radical-cybertools

All things RCT

Documentation

• Ensemble Toolkit
• Ensemble Toolkit Adaptive Execution Examples
• Workflow diagram

Concurrently run Adrian and Austin scripts via EnTK on Summit

It is assumed that:

• You have a current allocation on Summit
• You replace the location on the share filesystem in the following examples with your own location.

Prepare the virtual environment in which to run EnTK

• NOTE: This is NOT a conda environment and it is still python 2.7

module load py-virtualenv
virtualenv ve/inspire-experiments
. ve/inspire-experiments/bin/activate
git clone [email protected]:radical-cybertools/radical.pilot.git
git checkout experiment/prte_jsrun
git pull
pip install .
radical-stack
pip install radical.entk

• After the installation, check that the installed version of RADICAL-Pilot by running the following:

  radical-stak
  [...]   
  radical.pilot        : 0.70.3-v0.70.3-144-gdf51bd1@experiment-prte_jsrun
  [...]
  

Prepare code and data of the two tasks to run

• NOTE: It is assumed that you have the test dataset test.zip for the Model-generation.

git clone [email protected]:inspiremd/molecular-active-learning.git
git clone [email protected]:inspiremd/Model-generation.git
cp molecular-active-learning /gpfs/alpine/scratch/mturilli1/bip179/
cp Model-generation /gpfs/alpine/scratch/mturilli1/bip179/
mkdir /gpfs/alpine/scratch/mturilli1/bip179/data
cp -r test /gpfs/alpine/scratch/mturilli1/bip179/data

Prepare the wrappers around the two task to run

• Each wrapper sets up a specific environment for each task to run.

git clone [email protected]:inspiremd/radical-inspire.git
cd radical-inspire
mkdir /gpfs/alpine/scratch/mturilli1/bip179/bin/
cp mmgbsa_wrapper.sh /gpfs/alpine/scratch/mturilli1/bip179/bin/
cp run_learner_wrapper.sh /gpfs/alpine/scratch/mturilli1/bip179/bin/
chmod 777 /gpfs/alpine/scratch/mturilli1/bip179/bin/*.sh

Prepare and execute EnTK

• IMPORTANT: Before running the inspire_skeleton_summit_workflow.py script you have to edit the allocation that EnTK will use to submit a pilot job at line 78.
• IMPORTANT: Autotermination of the pilot job on Summit is not implemented. This means that if you want to minimize allocation consumption you should manually terminate your batch job once you EnTK script is done. On summit, you can use bkill <jobid>.

. setup_entk.sh
python inspire_skeleton_summit_workflow.py 

Output example

EnTK session: re.session.login2.mturilli1.018140.0007                               
Creating AppManagerSetting up RabbitMQ system                                 ok
                                                                              ok
Validating and assigning resource manager                                     ok
Setting up RabbitMQ system                                                   n/a
                                                                              ok
create pilot manager                                                          ok
submit 1 pilot(s)
        [ornl.summit_prte:336]
                                                                              ok    
Update: All components created
MD state: SCHEDULING
Update: MD.rescoring state: SCHEDULING
Update: MD.rescoring.task.0000 state: SCHEDULING
Update: ML state: SCHEDULING
Update: ML.learning state: SCHEDULING
Update: ML.learning.task.0001 state: SCHEDULING
Update: MD.rescoring.task.0000 state: SCHEDULED
Update: ML.learning.task.0001 state: SCHEDULED
Update: MD.rescoring state: SCHEDULED
Update: ML.learning state: SCHEDULED
                                                                              ok
add 1 pilot(s)                                                                ok
Update: MD.rescoring.task.0000 state: SUBMITTING
Update: ML.learning.task.0001 state: SUBMITTING
submit 2 unit(s)
        ..                                                                    ok
Update: MD.rescoring.task.0000 state: EXECUTED
Update: MD.rescoring.task.0000 state: DONE
Update: MD.rescoring state: DONE
Update: MD state: DONE
Update: ML.learning.task.0001 state: EXECUTED
Update: ML.learning.task.0001 state: DONE
Update: ML.learning state: DONE
Update: ML state: DONE
wait for 1 pilot(s)
                                                                              ok
closing session re.session.login2.mturilli1.018140.0007                        \
close pilot manager                                                            \
wait for 1 pilot(s)
                                                                         timeout
                                                                              ok
session lifetime: 171.1s                                                      ok
All components terminated
wait for 1 pilot(s)
                                                                         timeout
All components terminated