- launchd-dev - lists.macosforge.org

Disabling of launchd job if program is missing
by Vera Tkachenko 06 Jul '10

06 Jul '10

Hello, in my program I use quite a simple launchd jobs to perform some tasks. This job is configured to run some tool when certain conditions are met. All works fine except one case. When user deletes this tool (e.g. after incorrect uninstall of program), then launchd fails to find tool and therefore fails to run my job. The following error is logged: 25/04/2010 23:26:11 com.apple.launchd.peruser.501[224] (<job label>[89736]) posix_spawn(<path to tool>, ...): No such file or directory The problem is that after this launchd keep trying to run this job. Why it just don't disable this job? If this is desired behavior, is it possible to configure it somehow to disable job if it's tool is missing? This is not just my software problem. Lots of other programs have such issues. -- Thanks, Vera Tkachenko

4 5

Launch-on-Demand WWDC session
by Damien Sorresso 17 Jun '10

17 Jun '10

Hey folks, We're doing another launchd session this year! Whereas last year's was focused on launchd and launch-on-demand from a high-level, architectural perspective, this year's will drill into the details of writing a launch-on-demand daemon. We'll be covering how to effectively use GCD to set up the IPC channel, adopting Instant Off and incorporating privileged helper tools into your project with ServiceManagement. There's a lot of good stuff in this session, so if you'll be at WWDC this year, you won't want to miss it! -- Damien Sorresso BSD Engineering Apple Inc.

3 4

Bug in launchd? Agent sometimes just stops working until unloaded/reloaded
by Jerry Krinock 16 Jun '10

16 Jun '10

[Sorry... posted 3 days ago on darwin-userlevel(a)lists.apple.com before I knew about this list] I've seen several times in the last few months a randomly-occuring bad state in launchd wherein a LaunchAgent will fail silently; i.e. not run its program when triggered. Unloading and reloading the agent immediately knocks it out of this bad state. My app installs a launchd agent with a "WatchPaths" trigger. During a testing session in which the trigger is exercised several times, sometimes, if I change the watched path once again (after waiting the required ThrottleInterval), Finder indicates a new modification date, but launchd doesn't run my program nor log anything to console. Repeatable. While it is in this broken state, if I tell launchctl to 'list' my agent, it appears to be loaded. If I copy the ProgramArguments from the listing and run the target program in Terminal, it executes OK. Unload/reloading the agent like this: Jerrys-Mac-Mini:~ jk$ launchctl unload /Users/jk/Library/LaunchAgents/com.sheepsystems.BookMacster.10015DE4-45B0-4019-BAC7-A07229C22C07.000.001.plist Jerrys-Mac-Mini:~ jk$ launchctl load /Users/jk/Library/LaunchAgents/com.sheepsystems.BookMacster.10015DE4-45B0-4019-BAC7-A07229C22C07.000.001.plist fixes the problem immediately; the agent starts responding as expected to WatchPaths changes once again. The output from 'launchctl list MyAgent' is exactly the same before or after the unload/reload fix: Jerrys-Mac-Mini:~ jk$ launchctl list com.sheepsystems.BookMacster.10015DE4-45B0-4019-BAC7-A07229C22C07.000.001 { "Label" = "com.sheepsystems.BookMacster.10015DE4-45B0-4019-BAC7-A07229C22C07.000.001"; "LimitLoadToSessionType" = "Background"; "OnDemand" = true; "LastExitStatus" = 0; "TimeOut" = 60; "ProgramArguments" = ( "/Users/jk/Documents/Programming/Builds/Debug/BookMacster.app/Contents/Helpers/BookMacster-Worker"; "10015DE4-45B0-4019-BAC7-A07229C22C07"; "0"; ); }; The LimitLoadToSessionType key is mysterious. Besides the fact that I don't comprehend its documentation, this key is not in my plist file, and I do not believe that my code appends a -S option when invoking launchctl to load the job (which it does via NSTask). I was wondering if maybe my rebuilding of the program (i.e. rewriting the executable) was causing the problem. But it does not, at least not reproducibly. Has anyone ever noticed this bug? Can anyone suggest any conditions which might make it reproducible? Any weak spots where I should "push"? Thanks, Jerry Krinock Here's the agent's plist file:

