JRM Launcher

JRM Launcher is a tool designed to manage and launch Job Resource Manager (JRM) instances across various computing environments, with a focus on facilitating complex network connections in distributed computing setups.

JRM Launcher Deployment Overview

The following flow chart provides an overview of the JRM Launcher deployment process:

This diagram illustrates the key steps involved in setting up and deploying the JRM Launcher, including the setup of both the fw-lpad and fw-agent components, as well as the workflow management process.

Detailed Step-by-Step Guide

Part 1: Setting up JRM Launcher (fw-lpad)

For more detailed instructions on setting up the JRM Launcher, please refer to the fw-lpad/readme.md at jiriaf-fireworks.

1. Install prerequisites:

Have valid NERSC account, ORNL account, or other remote computing site account
Kubernetes control plane installed (Install Kubernetes in Docker (KinD))
kubectl command-line tool installed and configured to interact with your Kubernetes cluster.
Valid kubeconfig file for the Kubernetes cluster
MongoDB (for storing workflow of JRM launches)
Docker
Python 3.9 (for developers)

2. Set up MongoDB for storing Fireworks workflows:

# Create and start a MongoDB container
docker run -d -p 27017:27017 --name mongodb-container \
  -v $HOME/JIRIAF/mongodb/data:/data/db mongo:latest

# Wait for MongoDB to start (about 10 seconds), then create a new database and user
docker exec -it mongodb-container mongosh --eval '
  db.getSiblingDB("jiriaf").createUser({
    user: "jiriaf",
    pwd: "jiriaf",
    roles: [{role: "readWrite", db: "jiriaf"}]
  })
'

3. Prepare the site configuration file:

Use the template in site_config_template.yaml

Create a configuration file for your specific site (e.g., perlmutter_config.yaml or ornl_config.yaml)

Example configurations:

# perlmutter_config.yaml
slurm:
  nodes: 1
  constraint: cpu
  walltime: 00:10:00
  qos: debug
  account: m3792
  reservation: # 100G

jrm:
  nodename: jrm-perlmutter
  site: perlmutter
  control_plane_ip: jiriaf2302
  apiserver_port: 38687
  kubeconfig: /global/homes/j/jlabtsai/run-vk/kubeconfig/jiriaf2302
  image: docker:jlabtsai/vk-cmd:main
  vkubelet_pod_ips:
    - 172.17.0.1
  custom_metrics_ports: [2221, 1776, 8088, 2222]
  config_class:

ssh:
  remote_proxy: jlabtsai@perlmutter.nersc.gov
  remote: jlabtsai@128.55.64.13
  ssh_key: /root/.ssh/nersc
  password:
  build_script:

# ornl_config.yaml
slurm:
  nodes: 1
  constraint: ejfat
  walltime: 00:10:00
  qos: normal
  account: csc266
  reservation: #ejfat_demo

jrm:
  nodename: jrm-ornl
  site: ornl
  control_plane_ip: jiriaf2302
  apiserver_port: 38687
  kubeconfig: /ccsopen/home/jlabtsai/run-vk/kubeconfig/jiriaf2302
  image: docker:jlabtsai/vk-cmd:main
  vkubelet_pod_ips:
    - 172.17.0.1
  custom_metrics_ports: [2221, 1776, 8088, 2222]
  config_class:

ssh:
  remote_proxy:
  remote: 172.30.161.5
  ssh_key:
  password: < user password in base64 >
  build_script: /root/build-ssh-ornl.sh

4. Prepare necessary files and directories:

Create a directory for logs
Create a port_table.yaml file
Ensure you have the necessary SSH key (e.g., for NERSC access)
Create a my_launchpad.yaml file with the MongoDB connection details:

# my_launchpad.yaml
host: localhost
logdir: <absolute path to logs directory>
mongoclient_kwargs: {}
name: jiriaf
password: jiriaf
port: 27017
strm_lvl: INFO
uri_mode: false
user_indices: []
username: jiriaf
wf_user_indices: []

5. Copy the kubeconfig file to the remote site:

scp /path/to/local/kubeconfig user@remote:/path/to/remote/kubeconfig

6. Start the JRM Launcher container:

export logs=/path/to/your/logs/directory
docker run --name=jrm-fw-lpad -itd --rm --net=host \
  -v ./your_site_config.yaml:/fw/your_site_config.yaml \
  -v $logs:/fw/logs \
  -v `pwd`/port_table.yaml:/fw/port_table.yaml \
  -v $HOME/.ssh/nersc:/root/.ssh/nersc \
  -v `pwd`/my_launchpad.yaml:/fw/util/my_launchpad.yaml \
  jlabtsai/jrm-fw-lpad:main

7. Verify the container is running:

docker ps | grep jrm-fw-lpad

8. Log into the container:

docker exec -it jrm-fw-lpad /bin/bash

9. Add a workflow:

./main.sh add_wf /fw/your_site_config.yaml

10. Note the workflow ID provided for future reference

Part 2: Setting up FireWorks Agent (fw-agent) on Remote Compute Site

For more detailed instructions on setting up the FireWorks Agent, please refer to the fw-agent at jiriaf-fireworks.

