[THRIFT-1149] Nonblocking server fails when client connection is reset

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.6
• Fix Version/s: 0.7
• Component/s: Ruby - Library
• Labels: nonblocking patch

Environment

FreeBSD orion 8.2-RELEASE FreeBSD 8.2-RELEASE #19 r220462: Fri Apr 8 22:24:26 EDT 2011 alex@orion:/usr/obj/usr/src/sys/ORION amd64
Ruby 1.9.2

• Patch Info: Patch Available
• Assignee: Alex
• Reporter: Alex
• Created: 17/Apr/11 11:12
• Updated: 18/Apr/11 19:13
• Resolved: 18/Apr/11 17:29

Description

I've attempted to stress test the nonblocking server by writing a client which connects to said server and executes three operations as fast as possible in an infinite loop. I've found that if I kill the client (ctrl + C), an intermittent problem occurs where the server will stop accepting new connections from clients. In the case of multiple clients already connected, they are immediately disconnected.

I've tracked the bug down to the run method in nonblocking_server.rb, specifically the "if fd.handle.eof?" statement (around line 150). Under certain conditions executing this line will cause Errno::ECONNRESET to be raised. As such, the "@shutdown_queue.push :shutdown" statement is executed, which accounts for the behavior I am seeing. The attached patch rescues such an error and removes the connection. I am not sure if this is the best way to do it, however it solves the problem for me.