Feature request: reload action

[mailtoarlo@…]

When using ports that are started using launchd (e.g. apache2, mysql, php-fpm) the combination of port unload and port load is used to test configuration changes. Sadly typing this is long and cumbersome:

sudo port unload apache2 && sudo port load apache2

I suggest the port reload action that reduces the above to:

sudo port reload apache2

This is a common practice of "similar" commands like ​service.

[ryandesign (Ryan Carsten Schmidt)]

"sudo port load X" and "sudo port unload X" are only shortcuts for typing "sudo launchctl load -w /Library/LaunchDaemons/org.macports.X.plist" and "sudo launchctl unload -w /Library/LaunchDaemons/org.macports.X.plist" respectively. To my knowledge, launchctl does not provide a command for reloading or restarting something, therefore MacPorts does not provide such a shortcut.

[umassthrower@…]

I'm not sure that a deficiency in launchctl is an appropriate reason to have a deficiency in port.

To start, having the "port reload X" command be a shortcut for "launchctl load -w /Library/LaunchDaemons/org.macports.X.plist && launchctl unload -w /Library/LaunchDaemons/org.macports.X.plist" seems like it would sufficiently simulate the user's expected behavior. I for one was slightly shocked today when sudo port reload dnsmasq failed as the reload/restart common case optimization is supported in most analogous systems as it is by far the most common use case of the load/reload/unload or start/restart/stop paradigm.

[jmroot (Joshua Root)]

For any launchd service that has KeepAlive specified, you get an automatic restart when you stop it. So you can simply run launchctl stop org.macports.apache2 or even apachectl stop.

[ryandesign (Ryan Carsten Schmidt)]

Duplicate #42487 has a patch.

[xeron (Ivan Larionov)]

Is there any reason for not applying patch from #42487?

[neverpanic (Clemens Lang)]

Committed your patch in r123673 with the following changes:

• Removed the delay. I don't think that's necessary, especially since port unload; port load doesn't have the delay and worked before.
• Added the -w flag to unload and load just as it is in portload.tcl and portunload.tcl
• Switched to Tcl 8.5's -ignorestderr flag instead of redirecting stderr to itself
• Removed the change of src/port/port-help.tcl because that file is gone in trunk now. If you have some time, please consider writing a manpage (possibly for load, unload, and reload combined); see #44530 and wiki:NewHelpSystem for more information.

Thanks for the patch!

[xeron (Ivan Larionov)]

I saw errors using "port unload; port load" with redis, it doesn't release tcp port in time. So sleep could be helpful here.

[neverpanic (Clemens Lang)]

A reasonable point. I'm not sure how much wait time is appropriate, though. One second seemed too long to me because most daemons start in well under a second.

Also, this is a redis bug, IMO.

[ryandesign (Ryan Carsten Schmidt)]

Since we use daemondo to handle all launch daemons, and since daemondo is documented to restart the process it's monitoring when it receives a SIGHUP, why don't we send that instead of unloading and loading again?

[neverpanic (Clemens Lang)]

Replying to ryandesign@…:

Since we use daemondo to handle all launch daemons

We don't, see the mpstats and certsync ports, for example.

[xeron (Ivan Larionov)]

Replying to cal@…:

A reasonable point. I'm not sure how much wait time is appropriate, though. One second seemed too long to me because most daemons start in well under a second.

Also, this is a redis bug, IMO.

There're lot of "sleep 1" in linux init scripts.

[neverpanic (Clemens Lang)]

Oh well, better be safe than sorry then, I guess. r123716.