Hacker News new | past | comments | ask | show | jobs | submit login
Cron best practices (sanctum.geek.nz)
97 points by leejo on May 8, 2016 | hide | past | web | favorite | 19 comments



The crontab command has one of the worst UI problems:

1. "crontab -e" lets you edit your crontab file, and "crontab -r" deletes it (without asking). The two keys are next to each other.

2. "crontab" (no arg) reads the contents of the crontab file from stdin and replaces it. When you accidentally hit Ctrl-D after it, your crontab is replaced with an empty file.


man pages could use a new section that literally contains @euske comment.

The new section should be called: 'Gotcha'.


I did the "crontab -r" accidentally today . luckily I had a backup.


'luckily' - that's the strangest spelling of 'sanely' I've seen yet.


Cronic (http://habilis.net/cronic/) has been a real life-saver when it comes to using cron in production services. I've been using it for 4+ years without looking back.


There's also "chronic" in Moreutils https://joeyh.name/code/moreutils/ which seems to serve a similar purpose.


In the same-ish vein, may I suggest http://ikura.co for anyone wishing to link timely jobs w/ their web applications.


I never edit a crontab directly. I use external files and reinstall them with "crontab filename" as needed. Then of course the files can be revision-controlled.

Also, I name the files according to host so I remember exactly where they were installed in a networked environment (e.g. "host1.crontab", "host2.crontab").


Interesting, newbie question: where do you put the x.crontab file and how do you get the system to run it? Thanks


I put per-user crontab stuff in ~/.cron/ not only cron-specific scripts go there, but ~/.cron/crontab is the (inactive) master copy of the crontab itself.

Make changes active with `crontab ~/.cron/crontab`

Diff with `diff -du <(crontab -l) <(cat ~/.cron/crontab)`

For servers or not-my-own-user crontabs, config management is the way to go. The crontabs should be in a config management repository, so there's not just a single copy that's deployed and active.


x.crontab can live wherever you'd like; you can even delete it after you install the crontab. When you run the following command

  crontab x.crontab
your crontab file is replaced by the contents of x.contrab; or if you want to install x.crontab for another user

  crontab -u www-user x.crontab
The actual location of x.crontab does not matter as it is just used to setup your cron jobs, but the system does not actually read from this file to run the jobs.


0. Use systemd timers instead.

OnCalendar= has a less insane format than cron, exit code and output logging, locking, proper command line tools, and other niceties like Persistent= (catch up e.g. after system was powered off) etc.

(And yes, you can still have user "cronjobs" if you enable systemd user session lingering.)


There are lots of ways of scheduling stuff out there. The systemd stuff is Linux only, so there isn't really any good reason to bother to learn it for trivial stuff like timed jobs. Cron is much the same everywhere.


It depends how frequently you use a server OS that isn't UNIX. SmartOS and OpenBSD are great, but they're not popular.


I mean isn't Linux, obviously.


Having to create two files for a single task must get old fast though.


Which is why higher-level tools for system administration are awesome. NixOS will generate the .timer and .service files from two options sitting next to each other in a single file (script and startOn) - and you can put them right next to your application service definition and web server configuration. I assume there's systemd unit generators for Ansible, Puppet, etc etc too.


A few notes that I think are worth going over:

1. mysqldump can require quite a few more permissions than mentioned, depending on your utilization of mysql

2. The best way I've found to test the cron-like environmentis by following this: http://stackoverflow.com/questions/2135478/how-to-simulate-t...

Otherwise you run the risk of .profile (and similar) being loaded, where they will not be when run in the cron task.

3. Extra emphasis on using a lock file to avoid overlapping cron tasks!

4. I love using `time` or similar means to pass execution time somewhere (a log file, or a web hook), which helps with point 3.


Cron best practices: use anything else.




Applications are open for YC Winter 2020

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: