On Wed, Jan 16, 2013 at 11:26 AM, Jean-Baptiste Quenot <[email protected]> wrote:
> OK, I didn't know about the "attach-daemon" option.
Me neither. Looks like something I can use for inotify-hookable too.
> Here is the python reload script I finally came up with:
>
> #! /usr/bin/python
>
> import pyinotify, sys, os, signal
>
> def callback(evt):
> print "Change detected!", evt
> os.kill(os.getppid(), signal.SIGHUP)
> sys.exit(0)
>
> wm = pyinotify.WatchManager()
> notifier = pyinotify.Notifier(wm, default_proc_fun=callback)
> root = sys.argv[1]
> for parentdir, dirs, files in os.walk(root):
> for file in files:
> if file.endswith(".tmpl") or file.endswith(".py"):
> wm.add_watch(os.path.join(parentdir, file),
> pyinotify.ALL_EVENTS)
> notifier.loop()
>
> And the uwsgi config:
>
> {
> "uwsgi": {
> "attach-daemon": "uwsgi-watch-wc /path/to/my/app",
> "module": "server:application",
> "pythonpath": ["/path/to/my/app"],
> "uid": "www-data",
> "gid": "www-data"
> }
> }
>
> Thanks to all for your useful contributions!
I've only read the code for this and not actually tried it, but
there's a few major caveats with it that my inotify-hookable deals
with:
* You're only watching existing files, you need to be adding watches
on directories in case someone adds a file.
* You're going to restart as soon as you get one inotify event,
that's very wasteful and usually not what you want.
When someone e.g. does a "git pull" you'll restart right away when
1/100 files has been changed, and might possibly be back up before
all the files are changed, and before you can re-setup the watch.
inotify-hookable deals with this by waiting X ms when it gets a
inotify hit waiting for more events, as long as it keeps getting
events it'll extend the wait a bit more before finally running
hooks.
_______________________________________________
uWSGI mailing list
[email protected]
http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
