
- 23:24Replication and Reporting
- 25:34Work Environment
- 25:59Data Sources
- 26:53Speeds and Feeds and Dataset
- 28:29Early Voting 2008
- 30:03Sample Earlyvote Analytics Query
- 30:26Example Aggregated Table
- 31:56Florida: 11-04-2008
- 34:23Planning for Election Day
- 35:20The Pipeline: Architecture
- 35:49The Pipeline: Overview Input file (voterids)
- 37:38PA was called one minute after the polls closed, Why?
