JFR Setup on NCSA DeltaAI

Here, we have some instructions on setting up Jobflow-Remote on ﻿ 🤖⁠⁠NCSA DeltaAI⁠ .
Run mkdir ~/.jfremote  and then make a file ~/.jfremote/cms.yaml. Modify the username and password fields in the example bellow. Also update work_dir as desired and the time for the job as needed. For testing purposes, 30 mins is recommended.
The YAML below relies on a  MongoDB Atlas  cluster to use as the MongoStore. The free tier of MongoDB Atlas is fine, but you likely won't be able to store the output of the jobs in the database due to file size limitations (the free tier has a 512 MB limit). For a more performant solution, you can contact NCSA staff to give you access to a virtual machine where you can install and use MongoDB.
If you are making your own MongoDB Atlas database, you will need to swap out the host described below. To find the host, go to Database > Clusters > Connect > Drivers on MongoDB Atlas and copy the analogous part of the connection string.
name: cms
workers:
  basic_vasp:
    type: local
    scheduler_type: slurm
    work_dir: /work/nvme/bems/<USERNAME>/jfr_runs
    max_jobs: 1
    pre_run: |
      source ~/.bashrc
      conda activate cms
      module load vasp/6.5.1_gpu
      export MSGSIZE=16777216
      export ITERS=400
      export QUACC_VASP_PARALLEL_CMD="srun"
      export QUACC_WORKFLOW_ENGINE=jobflow
      export QUACC_CREATE_UNIQUE_DIR=False
    timeout_execute: 60
    max_jobs: 30
    resources:
      nodes: 1
      gres: gpu:1
      mem: 90G
      ntasks_per_node: 1
      cpus_per_task: 1
      partition: ghx4
      time: 00:30:00
      account: bems-dtai-gh
    batch:
      jobs_handle_dir: /work/nvme/bems/<USERNAME>/jfr_handle_dir
      work_dir: /work/nvme/bems/<USERNAME>/jfr_batch_jobs
queue:
  store:
    type: MongoStore
    host: mongodb+srv://<DB_USERNAME>:<DB_PASSWORD>@cluster0.clj61kx.mongodb.net
    database: mofs
    port: 27017
    collection_name: jf_jobs
  flows_collection: jf_flows
  auxiliary_collection: jf_aux
  batches_collection: jf_batches
exec_config: {}
jobstore:
  docs_store:
    type: MemoryStore
  additional_stores:
    data:
      type: MemoryStore
Run jf project check --errors to make sure everything works okay.
In order to have your Jobflow-Remote daemon run on the login node without being automatically killed, you will need to open a ticket with the NCSA staff to request an exemption for the jf and supervisord processes.