connection fails to clean up in-use state after interrupt

[tamird]

repro case: https://gist.github.com/tamird/a2ead907a70a70d54236
this code reliably produces Mysql2::Error: This connection is still waiting for a result, try again once you have the result on my machine (and we see it in production, too).

I don't know enough about the MRI C API, so I haven't had much luck fixing this issue myself. /cc @eam

[sodabrew]

This is a known issue on Ruby 2.1, they changed the Timeout.timeout exception mechanism to only catch if you provide an explicit class type as the second argument.

Which is to say, in your example, rescue Timeout::Error is not caught. Try putting a puts in there, you'll see that it isn't called.

[tamird]

huh? it is caught. also, I've added the reconnect option to the gist and now it is failing reproducibly in Ruby 2.0.0 as well.

[sodabrew]

I have only seen this error on Ruby 2.1, and it is resolved by using the second argument to Timeout.timeout. If you can reproduce it on Ruby 2.0, and/or it is not resolved even with the second argument to Timeout.timeout, then I can imagine that mysql2 will need some fixing. But I'm not sure what fixing is needed.

[tamird]

The gist i've provided fails on ruby 2.0.0.

[tamird]

I believe the fix needed is to make the cleanup call more resilient -- the connection should either disconnect or be marked inactive

[tamird]

@sodabrew ping, this is probably the same issue we "fixed" in mperham/connection_pool#67

Shall I attempt a similar fix?

[sodabrew]

I read through that issue, and I don't think I get it 🎱 What would you change in mysql2? PRs always welcome!

[tamird]

@sodabrew #592