Merge pull request #925 from barakda/new_runner_add

adding new runner for atom execution

.env

@@ -71,7 +71,7 @@ CEPH_SHA=latest
 CEPH_DEVEL_MGR_PATH=../ceph
 
 # Atom
-ATOM_SHA=3c0b7531fd1022d97d5600a8ead51992e2a40ec0
+ATOM_SHA=2f86e6fd28bb9f067fa3199e2d0041889f55dad5
 
 # Demo settings
 RBD_POOL=rbd

.github/workflows/build-container.yml

@@ -647,43 +647,40 @@ jobs:
   atom:
     needs: [build, build-ceph]
     if: github.repository == 'ceph/ceph-nvmeof'
-    runs-on: ibmcloud-1
+    runs-on: atomRunner
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
 
-      - name: Atom env initialization
-        run: |
-          . .env
-          ACTION_URL="https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }}"
-          ./tests/atom/atomEnvInit.sh $ATOM_SHA $ACTION_URL
-
       - name: Download container images
         uses: actions/download-artifact@v4
         with:
           pattern: container_images_nvmeof
           merge-multiple: true
 
-      - name: Load container images
-        run: |
-          docker load < nvmeof.tar
-          docker load < nvmeof-cli.tar
-
       - name: Cluster build and Atom tests run
-        if: always() || failure()
+        if: always() && github.event_name != 'schedule'
         run: |
           . .env
-          ./tests/atom/clusterBuildTestsRun.sh $NVMEOF_VERSION $CEPH_SHA $ATOM_SHA
+          ACTION_URL="https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+          ./tests/atom/clusterBuildTestsRun.sh $NVMEOF_VERSION $CEPH_SHA $ATOM_SHA $ACTION_URL
+            
+      - name: Cluster build and Atom nightly tests run
+        if: always() && github.event_name == 'schedule'
+        run: |
+          . .env
+          ACTION_URL="https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+          ./tests/atom/clusterBuildTestsRun.sh $NVMEOF_VERSION $CEPH_SHA $ATOM_SHA $ACTION_URL 'nightly'
 
       - name: Atom artifact build
         if: always()
         run: ./tests/atom/cpArtifactAndCleanup.sh
-
+  
       - uses: actions/upload-artifact@v4
         if: always()
         with:
           name: atom-artifact
-          path: /tmp/artifact/*
+          path: /home/cephnvme/artifact.tar.gz
 
   push-images-to-ceph-registry:
     if: github.event_name == 'release'

docker-compose.yaml

@@ -66,7 +66,7 @@ services:
     environment:
       TOUCHFILE: /tmp/ceph.touch
     entrypoint: >-
-      sh -c './vstart.sh --new $$CEPH_VSTART_ARGS &&
+      sh -c './vstart.sh --without-dashboard --new $$CEPH_VSTART_ARGS &&
       echo ceph dashboard nvmeof-gateway-add -i <(echo nvmeof-devel:5500) nvmeof.1 &&
       pushd /etc/ceph &&
       openssl req -x509 -newkey rsa:4096 -nodes -keyout server.key -out server.crt -days 3650 -subj /CN=my.server  -addext "subjectAltName = IP:192.168.13.3, IP:0.0.0.0" &&

tests/atom/clusterBuildTestsRun.sh

@@ -7,6 +7,52 @@ else
     CEPH_SHA=$2
 fi
 ATOM_SHA=$3
+ACTION_URL=$4
+NIGHTLY=$5
+
+RUNNER_FILDER='/home/cephnvme/actions-runner-ceph'
+
+# Check if cluster is busy with another run
+while true; do
+    if [ -f "/home/cephnvme/busyServer.txt" ]; then
+        echo "The server is busy with another github action job, please wait..."
+        sleep 90
+    else
+        echo "The server is available for use!"
+        echo $ACTION_URL > /home/cephnvme/busyServer.txt
+        chmod +rx /home/cephnvme/busyServer.txt
+        break
+    fi
+done
+
+# Remove previous run data
+hostname
+rm -rf $RUNNER_FILDER/ceph-nvmeof-atom
+sudo rm -rf /root/.ssh/atom_backup/artifact/multiIBMCloudServers_m6/*
+sudo ls -lta /root/.ssh/atom_backup/artifact/multiIBMCloudServers_m6
+
+# Cloning atom repo
+cd $RUNNER_FILDER
+git clone git@github.ibm.com:NVME-Over-Fiber/ceph-nvmeof-atom.git
+if [ $? -ne 0 ]; then
+    echo "Error: Failed to clone the atom repository."
+    exit 1
+fi
+
+# Switch to given SHA
+cd ceph-nvmeof-atom
+git checkout $ATOM_SHA
+if [ $? -ne 0 ]; then
+    echo "Error: Failed to checkout the specified SHA."
+    exit 1
+fi
+
+# Build atom images based on the cloned repo
+docker build -t nvmeof_atom:$ATOM_SHA .
+if [ $? -ne 0 ]; then
+    echo "Error: Failed to build Docker image."
+    exit 1
+fi
 
