Atomate2 Workflows

Atomate2 Workflows#

This guide discusses the steps to configure computational workflows on HPCs, mainly using pymatgen, jobflow, atomate2 and other related packages.

Note

The below instruction is for the HPE DSI Carya and Sabine cluster. It may need slight adjustments for other clusters.

Note

Make sure to have your database credentials ready before moving forward.

Setting up a conda environment#

Log into the cluster, and then create a conda environment:

$ conda create --name ht_mat

Activate the environment:

$ conda activate ht_mat

ht_mat is the name of the conda environment. Feel free to choose your own.

Install some basic packages:

$ conda install -c conda-forge pip pymatgen boto3

See Conda Environment for more info on using conda.

Installing workflow packages#

$ pip install atomate2

Configuring database outputs#

Create a directory, e.g. in your home directory, to store all the config files.

$ cd ~ && mkdir ht_config_files
$ cd ht_config_files

Let’s call this the <CONFIG_DIR>. In below, each time we use <CONFIG_DIR>, you should provide the full path, not the relative path. To get the full path, do $ pwd, and you should see something like:

/home/<your_username>/ht_config_files

jobflow.yaml#

This file specifics the MongoDB and MinIO info to store your calculation outputs. Create a jobflow.yaml in the <CONFIG_DIR> directory with the below content:

JOB_STORE:
  docs_store:
    type: MongoURIStore
    uri: <uri>
    database: <database>
    collection_name: jobflow_outputs
  additional_stores:
    data:
      type: S3Store
      endpoint_url: http://10.2.1.14:9000
      bucket: <minio_bucket_name>
      key: blob_uuid
      compress: true
      index:
        type: MongoURIStore
        uri: <uri>
        database: <database>
        collection_name: jobflow_minio_index
        key: blob_uuid
      s3_profile:
        aws_access_key_id: <minio_username>
        aws_secret_access_key: <minio_password>

Tip

Replace all info in bracket <> by your MongoDB and MinIO access credentials. For <uri>, don’t forget to replace <mongo_username> and <mongo_password> as well. Also, ensure the indentation is correct.

JOB_STORE:
  docs_store:
    type: MongoURIStore
    uri: <uri>
    database: <database>
    collection_name: jobflow_outputs
  additional_stores:
    data:
      type: S3Store
      bucket: <minio_bucket_name>
      key: blob_uuid
      compress: true
      index:
        type: MongoURIStore
        uri: <uri>
        database: <database>
        collection_name: jobflow_minio_index
        key: blob_uuid
      s3_profile:
        aws_access_key_id: <minio_username>
        aws_secret_access_key: <minio_password>
      ssh_tunnel:
        type: SSHTunnel
        tunnel_server_address: carya.rcdc.uh.edu:22
        remote_server_address: carya-nfs12:9000
        local_port: 9000
        username: <CARYA_USERNAME>
        password: <CARYA_PASSWORD>

Tip

Replace all info in bracket <> by your MongoDB and MinIO access credentials. For <uri>, don’t forget to replace <mongo_username> and <mongo_password> as well. Also, ensure the indentation is correct.

Also, replace <CARYA_USERNAME> and <CARYA_PASSWORD> by your Carya (yes, not Sabine) username and login password (i.g. your Cougarnet password). This is needed to establish a SSH tunnel to Carya from Sabine to access the MinIO server.

atomate2.yaml#

This file specifics atomate2 settings, the commands to run VASP, update of VASP INCAR parameters, and the maximum allowed number of custodian errors, etc. You can see the full list of available settings in the Atomate2Settings docs. For now we just config the commands used to run VASP.

Create an atomate2.yaml file in the <CONFIG_DIR> directory with the below content:

VASP_CMD: mpirun --bind-to core vasp_std
VASP_GAMMA_CMD: mpirun --bind-to core vasp_gam

Setting environment variables#

The last thing is to set environment variables to point to these files so that joflow and atomate2 can find them.

Edit your ~/.bashrc file and add the below two lines:

export JOBFLOW_CONFIG_FILE=<CONFIG_DIR>/jobflow.yaml
export ATOMATE2_CONFIG_FILE=<CONFIG_DIR>/atomate2.yaml

Remember you need to replace <CONFIG_DIR> by the full path to your config file directory.

Note

After editing ~/.bashrc, you need to $ source ~/.bashrc for the change to take effect.

Configuring pymatgen#

To run VASP workflows, create a ~/.pmgrc.yaml file with the below content:

PMG_VASP_PSP_DIR: /project/wen/commons/vasp/pp

This will allow the use of pymatgen to automatically create pseudopotentials for your VASP calculations.

Running a workflow#

Up to this point, you’ve fully configured the workflow packages. Plase try the example workflows from the atomate2 official tutorials before moving to the next session.

Next step#

Continue with submitting jobs with fireworks.

More information#

atomate2 documentation: https://materialsproject.github.io/atomate2/

Contents