Important .bashrc guidelines

From oldwiki.scinet.utoronto.ca
Jump to navigation Jump to search

The file .bashrc, located in your home directory, is read-in and executed whenever a bash script or bash shell is started. The exception is for login shells, in which case .bash_profile is started. The default .bash_profile calls .bashrc so that many scinet users never need to edit .bash_profile.

This means in particular that the file .bashrc is read in

  • when you log in
  • when you run compute jobs
  • when you run certain commands such as mpirun
  • at the end of a job, when copying output and error files.

And that both on the GPS and the TCS, which nonetheless often require quite a different setup.

Guidelines

To avoid difficulties, you are advised to follow these guidelines in .bashrc:

  • Sourcing /etc/profile and /etc/bashrc is required for various SciNet routines to work!
  • Do not execute any commands that write to standard output, because that would breaks things like copying-out output files and scp. So for instance, no echo or module list.
  • Load only commonly used modules, or preferrably, do not load any modules in your bashrc.
  • Distinguish, especially for the module command, between the gpc and tcs.
  • If you're trying out different implementations or versions of a module (e.g. mpi), you should definitely not load them in the .bashrc, but explicitly on the command line before compiling and once more explicitly in your job script.
  • To avoid accidentally compiling on the login nodes, you can 'shield' yourself from the compilers by wrapping the module load commands with hostname check in the .bashrc file, e.g. as follows:

<source lang="bash"> HOSTNAME=$(hostname) if [ "${HOSTNAME:0:6}" != "scinet" ]; then

   module load intel openmpi  # and any other modules

fi </source>

Default .bashrc and .bash_profile

The default startup scripts provided by SciNet are as follows. Certain things - like sourcing /etc/profile and /etc/bashrc are required for various SciNet routines to work!

Note that the "module load" command is commented out, as we highly recommend not loading any modules in the .bashrc. The comment merely serves as an example. This is a change from the old default, where the intel compiler and the openmpi library were loaded by default. The new default has been in use for new users since Oct 10, 2013.


.bashrc <source lang="bash"> if [ -f /etc/bashrc ]; then

      . /etc/bashrc

fi

  1. commands which work for both GPC and TCS can go here

HOST=$(uname)

if [ "${HOST}" == "AIX" ]; then

       # do things for the TCS machine
       # user environment for all shells goes here
       # replace colon with your own commands
       :

else

       # do things for the GPC machine
       #module load intel openmpi
       # user environment for all shells goes here
       # replace colon with your own commands
       :

fi </source>

.bash_profile <source lang="bash"> if [ -f /etc/profile ]; then

      . /etc/profile

fi

  1. commands which work for both GPC and TCS can go here

alias passwd='echo "Please use the SciNet portal to change password: https://por tal.scinet.utoronto.ca/change_password"'

HOST=$(uname)

if [ "${HOST}" == "AIX" ] then

       # do things for the TCS machine
       alias llq1='/xcat/tools/tcs-scripts/LL/jobState.sh'
       alias llstat='/xcat/tools/tcs-scripts/LL/jobSummary.sh'
       if [ "${TERM}" = "xterm-color" ]; then
               export TERM=xterm
       fi
       # user environment for login shells goes here
       # replace colon with your own commands
       :

else

       # do things for the GPC machine
       # user environment for login shells goes here
       # replace colon with your own commands
       :

fi

PS1="\h-\$ "

if [ -f ~/.bashrc ]; then

      . ~/.bashrc

fi </source>

Note that .bash_profile reads in the .bashrc, so that one in practice only has to add modify the .bashrc to load standard modules, set up aliases etc..

Distinguishing machines and operating systems in .bashrc and .bash_profile

These above startup scripts work well if you only use the gpc, or the gpc and tcs. If you're also using the more specialize machines like the arc or p7 cluster, or if you're using both centos6 and centos5 nodes (e.g. related to the ATLAS project), then you can distinguish what system you are on and what operating system it is running using the following .bash_profile and .bashrc:


.bashrc <source lang="bash"> if [ -f /etc/bashrc ]; then

      . /etc/bashrc

fi

  1. commands which work for both GPC and TCS can go here

export BASH_ENV=~/.bashrc

  1. what machine/architecture

HOSTNAME=$(hostname) case $(uname -p) in

   'ppc64')    MACHINE='p7';;
   'powerpc')  MACHINE='tcs';;
   'x86_64')   MACHINE='gpc';;
   *)          MACHINE='unknown';;

esac

if uname -r | grep -q el5 then

    OS='centos5'

elif uname -r | grep -q el6 then

    OS='centos6'

elif uname -o | grep -q AIX then

    OS='aix'

else

    OS='unknown'

fi

if [ "${MACHINE}" == "tcs" ] then

       # do things for the TCS machine
       # user environment for all shells goes here
       # replace colon with your own commands
       :

elif [ "${MACHINE}" == "p7" ] then

       # do things for the P7 machine
       module load pe vacpp xlf
       # note that this is an example only. Generally, we do
       # not recommended putting module commands in your .bashrc.
       # user environment for all shells goes here
       # replace colon with your own commands
       :

elif [ "${MACHINE}" == "gpc" ] then

       # do things for the GPC machine
       module load intel openmpi
       # note that this is an example only. Generally, we do
       # not recommended putting module commands in your .bashrc.
       # user environment for all shells goes here
       # replace colon with your own commands
       :
       
       # os specific commands/modules can be loaded next
       if [ "${OS}" == "centos5" ]
       then
           #centos 5 specific modules
           :
       elif [ "${OS}" == "centos6" ]
       then
           #centos 6 specific modules
           :
       fi

else

       echo "Not a recognized system"

fi </source>

.bash_profile <source lang="bash"> if [ -f /etc/profile ]; then

      . /etc/profile

fi

  1. commands which work for both GPC and TCS can go here

alias passwd='echo "Please use the SciNet portal to change password: https://portal.scinet.utoronto.ca/change_password"'

  1. what machine/architecture

HOSTNAME=$(hostname) case $(uname -p) in

   'ppc64')    MACHINE='p7';;
   'powerpc')  MACHINE='tcs';;
   'x86_64')   MACHINE='gpc';;
   *)          MACHINE='unknown';;

esac

if uname -r | grep -q el5 then

    OS='centos5'

elif uname -r | grep -q el6 then

    OS='centos6'

elif uname -o | grep -q AIX then

    OS='aix'

else

    OS='unknown'

fi

if [ "${MACHINE}" == "tcs" ] then

       # do things for the TCS machine
       alias llq1='/xcat/tools/tcs-scripts/LL/jobState.sh'
       alias llstat='/xcat/tools/tcs-scripts/LL/jobSummary.sh'
       if [ "${TERM}" = "xterm-color" ]; then
               export TERM=xterm
       fi
       # user environment for login shells goes here
       # replace colon with your own commands
       :

elif [ "${MACHINE}" == "p7" ] then

       # do things for the P7 machine
       # user environment for login shells goes here
       # replace colon with your own commands
       :

elif [ "${MACHINE}" == "gpc" ] then

       # do things for the GPC machine
       # user environment for login shells goes here
       # replace colon with your own commands
       :
       # os specific commands next
       if [ "${OS}" == "centos5" ]
       then
           #centos 5 specific login commands
           :
       elif [ "${OS}" == "centos6" ]
       then
           #centos 6 specific login commands
           :
       fi

else

       echo "Not a recognized system"

fi

PS1="\h-\$ "

if [ -f ~/.bashrc ]; then

      . ~/.bashrc

fi </source>