1. SSH into the remote compute site

2. Create a new directory for your FireWorks agent:

mkdir fw-agent
cd fw-agent

3. Create the requirements.txt file to this directory

click==8.0.3
FireWorks==1.9.7
Flask==2.0.2
flask-paginate==2021.12.28
gunicorn==20.1.0
itsdangerous==2.0.1
Jinja2==3.0.3
MarkupSafe==2.0.1
monty==2023.11.3
prettytable==3.10.0
pymongo==3.12.3
python-dateutil==2.8.2
ruamel.yaml==0.17.20
ruamel.yaml.clib==0.2.6
six==1.16.0
tabulate==0.8.9
tqdm==4.62.3
wcwidth==0.2.13
Werkzeug==2.0.2
requests==2.26.0

4. Create a Python virtual environment and activate it:

python3.9 -m venv jrm_launcher
source jrm_launcher/bin/activate

5. Install the required packages:

pip install -r requirements.txt

6. Create the fw_config directory and necessary configuration files:

mkdir fw_config
cd fw_config

7. Create and configure the following files in the fw_config directory: - my_fworker.yaml:

# For Perlmutter:
category: perlmutter
name: perlmutter
query: '{}'

# For ORNL:
# category: ornl
# name: ornl
# query: '{}'

- my_qadapter.yaml:

_fw_name: CommonAdapter
_fw_q_type: SLURM
_fw_template_file: <path to queue_template.yaml>
rocket_launch: rlaunch -c <path to fw_config> singleshot
nodes: 
walltime:
constraint:
account:
job_name:
logdir: <path to logs>
pre_rocket:
post_rocket:

- my_launchpad.yaml:

host: localhost
logdir: <path to logs>
mongoclient_kwargs: {}
name: jiriaf
password: jiriaf
port: 27017
strm_lvl: INFO
uri_mode: false
user_indices: []
username: jiriaf
wf_user_indices: []

- queue_template.yaml:

#!/bin/bash -l

#SBATCH --nodes=$${nodes}
#SBATCH --ntasks=$${ntasks}
#SBATCH --ntasks-per-node=$${ntasks_per_node}
#SBATCH --cpus-per-task=$${cpus_per_task}
#SBATCH --mem=$${mem}
#SBATCH --gres=$${gres}
#SBATCH --qos=$${qos}
#SBATCH --time=$${walltime}
#SBATCH --partition=$${queue}
#SBATCH --account=$${account}
#SBATCH --job-name=$${job_name}
#SBATCH --license=$${license}
#SBATCH --output=$${job_name}-%j.out
#SBATCH --error=$${job_name}-%j.error
#SBATCH --constraint=$${constraint}
#SBATCH --reservation=$${reservation}

$${pre_rocket}
cd $${launch_dir}
$${rocket_launch}
$${post_rocket}

8. Test the connection to the LaunchPad database:

lpad -c <path to fw_config> reset

If prompted "Are you sure? This will RESET your LaunchPad. (Y/N)", type 'N' to cancel

9. Run the FireWorks agent:

qlaunch -c <path to fw_config> -r rapidfire

Managing Workflows and Connections

Use the following commands on the fw-lpad machine to manage workflows and connections:

Delete a workflow:

./main.sh delete_wf <workflow_id>

Delete ports:

./main.sh delete_ports <start_port> <end_port>

Connect to database:

./main.sh connect db /fw/your_site_config.yaml

Connect to API server:

./main.sh connect apiserver 35679 /fw/your_site_config.yaml

Connect to metrics server:

./main.sh connect metrics 10001 vk-node-1 /fw/your_site_config.yaml

Connect to custom metrics:

./main.sh connect custom_metrics 20001 8080 vk-node-1 /fw/your_site_config.yaml

Troubleshooting

Check logs in the LOG_PATH directory for SSH connection issues
Ensure all configuration files are correctly formatted and contain required fields
Verify that necessary ports are available and not blocked by firewalls
For fw-agent issues:
- Ensure the FireWorks LaunchPad is accessible from the remote compute site
- Verify that the Python environment has all necessary dependencies installed
Consult the FireWorks documentation for more detailed configuration and usage information

For more detailed troubleshooting information, please refer to the readme at jiriaf-fireworks.

Network Architecture

The core functionality of JRM Launcher revolves around managing network connections between different components of a distributed computing environment. The network architecture is visually represented in the following flow chart.

This diagram illustrates the key components and connections managed by JRM Launcher (JRM-FW), including:

SSH connections to remote servers
Port forwarding for various services
Connections to databases, API servers, and metrics servers
Workflow management across different computing nodes

JRM Launcher acts as a central management tool, orchestrating these connections to ensure smooth operation of distributed workflows and efficient resource utilization.

Key Features

Workflow management
Flexible connectivity to various services
Site-specific configurations
SSH integration and port forwarding
Port management for workflows
Extensibility to support new computing environments

Extensibility

JRM Launcher is designed to be easily extensible to support various computing environments. For information on how to add support for new environments, refer to the "Customization" section in the fw-lpad at jiriaf-fireworks.

Deploy JRMs on NERSC and ORNL via Fireworks

Contents