Can't kill tracd with CTRL-C (or SIGTERM)

[davidbarnett2@…]

If I start up tracd like this:

tracd /path/to/project

and then hit it with the browser, I can't kill the process either by pressing CTRL-C or with kill -15 THE_PID. I can kill the tracd process normally if I do it before the server handles any requests.

[cboos]

At first this sounded familiar, as I thought I had experienced something similar on Windows recently. But I tested again and now it seems to work as expected.

I just tested on Linux (2.6.27-11-server x86_64 GNU/Linux lenny/sid) and interrupting tracd with a CTRL-C or kill -TERM works, with either plain tracd, or -r, or -d (well, obviously not CTRL-C in the latter case). This behavior is also highly dependent on your terminal settings (stty -a?)

[anonymous]

I confirm - Ctrl+C can't kill Trac 0.12 after default installation of openSUSE+Yakuake console.

[derek@…]

... same problem on Centos 5.

[roy@…]

I'm seeing the same on my Ubuntu box (2.6.32.16-linode28 #1 SMP Sun Jul 25 21:32:42 UTC 2010 i686 GNU/Linux), python 2.6.5, mysql 5.1 backend.

This smells like the kind of behavior you get if you've spawned a thread and are waiting for it to exit.

[roy@…]

Just to add some details, if I strace the tracd process, here's what I see when I Control-C it. The string of select() calls are the idle behavior before I hit Control-C.

[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = 0 (Timeout)
[pid  8062] select(4, [3], [], [], {0, 500000}) = ? ERESTARTNOHAND (To be restarted)
[pid  8062] --- SIGINT (Interrupt) @ 0 (0) ---
[pid  8062] rt_sigaction(SIGINT, {0x810fde0, [], 0}, {0x810fde0, [], 0}, 8) = 0
[pid  8062] sigreturn()                 = ? (mask now [])
[pid  8062] futex(0x923e398, FUTEX_WAIT_PRIVATE, 0, NULL

[jaroslav.benkovsky@…]

I have the same problem on Ubuntu 10.04, tracd is run gnome-terminal through ssh with

$ ssh server.example.com
...
$ su - www-data -c "tracd --port 8000 /var/trac"

The tracd process ignores ^C, ^Z, sigterm and sigint (sent as root, of course). Annoying

[jomae]

I have the same issue with 0.12-stable r10430. CTRL-C cannot stops tracd, however kill -TERM works.

I tested with the following environments and the same issue happens.

• python 2.4.6 (Debian 5.0)
• python 2.5.4 and python 2.6.4 (Ubuntu 9.10)
• python 2.6.6 (Windows XP sp3)

In sending CTRL-C, I think that tracd waits the termination of worker threads TracHTTPServer generates as a non-daemon thread.

In kill -TERM, SIGTERM just kills the tracd process because tracd don't handle SIGTERM.

The worker thread should become a daemon thread.

trac/web/standalone.py

@@ -90 +90 @@
 
 
 class TracHTTPServer(ThreadingMixIn, WSGIServer):
+    daemon_threads = True
 
     def __init__(self, server_address, application, env_parent_dir, env_paths,
                  use_http_11=False):

[cboos]

Are you sure about daemon_threads? ​16.2.1. Thread Objects talks about the daemon property.

Besides, how are you starting tracd? When I do simply tracd -p 9001 test-env/, I can interrupt it by ^C without doing anything special (Python 2.5 or 2.7.1, on Linux 2.6.27-11-server x86_64 GNU/Linux).

^C is mapped to intr for you, right? (check with stty -a).

[cboos]

Replying to cboos:

Are you sure about daemon_threads? ​16.2.1. Thread Objects talks about the daemon property.

Sorry, just looked closer. That was for Thread objects, and we're in a SocketServer (​20.17).

[rblank]

Replying to jomae:

The worker thread should become a daemon thread.

IIRC this could have a negative effect on the reload functionality.

[cboos]

Replying to rblank:

Replying to jomae:

The worker thread should become a daemon thread.

IIRC this could have a negative effect on the reload functionality.

That was #7049, and the trouble was with --daemon option which a different thing (the process is set up as daemon process).

Now I was able to reproduce the original issue: if there are many threads or long running threads, I indeed can't interrupt tracd right away.
The patch proposed in comment:8 fixes that. Well done, Jun!

Just to be sure, I also verified that the patch doesn't affect the -r behavior.

[jomae]

Replying to cboos:

Just to be sure, I also verified that the patch doesn't affect the -r behavior.

Thanks, Christian!

Committed in [10436].