 # Atom test script run
 #   Description of the uncleared flags with their default values
@@ -24,21 +70,45 @@ ATOM_SHA=$3
 #   - RBD size (200M)
 #   - Seed number (0)
 #   - FIO use (1=run fio, 0=don't run fio)
-sudo docker run \
-    -v /root/.ssh:/root/.ssh \
-    nvmeof_atom:"$ATOM_SHA" \
-    python3 cephnvme_atom.py \
-    quay.ceph.io/ceph-ci/ceph:"$CEPH_SHA" \
-    quay.io/ceph/nvmeof:"$VERSION" \
-    quay.io/ceph/nvmeof-cli:"$VERSION" \
-    None None None None None None 4 1 1 2 4 1024 2 2 200M 0 1 20 20 1 \
-    --stopNvmeofDaemon \
-    --stopNvmeofSystemctl \
-    --stopMonLeader \
-    --rmNvmeofDaemon \
-    --gitHubActionDeployment \
-    --dontUseMTLS \
-    --skiplbTest \
-    --journalctlToConsole \
-    --dontPowerOffCloudVMs noKey noKey \
-    --multiIBMCloudServers_m2
+
+set -x
+if [ "$5" != "nightly" ]; then
+    sudo docker run \
+        -v /root/.ssh:/root/.ssh \
+        nvmeof_atom:"$ATOM_SHA" \
+        python3 cephnvme_atom.py \
+        quay.ceph.io/ceph-ci/ceph:"$CEPH_SHA" \
+        quay.io/ceph/nvmeof:"$VERSION" \
+        quay.io/ceph/nvmeof-cli:"$VERSION" \
+        None None None None None None 1 1 4 1 1 2 4 1024 2 2 200M 0 1 20 10 1 \
+        --stopNvmeofDaemon \
+        --stopNvmeofSystemctl \
+        --stopMonLeader \
+        --rmNvmeofDaemon \
+        --gitHubActionDeployment \
+        --dontUseMTLS \
+        --skiplbTest \
+        --journalctlToConsole \
+        --dontPowerOffCloudVMs noKey noKey \
+        --multiIBMCloudServers_m6
+else
+    sudo docker run \
+        -v /root/.ssh:/root/.ssh \
+        nvmeof_atom:"$ATOM_SHA" \
+        python3 cephnvme_atom.py \
+        quay.ceph.io/ceph-ci/ceph:"$CEPH_SHA" \
+        quay.io/ceph/nvmeof:"$VERSION" \
+        quay.io/ceph/nvmeof-cli:"$VERSION" \
+        None None None None None None 1 1 4 1 1 10 90 1024 6 2 200M 0 1 20 10 1 \
+        --stopNvmeofDaemon \
+        --stopNvmeofSystemctl \
+        --stopMonLeader \
+        --rmNvmeofDaemon \
+        --gitHubActionDeployment \
+        --dontUseMTLS \
+        --skiplbTest \
+        --journalctlToConsole \
+        --dontPowerOffCloudVMs noKey noKey \
+        --multiIBMCloudServers_m6
+fi
+set +x

tests/atom/cpArtifactAndCleanup.sh

@@ -1,7 +1,16 @@
 #!/bin/bash
 
-sudo rm -rf /tmp/artifact/multiIBMCloudServers_m2
-sudo cp -r /root/.ssh/atom_backup/artifact/multiIBMCloudServers_m2 /tmp/artifact
-sudo ls -lta /tmp/artifact
-sudo chmod -R +rx /tmp/artifact
-rm -rf /home/cephnvme/busyServer.txt
+sudo rm -rf /home/cephnvme/artifact/*
+sudo ls -lta /home/cephnvme/artifact
+
+sudo rm -rf /home/cephnvme/artifact.tar.gz
+sudo ls -lta /home/cephnvme/
+
+sudo cp -r /root/.ssh/atom_backup/artifact/multiIBMCloudServers_m6 /home/cephnvme/artifact
+sudo ls -lta /home/cephnvme/artifact
+
+sudo tar -czf /home/cephnvme/artifact.tar.gz -C /home/cephnvme/artifact .
+sudo ls -lta /home/cephnvme/artifact
+sudo ls -lta /home/cephnvme
+sudo chmod +rx /home/cephnvme/artifact.tar.gz
+sudo rm -rf /home/cephnvme/busyServer.txt