Cronjob information

This page contains information about the cronjobs that are executed periodically to support the development of the mCRL2 toolset. We describe how these cronjobs are set up and how they can be configured. The information on this page is particularly interesting for people who want to maintain, extend or otherwise change the collection of cronjobs that are run.

The current host computer for the cronjobs is pclin164.win.tue.nl. It is expected that they can be migrated to another system without much effort. Please read this document carefully before doing so.

Overview

In essence, the system consists of a collection of Bash scripts that gets executed periodically by the host’s cron daemon. This collection is structured into one main script and a number of job scripts.

Main script

The purpose of the main script is to decide which jobs have to be executed and in what order, and executes these jobs in a proper shell environment. More precisely, its tasks are:

  • To set up any environment variables that are relevant to the job scripts.
  • To update the local SVN working copy of the mCRL2 source tree.
  • To decide which job scripts have to be run.
  • To execute the job scripts in the appropriate order.
  • To log any messages that the job scripts may produce, and upload these logs if necessary.
  • To send e-mail notifications in case certain jobs fail.

The main script gets called by the cron daemon at certain times as indicated by the installed crontab. We refer to the main script itself for more detailed information on its workings. See section Directory structure for its location on the host computer.

Job scripts

Every job script performs a certain, specific task related to the mCRL2 toolset. Examples include building the toolset, generating and uploading library documentation, and executing performance tests on the toolset.

A job script gets called by the main script. It may access environment variables or call Bash functions that have been exported by the main script. Of course, it may also define and use its own, local variables and functions.

When a job fails, the job script should signal this to the main script by a terminating with a non-zero exit status. Otherwise its exit status should be zero. Different non-zero exit statuses may be used to signal different kinds of failures to the main script. Of course, the main script should be made aware of these differences, so that it can take appropriate actions.

A job script should output relevant informative messages to stderr or stdout, so that the main script can redirect these to the appropriate log files. A job script is not supposed to write its own log files, unless this is absolutely necessary.

Directory structure

On the current host computer pclin164.win.tue.nl, all files and directories needed to run the cronjobs are located in /scratch/mcrl2-cron2

The most important subdirectories are the following:

  • install : installation of the mCRL2 toolset;
  • logs : logs from the various job scripts;
  • performance : scripts and collected data of the performance tests;
  • scripts : the cronjob scripts (see section “Overview”)
  • svn : working copy of the mCRL2 SVN source tree;
  • svn_oas : working copy of the mcrl2 directory in the OAS repository.

We describe these in more detail below.

install

The mCRL2 installation is used by the main script to determine the currently installed version of the toolset and by the performance job script for collecting the measurements.

logs

The logs directory is organized in the following way. For every job script, a subdirectory exists with a corresponding name. Whenever that job script is run, a log file is placed in that subdirectory by the main script. This log file will have the following name:

[revision]_[date]_[time].log

where:

  • [revision] is the SVN revision number associated with this run of the main script (see below);
  • [date] is the date on which the main script was called by the cron daemon;
  • [time] is the time on which the main script was called by the cron daemon.

The [revision] is the SVN revision number of the local working copy in the svn directory after the main script has attempted to update it to the latest SVN revision (a.k.a. the HEAD revision). So, if that update failed, [revision] is the revision number of the working copy prior to the attempt to update it by the main script.

Please note that the log messages of updating the SVN working copy by the main script are written to the log file of the build.sh job script.

performance

The most important files and directories in the performance directory are the following:

  • build_data: collected measurement data from building the toolset
  • cases: input files for the performance tests (being symbolic links to files in the local mCRL2 installation)
  • data: collected measurement data from the performance tests
  • make_plots.sh: script for generating tool performance plots from the data in data
  • make_build_plots.sh: script for generating build performance plots from the data in build_data
  • performance.py: the main performance test script

We do not describe the performance tests in more detail here.

scripts

The scripts directory contains the following files:

  • backup.sh: job script for backing up precious cronjob files
  • build_plots.sh: job script for generating build performance plots (calls make_build_plots.sh in performance)
  • cbuild.sh: job script for building the toolset using CMake
  • libdoc.sh: job script for updating online library documentation
  • main.sh: the main script
  • performance.sh: job script for performance tests (calls performance.py and make_plots.sh in performance)
  • source.sh: job script for updating the downloadable source package on the website (calls the script for generating this package in the packaging subdirectory of the SVN working copy)

These scripts should mostly speak for themselves.

svn

The working copy of the entire mCRL2 SVN repository. It is used by the job scripts for various purposes and updated by the main script before running any of the job scripts.

svn_oas

This working copy of the mcrl2 directory of the OAS repository is updated and used by the cbuild job script. It contains the configuration files that are necessary for building the toolset using CMake.

