ruby on rails 3 - Mysql 5.6 headaches on Mac OSX

Question

Several of my colleagues and I have recently upgraded from MySQL 5.5 to MySQL 5.6 using homebrew on our Macs to test locally before upgrading our servers. Since this upgrade, we all have been experiencing intermittent MySQL errors when running our rails code:

Lost connection to MySQL server at 'sending authentication information', system error: 32

We have tried re-making our usernames and passwords in our database, and upping the connection timeout, but neither have fixed the problem. The error logs do not mention the issue. The only workaround we have found when we run into the problem is to kill mysql and restart it. I have even noticed this error more recently using mysql -u root -p on the command line. It seems that once I start getting this error, I cannot exceed my current number of connections no matter what username I use. If I close a connection, then I can re-open one.

We have the following environments:

• some of us: Rails 3.2, Ruby 2, mysql2 0.3.13, MySQL 5.6.12, Mac OSX 10.8.4
• others of us: Rails 3.2, Ruby 1.9, mysql2 0.3.13, MySQL 5.6.10, Mac OSX 10.8.4

Any ideas what might be causing this?

Thanks! Julie

Answer 1

None of the answers here helped me, but finally I got MySQL 5.6 to work.

THREE options to fix MySQL 5.6:

1. (confirmed) Edit /etc/my.cnf (create if not exists) and add:

   [mysqld]
   innodb_file_per_table = OFF
   

and restart MySQL. Then for this to work you'll need to dump your databases into SQL file (mysqldump), then drop and re-create the databases, then load the data back.

2. Change default ulimit value of OSX (suggested by Github user sodabrew): https://superuser.com/questions/261023/how-to-change-default-ulimit-values-in-mac-os-x-10-6

3. Add the following option to [mysqld] section of my.cnf: table_open_cache = 250. By default it is set to 2000, which is way above OSX's default ulimit. This solution is also not recommended, because it hurts the performance of your MySQL - it forces MySQL to re-open tables often, if you have more than 250 tables: https://mariadb.com/kb/en/optimizing-table_open_cache/

Why this error is happening?

Since MySQL 5.6 innodb_file_per_table option is ON by default, which means that each table's data is stored in its own file. OSX default limit of the number of the open files is 256 per process. Normally this isn't a problem, but in my case I'm running unit tests in parallel, which creates 8 databases with 405 tables each. OSX has a limit of the number of open file handles per process. This StackOverflow answer suggests that this limit is 256, which explains my problem perfectly: before MySQL 5.6 all data from all these 8 databases was in ONE file.

Thanks to my colleague Thomas L. who found a MySQL bug report which hinted this solution!