PENG-2372 patch the demo cluster to install the agents from the Snap …

…Store
omnivector-solutions · Sep 27, 2024 · a3f10d1 · a3f10d1
1 parent 4bcf7b4
commit a3f10d1
Show file tree

Hide file tree

Showing 2 changed files with 20 additions and 57 deletions.
diff --git a/democluster/user-data b/democluster/user-data
@@ -319,6 +319,9 @@ packages:
 - mpich
 - python3-venv
 - apptainer
+- snap:
+  - [vantage-agent, --channel=stable]
+  - [jobbergate-agent, --channel=stable]
 bootcmd:
 - mkdir /run/packer_backup
 - mkdir /run/packer_backup/etc
@@ -390,38 +393,3 @@ runcmd:
   EOF
 
 - systemctl daemon-reload
-
-# create agents' venv
-- /usr/bin/python3 -m venv /srv/jobbergate-agent-venv
-- /srv/jobbergate-agent-venv/bin/pip install -U pip
-- /srv/jobbergate-agent-venv/bin/pip install jobbergate-agent
-- |
-  cat <<EOF > /srv/jobbergate-agent-venv/.env
-  JOBBERGATE_AGENT_X_SLURM_USER_NAME=root
-  JOBBERGATE_AGENT_BASE_API_URL=https://apis.@ENVIRONMENT@vantagehpc.io
-  JOBBERGATE_AGENT_OIDC_DOMAIN=auth.@ENVIRONMENT@vantagehpc.io/realms/vantage
-  JOBBERGATE_AGENT_OIDC_AUDIENCE=https://apis.vantagehpc.io
-  JOBBERGATE_AGENT_OIDC_CLIENT_ID=@CLIENT_ID@
-  JOBBERGATE_AGENT_OIDC_CLIENT_SECRET=@CLIENT_SECRET@
-  JOBBERGATE_AGENT_TASK_JOBS_INTERVAL_SECONDS=30
-  JOBBERGATE_AGENT_TASK_SELF_UPDATE_INTERVAL_SECONDS=30
-  EOF
-  chown root:root /srv/jobbergate-agent-venv/.env
-  chmod 0644 /srv/jobbergate-agent-venv/.env
-
-- /usr/bin/python3 -m venv /srv/vantage-agent-venv
-- /srv/vantage-agent-venv/bin/pip install -U pip
-- /srv/vantage-agent-venv/bin/pip install vantage-agent
-- |
-  cat <<EOF > /srv/vantage-agent-venv/.env
-  VANTAGE_AGENT_BASE_API_URL=https://apis.@ENVIRONMENT@vantagehpc.io
-  VANTAGE_AGENT_OIDC_DOMAIN=auth.@ENVIRONMENT@vantagehpc.io/realms/vantage
-  VANTAGE_AGENT_OIDC_CLIENT_ID=@CLIENT_ID@
-  VANTAGE_AGENT_OIDC_CLIENT_SECRET=@CLIENT_SECRET@
-  VANTAGE_AGENT_TASK_JOBS_INTERVAL_SECONDS=30
-  VANTAGE_AGENT_TASK_SELF_UPDATE_INTERVAL_SECONDS=30
-  EOF
-  chown root:root /srv/vantage-agent-venv/.env
-  chmod 0644 /srv/vantage-agent-venv/.env
-
-- systemctl daemon-reload
diff --git a/public-scripts/deploy-democluster.sh b/public-scripts/deploy-democluster.sh
@@ -62,9 +62,11 @@ launch_instance () {
 
   # Set the environment to the empty string if not supplied
   if [ -z $ENV ]; then
-      ENVIRONMENT=""
+      BASE_API_URL="https://apis.vantagehpc.io"
+      OIDC_DOMAIN="auth.vantagehpc.io/realms/vantage"
   else
-      ENVIRONMENT="${ENV}."
+      BASE_API_URL="https://apis.${ENV}.vantagehpc.io"
+      OIDC_DOMAIN="auth.${ENV}.vantagehpc.io/realms/vantage"
   fi
 
   # Create the cloud-init file and launch the demo cluster instance.
@@ -91,34 +93,27 @@ runcmd:
 
     REAL_MEMORY=\$(free -m | grep -oP '\\d+' | head -n 1)
     sed -i "s|@REAL_MEMORY@|\$REAL_MEMORY|g" /etc/slurm/slurm.conf
-  - |
-    sed -i "s|@CLIENT_ID@|$CLIENT_ID|g" /srv/jobbergate-agent-venv/.env
-    sed -i "s|@CLIENT_SECRET@|$CLIENT_SECRET|g" /srv/jobbergate-agent-venv/.env
-    sed -i "s|@ENVIRONMENT@|$ENVIRONMENT|g" /srv/jobbergate-agent-venv/.env
-  - |
-    sed -i "s|@CLIENT_ID@|$CLIENT_ID|g" /srv/vantage-agent-venv/.env
-    sed -i "s|@CLIENT_SECRET@|$CLIENT_SECRET|g" /srv/vantage-agent-venv/.env
-    sed -i "s|@ENVIRONMENT@|$ENVIRONMENT|g" /srv/vantage-agent-venv/.env
   - systemctl start slurmrestd
   - systemctl restart slurmdbd
   - systemctl restart slurmd
   - sleep 30
   - systemctl restart slurmctld
   - scontrol update NodeName=\$(hostname) State=RESUME
-  - systemctl start jobbergate-agent
-  - systemctl start vantage-agent
+  - snap set vantage-agent base-api-url=$BASE_API_URL
+  - snap set vantage-agent oidc-client-id=$CLIENT_ID
+  - snap set vantage-agent oidc-client-secret=$CLIENT_SECRET
+  - snap set vantage-agent task-jobs-interval-seconds=30
+  - snap set vantage-agent task-self-update-interval-seconds=30
+  - snap set jobbergate-agent base-api-url=$BASE_API_URL
+  - snap set jobbergate-agent oidc-client-id=$CLIENT_ID
+  - snap set jobbergate-agent oidc-client-secret=$CLIENT_SECRET
+  - snap set jobbergate-agent task-jobs-interval-seconds=30
+  - snap set jobbergate-agent task-self-update-interval-seconds=30
+  - snap set jobbergate-agent x-slurm-user-name=root
+  - snap start vantage-agent.start
+  - snap start jobbergate-agent.start
 EOF
 
-  if ! [ -z "${JG_VERSION}" ]; then
-      echo "  - systemctl stop jobbergate-agent" >> /tmp/cloud-init.yaml
-      echo "  - /srv/jobbergate-agent-venv/bin/pip install -U jobbergate-agent==$JG_VERSION" >> /tmp/cloud-init.yaml
-      echo "  - systemctl start jobbergate-agent" >> /tmp/cloud-init.yaml
-  fi
-  if ! [ -z "${VTG_VERSION}" ]; then
-      echo "  - systemctl stop vantage-agent" >> /tmp/cloud-init.yaml
-      echo "  - /srv/vantage-agent-venv/bin/pip install -U vantage-agent==$VTG_VERSION" >> /tmp/cloud-init.yaml
-      echo "  - systemctl start vantage-agent" >> /tmp/cloud-init.yaml
-  fi
   mkdir -p $HOME/democluster/tmp
 
   cat /tmp/cloud-init.yaml | multipass launch -c$(nproc) \