pserve --reload with --daemon

[mmerickel]

I'm not sure if this is supposed to work, but tizoc in #pyramid claimed it may be a problem. If this is supposed to work we should verify that it does. If it is not supposed to work (which I think would be fine) we should add an error message when trying to execute pserve with both options enabled.

[mcdonc]

It is not supposed to work AFAIK.

[mmerickel]

https://github.com/mmerickel/pyramid/compare/fix.pserve-monitor-daemon

Contains a fix adding support for the full blown combination of --reload --monitor-restart --daemon.

[digitalresistor]

Would this also solve the problem if the reload fails due to a Python exception? Right now I have pserve for development sitting in a loop with --reload so that if it fails, it sleep 10 and retries. This way I don't need to go restart it when I save a file and miss a comma or something along those lines.

[mmerickel]

This does allow for that. Although it attempts to restart with no delay, so we'd probably want to add a throttle to the --monitor-restart option or combine the logic such that we "don't restart until a file changes". A more ideal implementation would be something like https://pypi.python.org/pypi/ReloadWSGI which does not crash when your application fails to start, but continues to serve the old version. I had issues using ReloadWSGI on OS X last time I tried, and I think that it has no windows support at all.

[digitalresistor]

At least in my case, I have no need for it to continue serving the old version, but having it wait till a file has changed after failure would probably be best.

[joulez]

Slightly related (getting daemon functionality to be slightly more managable)
The pserve daemon feature needs some clean up, particularly when checking for a PID (which should be done early on). I'm not sure of the benefits of read_pidfile and live_pidfile being seperate functions? It seems to add some complexity (only care about a valid PID).
But this will break some test scripts that check for 1) The pid file 2) That the file contains a value 3) That the value is a valid PID.
Is the pserve daemon functionality and the subcommands widely used? {start|stop|restart|status}

[piotr-dobrogost]

From what I see --reload does not work with --daemon in Pyramid 1.4.5
What's preventing @mmerickel's fix from getting merged?

[mcdonc]

My $.02: I'd rather get rid of --daemon and --stop-daemon than trying to make pserve a full-blown process manager or add complexity to make daemon-stopping work with --monitor-restart. Very good process managers already exist and pserve doesn't really need to do their jobs.

That said, if @mmerickel wants to merge the patch, I'd be fine with it.

[mmerickel]

I'm a big fan of removing --daemon but it may be controversial.

Side note: my fix is something I would not consider production ready due to the infinite loop caused by saving a file with broken code while --monitor-restart is active.

[digitalresistor]

I'd be all for removing --daemon. I never understood why we try to be everything with pserve. Then again, I run on top of uWSGI.