Bug #4841: WEBrick threading leads to infinite loop

• Status: Closed
• Assignee: normalperson (Eric Wong)
• Target version: -
• ruby -v: run_this.rb
• Backport: 2.3: UNKNOWN, 2.4: UNKNOWN

[peakxu (Peak Xu)]

In a multi-threaded environment, the callback in WEBrick web server does not appear to work properly. If shutdown is called right after start, the server can entire an infinite loop and need to be killed using Ctrl-C.

Reproducible on 1.9.2 p180 on Windows and 1.8.7 on Cygwin.

ruby -v run_this.rb

ruby 1.8.7 (2008-08-11 patchlevel 72) [i386-cygwin]
[2011-06-06 10:12:58] INFO WEBrick 1.3.1
[2011-06-06 10:12:58] INFO ruby 1.8.7 (2008-08-11) [i386-cygwin]
[2011-06-06 10:12:58] INFO WEBrick::HTTPServer#start: pid=4860 port=19000
run_this.rb:95:in join': Interrupt from run_this.rb:95:in stop'
from run_this.rb:102

Note: killed manually to break out of infinite loop.

Recommended fix:
In webrick/server.rb around line 85-88, there's a callback as well as initialization of status to Running. If we move the callback below the initialization, this fixes the issue on 1.8.7 Cygwin but not in 1.9.2 Windows. Code should look like below:

@logger.info
"#{self.class}#start: pid=#{$$} port=#{@config[:Port]}"

    thgroup = ThreadGroup.new
    @status = :Running
    call_callback(:StartCallback)
    while @status == :Running
    <snip>

ruby -v output now looks like:

ruby 1.8.7 (2008-08-11 patchlevel 72) [i386-cygwin]
[2011-06-06 10:05:45] INFO WEBrick 1.3.1
[2011-06-06 10:05:45] INFO ruby 1.8.7 (2008-08-11) [i386-cygwin]
[2011-06-06 10:05:45] INFO WEBrick::HTTPServer#start: pid=5228 port=19000
[2011-06-06 10:05:45] INFO going to shutdown ...
[2011-06-06 10:05:45] INFO WEBrick::HTTPServer#start done.

[ko1 (Koichi Sasada)]

• Assignee set to gotoyuzo (GOTOU Yuuzou)

[ko1 (Koichi Sasada)]

• Status changed from Open to Assigned

[nahi (Hiroshi Nakamura)]

I guess that WEBrick::Server#{start,stop,shutdown} is not designed to be thread-safe and there's no simple fix for your usage. When I used WEBrick::Server for such purpose, I created my own lock outside of WEBrick::Server instances.

I'm going to ask gotoyuzo (the author) but please don't expect a response soon.

Do you think I understand the issue correctly?

[nahi (Hiroshi Nakamura)]

• Assignee changed from gotoyuzo (GOTOU Yuuzou) to nahi (Hiroshi Nakamura)

[mame (Yusuke Endoh)]

• Assignee changed from nahi (Hiroshi Nakamura) to normalperson (Eric Wong)

normal, could you please check this ticket?

• Status changed from Assigned to Closed

Applied in changeset trunk|r61083.

---

webrick: allow shutdown after StartCallback

We must to ensure the @status ivar is set to :Running before
running StartCallback, otherwise Webrick::Server#stop will not
change the @status to :Shutdown properly.

Note: I have not been able to reproduce the original issue but
understood at least part of the problem and fixed it with this
commit. However, the original reporter (Peak Xu) was still able
to reproduce the problem on 1.9.2 p180 on Windows, so I'm not
sure what else might be going on. Ruby threading and
synchronization primitives have changed a lot since 1.9.2, so
maybe that was fixed elsewhere.

• lib/webrick/server.rb: call StartCallback sooner [Bug #4841]

[normalperson (Eric Wong)]

[e-mail] wrote:

normal, could you please check this ticket?

---

Bug #4841: WEBrick threading leads to infinite loop
https://bugs.ruby-lang.org/issues/4841#change-68231

Sure, I think the problem is fixed in r61083.

Looking at the original run-this.rb script, it seems racy
in that the Mutex does not synchronize enough scope for
the ConditionVariable.

[mame (Yusuke Endoh)]

Thank you for your quick response!