ºÝºÝߣ

ºÝºÝߣShare a Scribd company logo

Mysql 2007 Tech At Digg V3

?Download as ODP, PDF?
?
E
epee

Technology at digg.com initially used a single Linux server with Apache, PHP, and MySQL. It grew to use multiple servers with load balancing, master-slave replication, and Memcached for caching. Purpose-driven MySQL pools were created to isolate high database loads. While sharding was considered, it was not implemented due to issues with SQL support and increased complexity. Current challenges include physical I/O performance, optimizing queries, and managing monitoring systems.

1 of 16
Technology at digg.com Elliott White III - Eli Tim Ellis - Time
What's all this then? History of Tech at Digg Standard Setup Memcached Purpose Driven MySQL Server Pools Sharding Other MySQL details
History of Tech at Digg Initially one Linux server with Apache 1.3 and PHP 4.x MySQL 4.0 and myisam tables & MySQL full-text search Started with as many open source packages as possible for rapid development. ImageMagick, Ispell, prototype/scriptaculous, etc Grew quick. One server became two. myisam become innodb. We moved to 3 servers, Apache 2.x, MySQL master-slave?replication, started using memcached, moved to PHP 5.x & hired a dba And then the pace picked up yet again...
A Standard Setup Many PHP Servers behind a load balancer:
A Standard Setup Many MySQL slaves, talking to a master
A Standard Setup Randomized PHP to MySQL relations
Memcached What is it & Why use it? Why not just pregenerate pages of content? Performance gains: Caching certain chunks of data that may be used on many different pages. From that, still being able to dynamically create the page, but using cached data.
Memcached Server Farm Issues to be solved: Allowing for Failover and Redundancy Removing or ignoring stale data upon machine failure Stopping perpetuation of database slave lag Making the actual decisions of what and when to cache Solution: Write generic code that understands your data.
Purpose Driven MySQL Pools Creating separate slave pools, that are close to identical in order to isolate high database load.
Purpose Driven Pool Example
Sharding Simplest Definition: Breaking up your database into a number of smaller ones. Pros Greater performance Tweakable / Scalable Cons Loss of SQL support (JOIN) Increased PHP load Complicated programming
Possible Types of Sharding Table-based Range-based Date-based Hashed Partial Sharding
How MySQL Can Do Sharding Partition tables Federated tables MySQL Cluster (NDB) Which of these technologies does Digg use? None of them They weren't ready when we needed them
Digg MySQL Specifics Mix of MySQL 5.0 versions: 5.0.22, 5.0.27, 5.0.30, 5.0.32 A single MySQL 4.1 supporting Cacti MySQL installation & patching done via Debian apt 5.0 performs fine for us compared to 4.1 OLTP DBs are InnoDB & OLAP DB is MyISAM InnoDB backup DB for dumps and slave creation InnoDB recovery faster after slave hardware dies MyISAM great for loads and date-versioned DBs
Current DB Challenges Scaled via ¡°buy more RAM,¡± can't afford that anymore Physical I/O common, so rewrite queries Convincing site architects to change features to more scalable alternatives MySQL debug binary testing Schema cruft Managing monitoring & alerting systems (Cacti) Master is a single point of failure (not as bad as it sounds) Why are our disks lying? Use diskTest.pl from: http://faemalia.net/mysqlUtils
Any Questions? For this presentation and more: http://eliw.com/ Visit http://digg.com/

More Related Content

Mysql 2007 Tech At Digg V3

  • 1. Technology at digg.com Elliott White III - Eli Tim Ellis - Time
  • 2. What's all this then? History of Tech at Digg Standard Setup Memcached Purpose Driven MySQL Server Pools Sharding Other MySQL details
  • 3. History of Tech at Digg Initially one Linux server with Apache 1.3 and PHP 4.x MySQL 4.0 and myisam tables & MySQL full-text search Started with as many open source packages as possible for rapid development. ImageMagick, Ispell, prototype/scriptaculous, etc Grew quick. One server became two. myisam become innodb. We moved to 3 servers, Apache 2.x, MySQL master-slave?replication, started using memcached, moved to PHP 5.x & hired a dba And then the pace picked up yet again...
  • 4. A Standard Setup Many PHP Servers behind a load balancer:
  • 5. A Standard Setup Many MySQL slaves, talking to a master
  • 6. A Standard Setup Randomized PHP to MySQL relations
  • 7. Memcached What is it & Why use it? Why not just pregenerate pages of content? Performance gains: Caching certain chunks of data that may be used on many different pages. From that, still being able to dynamically create the page, but using cached data.
  • 8. Memcached Server Farm Issues to be solved: Allowing for Failover and Redundancy Removing or ignoring stale data upon machine failure Stopping perpetuation of database slave lag Making the actual decisions of what and when to cache Solution: Write generic code that understands your data.
  • 9. Purpose Driven MySQL Pools Creating separate slave pools, that are close to identical in order to isolate high database load.
  • 11. Sharding Simplest Definition: Breaking up your database into a number of smaller ones. Pros Greater performance Tweakable / Scalable Cons Loss of SQL support (JOIN) Increased PHP load Complicated programming
  • 12. Possible Types of Sharding Table-based Range-based Date-based Hashed Partial Sharding
  • 13. How MySQL Can Do Sharding Partition tables Federated tables MySQL Cluster (NDB) Which of these technologies does Digg use? None of them They weren't ready when we needed them
  • 14. Digg MySQL Specifics Mix of MySQL 5.0 versions: 5.0.22, 5.0.27, 5.0.30, 5.0.32 A single MySQL 4.1 supporting Cacti MySQL installation & patching done via Debian apt 5.0 performs fine for us compared to 4.1 OLTP DBs are InnoDB & OLAP DB is MyISAM InnoDB backup DB for dumps and slave creation InnoDB recovery faster after slave hardware dies MyISAM great for loads and date-versioned DBs
  • 15. Current DB Challenges Scaled via ¡°buy more RAM,¡± can't afford that anymore Physical I/O common, so rewrite queries Convincing site architects to change features to more scalable alternatives MySQL debug binary testing Schema cruft Managing monitoring & alerting systems (Cacti) Master is a single point of failure (not as bad as it sounds) Why are our disks lying? Use diskTest.pl from: http://faemalia.net/mysqlUtils
  • 16. Any Questions? For this presentation and more: http://eliw.com/ Visit http://digg.com/