Merge pull request #3268 from teojgo/ci/slurm_multinode

[ci] Run scheduler unittests with GitHub Actions

.github/pseudo-cluster/docker-compose.yml

@@ -0,0 +1,114 @@
+services:
+  munge-key-generator:
+    image: ghcr.io/reframe-hpc/munge-ubuntu:20.04
+    hostname: munge-host
+    healthcheck:
+        test: ["CMD-SHELL", "test -f /scratch/munge.key"]
+        interval: 10s
+        timeout: 10s
+        retries: 5
+    volumes:
+      - shared-scratch:/scratch
+
+  frontend:
+    image: slurm-reframe
+    container_name: frontend
+    build:
+      dockerfile: .github/pseudo-cluster/reframe/Dockerfile
+      context: ../../
+    hostname: login
+    user: admin
+    init: True
+    volumes:
+      - shared-home:/home/admin:rw
+      - shared-scratch:/scratch:rw
+    links:
+      - slurm-master
+    depends_on:
+      munge-key-generator:
+        condition: service_healthy
+      slurm-master:
+        condition: service_started
+      node0:
+        condition: service_started
+      node1:
+        condition: service_started
+      node2:
+        condition: service_started
+    environment:
+      - SLURM_CPUS_ON_NODE=1
+      - BACKEND=${BACKEND:-squeue}
+
+  slurm-master:
+    image: ghcr.io/reframe-hpc/slurm-master-ubuntu:20.04
+    hostname: slurm-master
+    user: admin
+    volumes:
+      - shared-home:/home/admin
+      - shared-scratch:/scratch:rw
+    depends_on:
+      munge-key-generator:
+          condition: service_healthy
+    environment:
+      - SLURM_CPUS_ON_NODE=1
+
+  node0:
+    image: ghcr.io/reframe-hpc/slurm-node-ubuntu:20.04
+    hostname: nid00
+    container_name: slurm-node0
+    user: admin
+    volumes:
+      - shared-home:/home/admin
+      - shared-scratch:/scratch:rw
+    environment:
+      - SLURM_NODENAME=nid00
+      - SLURM_CPUS_ON_NODE=1
+    depends_on:
+      munge-key-generator:
+          condition: service_healthy
+      slurm-master:
+          condition: service_started
+    links:
+      - slurm-master
+
+  node1:
+    image: ghcr.io/reframe-hpc/slurm-node-ubuntu:20.04
+    hostname: nid01
+    container_name: slurm-node1
+    user: admin
+    volumes:
+      - shared-home:/home/admin
+      - shared-scratch:/scratch:rw
+    environment:
+      - SLURM_NODENAME=nid01
+      - SLURM_CPUS_ON_NODE=1
+    depends_on:
+      munge-key-generator:
+          condition: service_healthy
+      slurm-master:
+          condition: service_started
+    links:
+      - slurm-master
+
+  node2:
+    image: ghcr.io/reframe-hpc/slurm-node-ubuntu:20.04
+    hostname: nid02
+    container_name: slurm-node2
+    user: admin
+    volumes:
+      - shared-home:/home/admin
+      - shared-scratch:/scratch:rw
+    environment:
+      - SLURM_NODENAME=nid02
+      - SLURM_CPUS_ON_NODE=1
+    depends_on:
+      munge-key-generator:
+          condition: service_healthy
+      slurm-master:
+          condition: service_started
+    links:
+      - slurm-master
+
+volumes:
+  shared-home:
+  shared-scratch:

.github/pseudo-cluster/reframe/Dockerfile

@@ -0,0 +1,42 @@
+FROM ubuntu:20.04
+
+ARG DEBIAN_FRONTEND=noninteractive
+
+RUN apt update -y && \
+    apt install -y \
+    build-essential \
+    clang jq libomp-dev tree vim \
+    git \
+    mariadb-client \
+    munge \
+    slurm-client \
+    slurm-wlm-torque \
+    sudo \
+    python3 \
+    python3-pip \
+    wget \
+    curl \
+    mpich \
+    libmpich-dev && \
+    rm -rf /var/lib/apt/lists/*
+
+RUN useradd -m admin -s /usr/bin/bash -d /home/admin && \
+    echo "admin:admin" | chpasswd && adduser admin sudo && \
+    echo "admin     ALL=(ALL) NOPASSWD:ALL" >> /etc/sudoers
+
+COPY .github/pseudo-cluster/reframe/slurm.conf /etc/slurm-llnl/
+COPY .github/pseudo-cluster/reframe/cgroup.conf /etc/slurm-llnl/
+COPY .github/pseudo-cluster/reframe/docker-entrypoint.sh /etc/slurm-llnl/
+COPY . /usr/local/share/reframe
+
+RUN mkdir /scratch && \
+    chown -R admin:admin /scratch
+
+RUN chmod +rx /etc/slurm-llnl/docker-entrypoint.sh
+
+WORKDIR /home/admin
+
+ENV USER admin
+ENV SHELL bash
+
+ENTRYPOINT ["/etc/slurm-llnl/docker-entrypoint.sh"]

.github/pseudo-cluster/reframe/cgroup.conf

@@ -0,0 +1 @@
+../../../examples/tutorial/dockerfiles/slurm-cluster/reframe/cgroup.conf

.github/pseudo-cluster/reframe/docker-entrypoint.sh

@@ -0,0 +1,24 @@
+#!/bin/bash
+
+trap exit 0 INT
+
+while [ ! -f /scratch/munge.key ]
+do
+  sleep 1
+done
+
+sudo cp /scratch/munge.key /etc/munge/munge.key
+sudo service munge start
+sudo sed -i "s/REPLACE_IT/CPUs=${SLURM_CPUS_ON_NODE}/g" /etc/slurm-llnl/slurm.conf
+
+# Needs to be copied in the shared home directory
+cp -r /usr/local/share/reframe .
+cd reframe
+./bootstrap.sh
+
+echo "Running unittests with backend scheduler: ${BACKEND}"
+
+tempdir=$(mktemp -d -p /scratch)
+TMPDIR=$tempdir ./test_reframe.py -v \
+    --rfm-user-config=ci-scripts/configs/ci-cluster.py \
+    --rfm-user-system=pseudo-cluster:compute-${BACKEND:-squeue}

.github/pseudo-cluster/reframe/slurm.conf

@@ -0,0 +1 @@
+../../../examples/tutorial/dockerfiles/slurm-cluster/reframe/slurm.conf

.github/workflows/test-schedulers.yaml

@@ -0,0 +1,24 @@
+name: ReFrame CI / Scheduler backend tests
+on:
+  pull_request: []
+
+jobs:
+  scheduler-test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        scheduler: ['pbs', 'squeue', 'torque']
+    steps:
+    - uses: actions/checkout@v4
+    - name: Login to GitHub Container Registry
+      uses: docker/login-action@v2
+      with:
+        registry: ghcr.io
+        username: ${{ github.actor }}
+        password: ${{ secrets.GITHUB_TOKEN }}
+    - name: Build Images
+      run: |
+        docker compose -f .github/pseudo-cluster/docker-compose.yml build
+    - name: Run Unittests with ${{ matrix.scheduler }} sceduler
+      run: |
+        BACKEND=${{ matrix.scheduler }} docker compose -f .github/pseudo-cluster/docker-compose.yml up --abort-on-container-exit --exit-code-from frontend