Improved Deadlock Retry


Anyone who’s built a reasonably complex web site on top of a database has seen database deadlock. MySQL will throw a timeout error after waiting for 30 seconds, thereby ending the deadlock in a brutal but effective manner. Rails has a plugin, known as deadlock_retry, which will catch those errors and simply retry the transaction. There’s a good chance this will work since deadlocks are race conditions but it’s not very effective for finding and solving the underlying problem.

So I’ve forked the code and spruced it up a bit:

  1. If a deadlock happens, the code will log the output of SHOW INNODB STATUS, which will indicate who was waiting for what. Knowing the tables in contention goes a long way to understanding the problem.
  2. The code now runs as a gem, not as a plugin. You don’t need to put the code in your application, just add this config.gem reference in config/environment.rb and use rake gems:install to install it:
config.gem 'mperham-deadlock_retry', :lib => 'deadlock_retry', :source => ''

This is code I wrote for our internal usage of deadlock_retry at FiveRuns. Please let me know if you find this improvement useful.