Backport #2473

Handle leak in TCPServer on Windows

Added by TimWalters (Tim Walters) about 10 years ago. Updated 7 months ago.




  1. Run the following program on Windows (XP, haven't tried others) using Ruby 1.9.1p0:

require 'socket'

server =
while (true)
socket = server.accept
cmd = socket.readline.chomp
puts cmd

  1. Open the task manager, add "Handles" to the columns being viewed, and observe how many handles are being used.

  2. Open an irb window and execute the following lines:

s ='localhost', 2000)

  1. In the task manager, note that the number of handles used by the first process has increased. Every time you connect, three handles are added; when you write and the socket is closed, only two are released.

  2. Try the same instructions using Ruby 1.8.2. All the handles are released, and there is no leak.


Updated by usa (Usaku NAKAMURA) about 10 years ago

  • Status changed from Open to Assigned
  • Assignee set to yugui (Yuki Sonoda)

It's fixed on trunk, in r25569.
Yugui, can you backport the first hunk of the changeset?


Updated by yugui (Yuki Sonoda) about 10 years ago

  • Category set to core
  • Assignee changed from yugui (Yuki Sonoda) to usa (Usaku NAKAMURA)

I did not merge r24218 into ruby_1_9_1, so r25569 is not applicable.

Is the reported problem still reproducible? < unak


Updated by usa (Usaku NAKAMURA) about 10 years ago

Sorry, I've mistaken.
Please backport the first hunk of r25611.


Updated by usa (Usaku NAKAMURA) almost 10 years ago

  • Assignee changed from usa (Usaku NAKAMURA) to yugui (Yuki Sonoda)




Updated by jeremyevans0 (Jeremy Evans) 7 months ago

  • Status changed from Assigned to Closed
  • Description updated (diff)

Also available in: Atom PDF