aiida-raspa-EQeq

A prototypal package to test a screening study on a MOF database using a state-of-the-art calculation : Equilibration of Charges.

Installation

Virtual environment with Conda

conda create -n aiida python=3.11

Other dependies

Python

pip install -e .

Slurm

Installing and configuring Slurm scheduler on a local computer may be painful. We recommend to use SSH connections to a cluster node, where one can find a configured scheduler is installed (jean-zay:slurm, gricad:oar).

For quick tests, on Ubuntu (22.04 tested), one can use :

sudo apt-get install slurm-wlm

Initialize the aiida framework

We need to install aiidausing a conda environment following the installation procedure. Then, to run code using the pre-defined workchain and calcJob objects in the plugin aiida-raspa, we can follow the following steps using the documentation:

• set up a computer

verdi computer setup --config config/computer_slurm.yml

• create a code

verdi code create core.code.installed
        --label raspa_run_slurm \
        --computer curie \
        --filepath-executable "/opt/raspa_v2.0.48/bin/simulate"

Indicate after --filepath-executable the executable path of RASPA installed in the computer.

or use a existing configuration :

verdi computer setup --config config/code_raspa_slurm.yml

To check how the aiida code looks like, use verdi code show <PK>:

-----------------------  -----------------------------------------
PK                       2240
UUID                     72fc9725-3c2d-4cd0-b6c3-4296d9d1e00d
Type                     core.code.installed
Computer                 localhost_slurm (localhost), pk: 2
Filepath executable      /opt/raspa_v2.0.48/bin/simulate
Label                    raspa_run_slurm
Description              Run a simulation with RASPA
Default calc job plugin  raspa
Use double quotes        False
With mpi
Prepend text             export RASPA_DIR=/opt/raspa_v2.0.48
                         export DYLD_LIBRARY_PATH=${RASPA_DIR}/lib
                         export LD_LIBRARY_PATH=${RASPA_DIR}/lib
Append text
-----------------------  -----------------------------------------

Note : if one want to use the direct scheduler from Aiida, it results in the following example that all processes are ran simultaneously, independently of the number of available CPUs. To avoid this behavior, one use the SLURM scheduler

Example

Run on CoRE MOF 2019

• Input structures : examples/subset-coremof/cif As an input, we provide a folder with 4140 cif files obtained by :

  • downloading the dataset coremof-2019 through MOFXDB (12018 MOFs)
  • converting CIF format using openbabel
  • selecting only the cif files we already have charges calculated using lsmo-epfl/EQeq (4140 MOFs)
  • correcting the name of the CIF file with python check_and_convert_filenames.py. aiida-raspa do not accept nonalphanumeric (exept underscore) in the filename (e.g. j.inoche.2015.06.036_inc06102-mmc1_clean_openbabel.cif).

• Running the calculation command to submit all jobs at once from the root directory:

verdi run eqeq_raspa_screening.py raspa_run_slurm \
    --inputpath examples/subset-coremof/cif_converted/ \
    --num_max 80

To test the submission on a few structures rather than the whole database, one can pass the maximum number of submission with --num_max.

Troubleshooting

Aiida

Sometimes, we need to restart the local database :

pg_ctl -D /home/hardiagon/mylocal_db start

where -D flag denotes the path to database postgresql settings (by default $HOME/mylocal_db) if the database server has been initialized using initdb -D mylocal_db (see also aiida doc).

To restart the rabbitmq server :

rabbitmq-server -detached

To restart the daemon :

verdi daemon start

Then one can check everything is working with verdi status.

Slurm

The default configuration of Slurm using apt might be limited. We noticed for example that we could not run more than one job in the same time and Slurm was chaining jobs. To fix this, we changed the configuration file for Slurm, usually in /etc/slurm/slurm.conf, and add the following line at the end of the file :

SelectType=select/cons_res
SelectTypeParameters=CR_Core

# Define a single compute node with only 40 cpus over the total 40 cpus
NodeName=localhost CPUs=40 Boards=1 SocketsPerBoard=1 CoresPerSocket=40 ThreadsPerCore=1 RealMemory=64031

PartitionName=debug Nodes=localhost Default=YES MaxTime=INFINITE State=UP

Additional scripts

Cif name checker

If you use as inputs CIF files with filenames using characters different from _ and alphanumeric characters, the screening will fail. To fix that, one can copy and rename the problematic CIF filenames with :

check_and_convert_filenames

The input set of CIF files must be present in examples/cif/ and the output CIF files will be copied to examples/cif_converted/.

Visualize output data

It might be a bit complicated to look for the data located in the database for a non-expert of Aiida, the following script allows to quickly identify the path directory of a specific calculation :

calcjob2path

You will need to enter your aiida (database) profile (see verdi profile list) and the PK, the index of your calculation process (e.g. found using verdi process list). It will return the absolute path where the remote data is stored (e.g. /data/hardiagon/.aiida/f7/9b/8e3b-dee2-4f0c-be99-45a2e47ccb21)

Check number of CPUs used

Before running the workflow, run this script in another terminal to follow the CPU usage :

print_cpus_used

After all your simulations have completed, plot the results :

plot_cpu_time ./cpu_usage.log

It only detects the number of CPUs that are busy with processes running the RASPA executable simulate.