1 0

How can I use launchd to do an "at" job ?
by Dan White 04 May '10

04 May '10

"at" jobs are one-shot cron-like commands <http://unixhelp.ed.ac.uk/CGI/man-cgi?at> I can see how to do a cron-like job, but I see nothing in launchd that lets me run something once at a given date-time. “Sometimes I think the surest sign that intelligent life exists elsewhere in the universe is that none of it has tried to contact us.” Bill Waterson (Calvin & Hobbes)

1 0

launchd check-in returns error 2 (ENOENT)
by Brent Gulanowski 23 Apr '10

23 Apr '10

This is not urgent but I clearly don't understand something about launchd check-in. We have a couple of daemons that are launched in response to a socket connection. Now, I'm going to straight-up guess that my problem is because I'm not playing by the "rules" -- the daemon's are in inetd mode with Wait as NO. So perhaps launchd just assumes the socket is no good to me and therefore nothing else in the job's plist is of any use. Not sure why it would make such an assumption, but I suppose it's because I can easily read the job property list myself from the file system. Although since you can submit a job directly without a property list file, it would be nice to access it through check-in anyway, say, to read the name of the StandardErrorPath. The daemon runs as a dedicated background user in the root launchd domain. check-in returns NULL and errno was not 2 before calling launch_msg(). As an aside, when I run the daemon from Xcode, errno is 1 (EPERM), which seems weird (what don't I have permission to do?). It would be cool if someone would write a good book on launchd. It's so awesome in so many ways, but it's just not getting the support it deserves. It would also be nice if it had a full-featured Cocoa API. -- #pragma mark signature [[self mailClient] send:[Mail messageWithText:@"From: Brent Gulanowski\nTo: You"];

1 0

launchctl - LAUNCHD_SOCKET
by Eric Long 19 Apr '10

19 Apr '10

Hi, I have a daemon and one of its tasks is to check for launchd jobs in user LaunchAgents folders, see if they failed to execute because the machine was off, or whatever, and start the jobs that were missed. These jobs are all directly related to a feature in our software. The daemon runs in the global root context. When it performs checks it iterates jobs for all logged in users. To start a job, I use NSTask to launch launchctl. If I fail to set the euid to the user whose job I want to start, I get an error saying no such process, or something to that effect. When I set the euid before launching the task, I get back an error saying I must be the root user to conduct the operation and also a socket not connected error message. Looking at the man page for launchctl I see an environment variable named LAUNCHD_SOCKET. My guess is that even though I'm launching the task while the euid is set to the correct user, its still not executing in the user context. I'd attempt to set the LAUNCHD_SOCKET path, but I don't know what the path should point to. Looking at the docs, I see: > Each instance of launchd has a directory in /var/launchd, identified by the > uid or by the uid followed by a pid number, containing a socket for > communicating with it. The environment variable LAUNCHD_SOCKET, when set to > the full path of one of these socket files, specifies a specific instance of > launchd to communicate with. I see directories for users, but the directories are empty. What is the proper way to start the missed job for a given user from my daemon? Thanks, Eric

2 1

WatchEvents and unlink
by Alfred Labrie 24 Mar '10

24 Mar '10

Hello It sounds like there is a new behavior with WatchEvents on Snow Leopard. Let's say, I watch for this file path: "/private/tmp/watch" - if I create the file, launch daemon is triggered - if I write + close the file, launch daemon is triggered - if I remove the file then I create a new file, launch daemon is no longer triggered On Leopard, remove then create the file would trigger the launch daemon. Typical watch.plist syntax would be: ... <dict> <key>KeepAlive</key> <false/> <key>Label</key> <string>inc.ninja.watch</string> <key>ProgramArguments</key> <array> <string>/path/to/execute</string> </array> <key>WatchPaths</key> <array> <string>/private/tmp/watch</string> </array> </dict> </plist> This is tested on Mac OS X 10.6.2 and above. Is it by design or is it a bug? Regards.