File system permissions

The directory for the cronjob files and all of its contents are shared among the members of the mcrl2 Unix group. Directories have the group sticky bit set, which ensures that any file/directory that is created within that directory will automatically inherit the mcrl2 group ownership.

Any file or directory should have both read and write permissions for group, to ensure that any member of the mcrl2 group has full access to that file or directory. Please note that the system does not give group write permissions to newly created files and directories by default. This can be dealt with in the following ways:

  1. If you only have to create a small number of files/directories, simply set the proper permissions after creation:

    chmod 775 [files]
    
  2. If you have to create a large number of files/directories (or when some external program will be creating those on your behalf), it’s probably more convenient to set your umask beforehand:

    umask 002
    

    Do not forget to set your umask back to its original value when you are done, to prevent any new personal files that you create from getting group write permissions.

If you want your user account to be added to the mcrl2 group, please contact the BCF Unix helpdesk.

Setting up a cronjob

To set up a cronjob on the host machine, login to that machine and run the following command:

crontab -e

Now, add a line to the list of cronjob entries, calling the main script. For example:

0 5,17 * * * /scratch/mcrl2-cron2/scripts/main.sh &> /scratch/mcrl2-cron2/logs/latest.log

This will run the main script every day at 5am and 5pm, and redirect any output messages that are not captured by the main script to the indicated log file.

Note that the script will be run under the user account that executed the crontab command and will inherit that account’s privileges.

Dependencies

System requirements

In order for the cronjobs to work correctly, recent versions of the following software has to be installed on the host computer.

Name Commands
Cron daemon  
Bash shell and builtins cd, echo, export, pwd, ulimit, umask, etc.
Core Linux tools awk, bc, date, find, grep, ps, sed, tar, time, which, xargs, etc.
SSH secure shell client ssh, scp
Mailx client mail
SVN client svn, svnversion
Tools required for building mCRL2 make, gcc, etc.
CMake cmake
Rsync client rsync
Gnuplot gnuplot
Python python
Doxygen doxygen
LaTeX pdflatex, bibtex

These should ideally be installed in locations that are listed in the PATH environment variable by default (e.g. /bin and /usr/bin). Otherwise, the PATH variable should be extended in the main script to include the proper locations.

User requirements

The user account under which the cronjobs are executed, must meet the following requirements:

  • The account must be a member of the mcrl2 Unix group. This is because the scripts need to have read- and write-access to the directory tree. To become a member of that group, please contact the BCF Unix helpdesk.

  • The account must have a directory called backups in its home directory. The backup.sh job script will place an archive containing precious cronjob files in that directory. This archive is overwritten every time the backup.sh script is run. To achieve this, simply run:

    mkdir ~/backups
    
  • The account must be able to login to the faculty’s webserver as user mcrl2 via SSH, ‘’without the need to supply a password’’. This is because the scripts need to be able to upload files to the webserver. Secure, passwordless login via SSH can be set up as follows, using RSA key authentication (DSA is very similar): #. If your account has no SSH key pair yet, run the following command:

      ssh-keygen -t rsa
    
    Save the key to the default location (``$HOME/.ssh/id_rsa``) and do
    **not** enter a passphrase (simply hit enter when prompted, twice).
    
    1. Upload the public key of the account’s SSH key pair to the webserver:

      scp ~/.ssh/id_rsa.pub mcrl2@www.win.tue.nl:~/
      

      You will need to enter the password for the mcrl2 web account here.

    2. Login to the webserver, append the uploaded key to the list of authorized keys and remove the key file:

      ssh mcrl2@www.win.tue.nl
      (enter password)
      cat id_rsa.pub >> .ssh/authorized_keys
      rm id_rsa.pub
      
    3. Log out and log back in using the same command as above. This time, SSH should not prompt you for the password and should log you in immediately.