Can't start killed process about daemons

[kevinconway]

What signal did you use to kill the process?

[coffebar]

SIGKILL

[kevinconway]

Ah, yeah. SIGKILL terminates the process without allowing it to run the cleanup logic. It's a known shortcoming of using signal trapping to perform pidfile cleanup. I think something like psutils would be a good alternative. I'm trying to keep the project dependency free so I will likely make psutils an optional driver for pid management.

Is using SIGKILL a requirement for your project? If not you should consider using another signal such as SIGTERM as it will properly tigger the cleanup logic and remove the pidfile.

[coffebar]

Ah, yeah. SIGKILL terminates the process without allowing it to run the cleanup logic

it's not a secret

I'm trying to keep the project dependency free so I will likely make psutils an optional driver for pid management.

it would be great

Is using SIGKILL a requirement for your project?

I think that this is possible as the external impact

[tenortim]

The problem isn't just if the daemon is killed. The same issue happens if it dies unexpectedly. Or the host system crashes or loses power. Or any number of other possibilities. That is why full pidfile handling needs to test that the pid is valid.

I need to test portability of the method, but on POSIX systems, there is no need for external dependencies. It's as easy as something like

try:
    os.kill(pid, 0)
except OSError as e:
    if e.errno == errno.EPERM:
        print("pid %d is running under a different username" % pid)
    elif e.errno != errno.ESRCH:
        print("kill of pid %d failed with error %s" % (pid, e.strerror))
    else:
        print("pid %d is running" % pid)

Windows supports os.kill in 2.7 onwards. I'll see if it works there.

[tenortim]

The code as it stands doesn't work on Windows for a number of unrelated reasons.
The signal handling needs SIGQUIT removed since it doesn't exist on Windows (trivial), but then daemonize fails because os.fork doesn't exist either.

As such, I already have a fix (plus tests) for the pidfile handling in my clone. I'll submit a pull request.

[kevinconway]

Thanks to @tenortim I can finally close this bug!