MySQL5, OS X 10.4.10, startups, etc.

[paul beard]

On 10/1/07, Ryan Schmidt <ryandesign at macports.org> wrote:
>
> On Oct 1, 2007, at 17:52, paul beard wrote:
>
> > On 10/1/07, Ryan Schmidt wrote:
> >
>
> Yes, we do want MacPorts to install as much as possible inside $
> {prefix}. But currently some things do get installed outside of the
> prefix. But IMHO the current uninstall instructions adequately
> address this issue. BTW, "rm -rf /opt/local" is not the extent of the
> uninstall instructions:


Um, yes, I realize that. Allow the possibility that someone will just do rm
-rf /opt/local when they get stuck without realizing that they are leaving
some cruft behind.

> And as for invoking launchct load statements, how is this different?
> >
> > launchctl  load /opt/local/etc/LaunchDaemons/org.macports.dbus/
> > org.macports.dbus.plist
> > launchctl  load /opt/local/etc/LaunchDaemons/org.macports.apache2/
> > org.macports.apache2.plist
> > launchctl  load /opt/local/etc/LaunchDaemons/org.macports.mysql5/
> > org.macports.mysql5.plist
> >
> > from putting the plists in launchd's default search path? As
> > configured, launchctl looks in two places for plists: in /System/
> > Library/LaunchDaemons, for Apple-owned stuff, in /Library/
> > LaunchDaemons for user-installed/third-party stuff. I don't see a
> > big deal about another repository for plists. Maybe in future
> > releases launchd will allow additional directories to be added
> > instead of individual lines.
> >
> >> Your way would require someone to manually edit the file /etc/
> >> launchd.conf, at least to remove a line from it. That's more
> >> complicated than the current way, where a single launchctl line
> >> starts or stops the service. Also, with your way, the service
> >> wouldn't start or stop until the next startup. That's worse than what
> >> we have now, where the service starts or stops immediately.
> >
> > see above: if launchctl works as it says on the tin, there is not
> > different between symlinks that point to /Library/LaunchDaemons and
> > the lines I have in /etc/launchd.conf.
>
> Wait. Currently, MacPorts automatically puts the symlinks in /Library/
> LaunchDaemons, then instructs you how to load (start) and unload
> (stop) as needed. Your proposition as I understood it was that the
> user would manually add and remove lines in /etc/launchd.conf. Are
> you now suggesting instead that MacPorts would automatically put the
> lines in /etc/launchd.conf? If so, why is that preferable to the
> current way? Currently, we add symlinks in /Library/LaunchDaemons,
> which is outside of ${prefix}. You propose modifying /etc/
> launchd.conf, which is also outside of ${prefix}. At least the
> symlinks in /Library/LaunchDaemons have "macports" in the filenames
> so they're easier to find, and getting rid of them is as simple as a
> single "rm" command. Getting rid of a line in a conf file is more
> involved.



Easy to find if someone knows to look for them.

Also, since I'm not familiar with /etc/launchd.conf, I assume that
> the file is only scanned at system startup. Is that the case? If so,
> then adding to that file is worse than what we have now, where
> "launchctl load" will load the service immediately, and cause it to
> start at every subsequent system startup. Unless you're suggesting
> that the user should modify /etc/launchd.conf, and then also use the
> launchctl command to start the service now. If so, you've just added
> an extra step for the user, again for no apparent benefit.


man launchd.conf?

if you understand how launchd/launchctl works, you would know that a
launchctl load statement means "load this plist file and do what it says
until instructed otherwise." It doesn't matter if it's loaded on the command
line or at boot time: once the instructions are loaded, they stay there.

>> Running a single launchctl command to start or stop a service is
> >> easy. Conversely, adding a line to a file is easy, as you showed, but
> >> removing a line from a file requires a more elaborate script. So why
> >> change it to make it more difficult?
> >
> > I'm not arguing in favor of it. I'm just explaining it.  I have a
> > hard time believing that, given the power on tap, that there is no
> > way to add a commented line to /etc/launchd.conf
> >
> > launchctl  load /opt/local/etc/LaunchDaemons/org.macports.dbus/
> > org.macports.dbus.plist # installed by macports
> >
> >
> > and add a # character to the beginning of the line containing the
> > comment:
> >
> > sed -e s/^launchctl/#&/g <-  yes, I know that won't work, but you
> > get the idea: sed is not my native language.
>
> Yes, of course it's possible to write something with sed or similar
> that adds or removes a specific line from a file. My point is that
> it's more difficult than executing a single launchctl statement. And
> since the current way works, I didn't understand why you think it's
> better to complicate things.


Perhaps I don't think it's complicated. It's not clear that the other way
works optimally.  The process of starting up daemons since Tiger is a
departure and for a lot of people it doesn't always work properly (check the
subject line of this thread and see if there are any others similarly named
in the archives).

What I'm hearing is "this way works because I understand it, and I don't
understand your way. Different == bad." I really don't care what anyone
uses. I found a way that works for me, even if it repeats steps that
MacPorts already does.

>> There's a slight problem with the current way. If the service is
> >> running at the time that you uninstall the port, the software stays
> >> running. And if the service is running at the time that you upgrade
> >> the port, the old version stays running and the new version's plist
> >> says the software isn't running and it's inconvenient to fix that
> >> (manually edit the plist to show that the software is running,
> >> launchctl unload the software, launchctl load it again). Would your
> >> new way solve either of these problems? (I haven't tried so I don't
> >> know.)
> >
> > Is there no pre-install step that could stop the service with the
> > existing plist, install, restart?
>
> I don't think MacPorts should be stopping or starting any services
> automatically. That could be unexpected and bad.


Hmm. Sounds like that was what you were asking for. I think upgrading a
process and having it stopped and *not* restarted is worse. I have my
FreeBSD installation set to restart any process for which an rc script is
installed.

So at upgrade time, stop the process with its plist file, do the upgrade,
then let the administrator know that the port is upgraded and not running,
with the requisite launchctl instruction to start it. Conversely, I would
make automatically starting upgraded daemons an option for people in
high-availability environments.

I've been meaning to submit bug reports about these situations but
> hadn't gotten around to it yet.
>
> > Again, I'm not arguing in favor of doing it this way, I'm just
> > explaining it. Perhaps if it's more clearly understood, a
> > comparison can be made.
>
> Ok. Then I'll say that I acknowledge that MacPorts could do it your
> way instead of how it does it now, but I don't see any advantage to
> making that change at this point, rather I currently see
> disadvantages, so I don't think MacPorts should change this right
> now, until someone shows why it's better.


Again, I simply offer this an alternative way that leverages the
launchd/launchctl infrastructure.

I'm done with this thread. It's diverged quite a bit and not very
productively.


-- 
Paul Beard / www.paulbeard.org/
<paulbeard at gmail.com/paulbeard at mac.com>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.macosforge.org/pipermail/macports-users/attachments/20071001/136f63f0/attachment.html