The host on the SSL certificates for EMR (<region>.elasticmapreduce.amazonaws.com) is different from the host in the AWS documentation (elasticmapreduce.<region>.amazonaws.com).

Pull request #705 changes us over from the host in AWS's documentation to the host that matches the SSL certificates, which should fix #621.

However, this makes me very nervous. The documentation is wrong now, but someday it may be right, and then mrjob will not work correctly.

What we should do, before we release v0.4.1, is something I mentioned in #658: wrap the connections in such a way that we first try the host in the documentation, and if we get an SSL error, we fall back to the host that we know matches the SSL certs.