cronrepo 0.3.3

Maintain a set of cron jobs in your code repository

cronrepo: Maintain a set of cron jobs in your code repository

In Unix conventions, periodic tasks are invoked by cron jobs. These jobs are normally configured by the users interactively. When building a complex system, one usually needs to configure many related cron jobs and to ensure that they are installed when the repository gets deployed. This is cumbersome and error-prone.

The cronrepo system eases that pain.

Cron job files

A directory should be created for the cron job files. They are normally shell scripts, although you can use other types of programs as well. They need to be line-based, and need to allow a comment style headed by "#" (or otherwise allowing such lines to be inserted freely, e.g., through a multi-line string syntax). So using Perl or Python scripts or even Makefile as such jobs are okay.

The cron job files are tagged by "taglines" to tell what cron jobs should be installed on each target. The simplest ones look like this:

# CRON@alice::1-10/2 05 01-07 * 2,4

The above tagline configures a cron job running when the following criteria is matched:

• minute: between 1 to 10, if divisible by 2.
• hour: equals 5
• day: between 01 and 07
• month: any month
• day of week: 2 (Tuesday) and 4 (Thursday)

So it is a job which is invoked 5 times on the first Tuesday and first Thursday of every month, at 05:02, 05:04, 05:06, 05:08 and 05:10.

Not all cron time formats are supported, and the above demonstrated all the supported types.

The "alice" is called the "target" of the cron job. When the cron jobs are installed on the system, one target is installed at a time. This allows you to have a cron directory containing jobs that runs differently on different targets, e.g., different machines.

Multiple taglines may be created for the same target, in the same job file. It is at times handy to be able to differentiate them. We can add a "job ID" to the above line, like this:

# CRON@alice%second:5:11-20/2 05 01-07 * 2,4 + foo bar

The job ID consists of word characters (letters, digits and underscores). The job ID is set as the environment variable CRONREPO_JID during the execution of the job.

The above job shows two more features of the tagline:

• We can put an integer between the double-colon to give a level number to the job. The default level is 0. This is useful in the list-inv command described below.
• We can add parameters to the cron job, by adding it to the tagline after a "+" character. The above job will be executed with two arguments "foo" and "bar".

The cronrepo program

The cronrepo program manages the cron jobs given a directory of cron job files. This is done by the followings:

• Generation and installation

  # cronrepo generate <dir> --target <target>
  # cronrepo install <dir> --target <target>
  

  Generate cron jobs entries and show it on the command line or install them as cron jobs. Only jobs of the specific target are generated or installed. If not specified, generate/install all jobs. The job is started by a "cron runner" generated if you use "install". If the crontab already contains a previous installation, they are updated.

  One of the design objectives is that it is possible to create a cron job which updates the crontab. This is done by a job which calls this after updating the working directory.

• Uninstallation

  # cronrepo uninstall <dir> --target <target>
  

  This undos the modification to your crontab, thus uninstalling the cron job. It also uninstalls the cron runner file generated during installation.

• Listing invocations

  # cronrepo list-inv <dir> --target <target> --minlevel <level> \
        --start <dt> --end <dt>
  

  This lists the expected invocations of the cron job entries that occur between the specified <dt> (datetime, in 'YYYY-mm-ddTHH:MM'), inclusive. Jobs are listed only if it has a level of at least <level>. The output is in a format that you can save and run in the shell.

The cron runner

If you ever written a cron job, you know that the environment as seen by the cron job is quite different from your normal environment: PATH is very simplistic (usually so simplistic that you end up setting up your PATH in your script as the first step), all the environment variables set in your shell init script are not present, etc. To make it easier to use, cronrepo creates an environment before running the cron job files.

More precisely, when you use cronrepo install, a "runner script" is created. All current environment variables (except a few) are converted into variable exporting commands in the runner script, and the current directory is also set in the script. So at the end, the cron jobs will run in a very similar environment as simply running the job on the command line of your terminal running "cronrepo install".

The trampoline

Normally, if a cron job emits output, notification will be sent to the owner of the cron job via E-mail. You can globally disable this feature (by setting variables in the crontab), but having nobody to take care of cron job failures is not a good idea.

You can write a program to be run by the runner to actually run your cron jobs, and have the runner file to run that instead of the cron job. This is done by adding --trampoline "your_program" when you run cronrepo install. The arguments to your_program is simply the path to the cron job file, followed by all the arguments to be passed to your job (as specified in + ... in the cron job file).

The default trampoline: cronrepo_run

If you do not want to code your own trampoline, the cronrepo system provides a default trampoline called cronrepo_run. By default it simply exec the program specified. But if the cronrepo.rc file exists, it looks for lines like the following to monitor the execution:

• LOG=<CRONREPO_LOG> (mandatory): Define the log directory, which is used to contain three types of files:
  • The standard output and error streams are redirected to the file <CRONREPO_LOG>/<CRONREPO_NAME>.log (see below for the definition of <CRONREPO_NAME>). If the file already exists, it is rotated.
  • An empty file <CRONREPO_LOG>/<CRONREPO_NAME>.running is created when the job starts running. It is renamed to <CRONREPO_LOG>/<CRONREPO_NAME>.completed or <CRONREPO_LOG>/<CRONREPO_NAME>.failed depending on the execution is successful or not. This is for other scripts to detect the current progress of the job.
  • If the job fails, the file <CRONREPO_LOG>/<CRONREPO_NAME>.failed contains a single line containing the exit code, or the negated signal number if negative.
• NOTIFY=<notifier> (optional): Invoke <notifier> if the command failed. The notifier is invoked through the shell, so that you can have a notifier that performs output redirection, uses $CRONREPO_NAME, etc.
• ROTATE=<N> (optional): Maximum number of backup files to keep for the <CRONREPO_LOG>/<CRONREPO_NAME>.log files.

<CRONREPO_LOG> is actually a strftime template: you can use %Y, %m, etc., to specify that the directory is dependent on the date (but not time, the time is always set to 00:00:00). Environment variables and user directories are also expanded. If the directory does not exist it is created.

If the job does not have a job ID, <CRONREPO_NAME> is the name of the cron job file with last . and anything that follows removed. So if the cron job file is job.sh, the log file is job.log. If the job has a job ID, the ID is appended before .log, separated from the cron job file name by %. E.g., if the above job has job ID home, the log file would be job%home.log.

When running the programs (both the cron file and the notifier), some environment variables are defined:

• CRONREPO_LOG: as defined above.
• CRONREPO_NAME as defined above.
• CRONREPO_DATE: the date that defines CRONREPO_LOG, in %Y-%m-%d format.

If cronrepo_run itself has an error, it is written to the standard error stream. If the job runs from cron, it normally ends up in the mailbox. Before installing the job, it is a good idea to test the job by running it from the console. As a reminder, the job ID can be defined with the CRONREPO_JID variable.

The trampoline cronrepo_run is written so that it ignores the signals SIGINT, SIGQUIT, SIGTERM and SIGPIPE. This is to allow users to run the program interactively, and do things like pressing Control-C, pressing Control-\ and exiting the terminal. Such actions affect the program it runs, without causing cronrepo_run to stop and skip logging. Of course if you send a SIGKILL to it, all bets are off.

At times you want to run cronrepo_run interactively, without causing the notifier program to be invoked even if the program fails. To do this, you can add a -d option right after cronrepo_run, like cronrepo_run -d <crondir>/<cronfile> .... In case of errors, the exit code is printed and becomes the exit code of the program itself, but the notifier is not invoked.