[THRIFT-1720] JRuby times out on successful connection

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Not A Problem
• Affects Version/s: 0.8, 0.9
• Fix Version/s: 0.9.1
• Component/s: Ruby - Library

Environment

jruby 1.6.7 and 1.6.8

• Patch Info: Patch Available
• Assignee: Jake Farrell
• Reporter: Hawk Newton
• Created: 10/Oct/12 02:33
• Updated: 09/Jun/13 02:38
• Resolved: 09/Jun/13 02:38

Description

Under JRuby any thrift socket that doesn't connect immediately will result in a connection timeout.

Under MRI IO.select returns an array of IO objects.
Under JRuby IO.select returns nil, even when an object is ready.

Connections to a cassandra instance on localhost will usually be successful, as socket.connect_nonblock actually connects synchronously.

The following code fails with no pause under JRuby (but works under MRI) when connecting to a remote cassandra cluster:

Thrift::Socket.new(remote_server, 9160, 10).open