C-LAM

C-LAM - Cloud-computing based template for NWP Local Area Model.

The work was an outcome of the ECMWF Code for Earth 2026 edition.

It is an automated installation workflow for building a reproducible WRF/WPS environment on Virtual Machines, with a primary focus on the European Weather Cloud.

Two deployment modes with the same Ansible playbook

Selected by the boolean clam_use_container in group_vars/all.yml (default false). The two role/variable sets are disjoint.

* intel_mpi only runs with clam_install_oneapi=true (multinode hosts); eccodes only with install_eccodes=true.

Requirements

Control node (where you run Ansible):

• ansible-core only; the playbook uses ansible.builtin modules exclusively, so no Galaxy collections are required.

  sudo apt install -y ansible        # or: pipx install ansible-core
  

• For container mode: a prebuilt SIF (clam-wrf.sif), produced once via ./container/build_sif.sh.

Target hosts:

• Ubuntu 24.04 (enforced by a pre_tasks assertion).
• SSH access as a normal user with passwordless sudo (play runs as the ordinary user; do not use --become for the whole play).
• The target may be the control node itself (localhost ansible_connection=local).
• Source build needs ~60 GB under arw_root and several hours (run under tmux/screen/nohup).

Quickstart

a. Source build deployment

cd ansible
# Full stack (open-data variant + ecCodes by default):
ansible-playbook site.yml -e clam_use_container=false

# On EWC, stack on the data volume:
ansible-playbook site.yml -e clam_use_container=false -e arw_root=/data/ARW

# Run a single step via tags:
ansible-playbook site.yml -e clam_use_container=false --tags oneapi

Tags: packages, oneapi, libs, wrf, wps, wrf_wps, geog, eccodes.

b. Pre-built container deployment

# From repo root build the SIF once:
./container/build_sif.sh                      # produces clam-wrf.sif

cd ansible
# Install Apptainer + deploy SIF and run scripts to all hosts:
ansible-playbook site.yml -e clam_use_container=true

# Multi-node hosts: also install the host Intel MPI launcher:
ansible-playbook site.yml -e clam_use_container=true -e clam_install_oneapi=true

# Custom SIF path + deploy/data dirs:
ansible-playbook site.yml \
  -e clam_use_container=true \
  -e clam_sif_src=/mnt/data1/clam-wrf.sif \
  -e clam_deploy_dir=/opt/clam -e clam_data_dir=/data/clam

# Build SIF on the target from a registry instead of copying:
ansible-playbook site.yml \
   -e clam_use_container=true \
   -e clam_sif_registry=docker://<registry-url>/<project>/clam-wrf:latest

Tags: apptainer, oneapi, deploy, geog (e.g. --skip-tags geog).

Launch a run after deploy:

# Single node:
IMG=~/clam-ewc/clam-wrf.sif DATA_DIR=~/clam-data ~/clam-ewc/container/run_apptainer.sh
# Multi node:
HOSTS=vm01,vm03 PPN=16 IMG=~/clam-ewc/clam-wrf.sif DATA_DIR=~/clam-data ~/clam-ewc/container/run_multinode.sh

Key variables

Source build deployment mode (group_vars/all.yml):

Container deployment mode :