2 3

Launchd and python
by Michael McCracken 15 Mar '10

15 Mar '10

Hi, I just spent some time figuring out how to get a launchd agent written in python (and using UNIX domain sockets) up and running. I think this was asked previously on the list, and since it wasn't straightforward to figure out how to make it work, I decided to write a post about it and put up some public example code. My post is here: http://michael-mccracken.net/2010/03/snakes-on-demand-how-to-write-a-python… The example code is on bitbucket here: http://bitbucket.org/mikemccracken/py-launchd/ I'd be interested to hear comments on how I've implemented it. In particular: - I use a socket in /private/tmp/ but saw some comments saying that that might not be good for long-lived agents (mine won't be, but for sample code it might be better to use something else?) - To avoid forcing a restart on first-run, I use system() to unload, then load the plist when my app starts up. This smells - what happens if I'm doing this while a second app is talking to a running instance of this agent? Is there a better way? Thanks for any comments - I'm interested in updating my public code with the *right way* once I figure it out :) Cheers, -mike

2 2

Re: [launchd-dev] A question about launchd and cron
by wierzbowski 06 Mar '10

06 Mar '10

> Date: Fri, 5 Mar 2010 23:04:47 +0000 > From: Quinn <eskimo1(a)apple.com> > To: launchd-dev(a)lists.macosforge.org > Subject: Re: [launchd-dev] A question about launchd and cron > Message-ID: <p06240832c7b73e8627e1(a)[10.0.40.15]> > Content-Type: text/plain; charset="us-ascii" ; format="flowed" > > At 14:34 -0800 5/3/10, wierzbowski wrote: > >> Please let me know if this would be an appropriate subject for this list. > > Sure. Fire away! > > S+E For demonstration purposes I've set up a simple crontab and cronjob running as myself (i.e. non-root). Here is the crontab: SHELL="/bin/sh" PATH="/usr/bin:/bin:/usr/sbin:/sbin:/usr/local/bin:/usr/X11/bin" HOME="/Users/wierzbowski" MAILTO="wierzbowski@somewhere" * * * * * /Users/wierzbowski/bin/helloworld Here is the content of /Users/wierzbowski/bin/helloworld: #!/bin/sh /bin/echo "hello world." exit 0 Here is the output when I run the script manually: hello world. Here is the output that is mailed to me when the script runs via cron: hello world. Here is the error recorded in the syslog: Mar 5 12:17:00 macbookpro com.apple.launchd[1] (0x10e4d0.cron[97753]): Could not setup Mach task special port 9: (os/kern) no access Basically, on all the systems I manage I get that "special port 9" error in the syslog every time cron executes a job. And it doesn't seem to matter if it's a root or non-root job, if I have the environment explicit defined or not, if the cronjob is a one-liner right in the crontab or if it calls another shell script, if the script generates output or not, if the script redirects any i/o or not, if the script writes any files or not, etc. I'm just trying to figure out if there's a fundamental flaw in my cron usage on OS X, or if I've hit a bug and I just have to accept a few thousand extra lines in the syslog every day. I also realize that cron has been deprecated in favor or straight up launchd, and I'll get around to converting one of these day I swear, but for the moment I don't have the luxury of bumping that project to the top of my stack. :) I did take a quick look at the launchd source code, and I see where the error message is coming from, but I'm definitely no programmerologist and I have no idea what triggers that particular bit of code. Thanks in advance for any insight. Wierzbowski

1 0

A question about launchd and cron
by wierzbowski 05 Mar '10

05 Mar '10

Greetings! I'm a user, not a developer, and I'd like to ask a question about an error that launchd generates in the syslog when running cronjobs on 10.5.8. I've tried some general purpose Mac OS X forums but haven't found an answer. Please let me know if this would be an appropriate subject for this list. Thank you! Wierzbowski

2 1