Skip to content

Latest commit

 

History

History
81 lines (65 loc) · 3.22 KB

README.md

File metadata and controls

81 lines (65 loc) · 3.22 KB
Raw

HAL::docker

Tool to run binaries in a docker image on the HAL cluster (OAR based)

Permissions

In order to be able to write in the shared /host folder (see --dir option) the script gives to your docker container your uid and gid.

Hence you won't be able to modify the container content but you can still use the already installed libraries

If you want to install another version of one of the library you need to follow the compile_autotune.sh script example.

The principle is to install yout librairy inside the /host folder and to give your project the correct library paths You also need to change the LD_LIBRARY_PATH and PATH accordingly See compile_autotune.sh and run_autotune.sh

Examples

Compile and run in the docker

hal-docker.py --image registry.gitlab.inria.fr/locolearn/public/docker_inria_wbc/inria_wbc:latest  --compile examples/compile.sh  --script examples/test_script.sh --dir /nfs/hal01/jmouret

Compile and submit with 6 replicates

hal-docker.py --image registry.gitlab.inria.fr/locolearn/public/docker_inria_wbc/inria_wbc:latest  --compile examples/compile.sh  --script examples/test_script.sh --dir /nfs/hal01/jmouret --submit --replicates 6 --cores 24 --walltime 1:00

Please note that it is your responsibility to make it sure that the replicates do not overwrite their output files (as they are running concurently). For instance, in your script:

mkdir /host/data/$RESULT_DIR
FILE=/host/data/$RESULT_DIR/test_result_talos-`date +%m-%d-%H-%M-%S`-$RAND.dat
cd tests
./test_behaviors_talos > $FILE

$RAND is filled by our system to give you a random number between 1 and 10000.

You can create directories using the same principle.

Kill a job

Do not use oardel directly because it does not clean properly the docker containers!

hal-docker.py --kill  745342

Help

usage: hal-docker.py [-h] [-a ARGS [ARGS ...]] [-c CMD] [-C COMPILE] [-d DIR]
                     [-i IMAGE] [-k KILL [KILL ...]] [-n CORES]
                     [-r REPLICATES] [-R RESULT_DIR] [-S SCRIPT] [-s] [-v]
                     [-w WALLTIME]

optional arguments:
  -h, --help            show this help message and exit
  -a ARGS [ARGS ...], --args ARGS [ARGS ...]
                        arguments to the run script
  -c CMD, --cmd CMD     command to run
  -C COMPILE, --compile COMPILE
                        compilation script
  -d DIR, --dir DIR     directory to be mounted as /host
  -i IMAGE, --image IMAGE
                        DOCKER image
  -k KILL [KILL ...], --kill KILL [KILL ...]
                        kill a job [--kill job_number]
  -n CORES, --cores CORES
                        number of cores for OAR [use 24 for a full node, 32
                        for the 'fat nodes']
  -r REPLICATES, --replicates REPLICATES
                        number of replicates
  -R RESULT_DIR, --result_dir RESULT_DIR
                        set $RESULT_DIR in your execution script
  -S SCRIPT, --script SCRIPT
                        script to compile and run
  -s, --submit          submit the job to OAR
  -v, --verbose         print more information
  -w WALLTIME, --walltime WALLTIME
                        walltime for the OAR job (warning the job will be
                        killed after this time)