DAOSGCP-224 Updates required for HPC Toolkit hpc-slurm6-daos example (#…

…99)

Added retry logic for daos-* package installs

Commented out hard coded network interface config in daos_client.yml since
slurm instances do not force eth0 name for primary ethernet interface.

Signed-off-by: Mark Olson <115657904+mark-olson@users.noreply.github.com>

terraform/modules/daos_server/scripts/client_install.sh

@@ -59,18 +59,35 @@ install_epel() {
 add_daos_repo() {
   local repo_file="${REPO_PATH}/daos.repo"
   rm -f "${repo_file}"
-  echo "Adding DAOS v${DAOS_VERSION} packages repo"
+  echo "Downloading DAOS v${DAOS_VERSION} packages repo file"
   curl -s -k --output "${repo_file}" "https://packages.daos.io/v${DAOS_VERSION}/${DAOS_OS_VERSION}/packages/x86_64/daos_packages.repo"
-  if [[ "${OS_VERSION_ID}" == "opensuse-leap_15" ]]; then
-    sed -i 's|gpgkey=.*|gpgkey=https://packages.daos.io/RPM-GPG-KEY|g' "${repo_file}"
+  if [[ -f ${repo_file} ]]; then
+    echo "Download of DAOS v${DAOS_VERSION} packages repo file was successful"
+  else
+    echo "Download of DAOS v${DAOS_VERSION} packages repo file failed. Exiting."
+    exit 1
   fi
 }
 
 install_daos_client() {
-    "${PKG_MGR}" install -y daos-client daos-devel
-    # Disable daos_agent service.
-    # It will be enabled by a startup script after the service has been configured.
-    systemctl disable daos_agent
+    local max_attempts=5
+    local attempt_num=1
+    local success="false"
+    while [[ "${success}" == "false" ]] && [ $attempt_num -le $max_attempts ]; do
+      echo "Installing DAOS v${DAOS_VERSION} admin, client, and develop packages. Attempt: ${attempt_num}"
+      "${PKG_MGR}" install -y daos-admin daos-client daos-devel
+      if [[ $? -eq 0 ]]; then
+        success="true"
+        echo "DAOS admin, client, and develop packages installed successfully"
+        echo "Disabling daos_agent service"
+        echo "daos_agent service will be enabled after the service has been configured."
+        systemctl disable daos_agent
+      else
+        echo "DAOS client install attempt ${attempt_num} failed. Sleeping for 30 seconds before retry ..."
+        ((attempt_num++))
+        sleep 30
+      fi
+    done
 }
 
 main() {

terraform/modules/daos_server/templates/daos_agent.yml.tftpl

@@ -14,8 +14,8 @@ transport_config:
   key: /etc/daos/certs/agent.key
   %{ endif }
 
-fabric_ifaces:
-- numa_node: 0
-  devices:
-  - iface: eth0
-    domain: eth0
+# fabric_ifaces:
+# - numa_node: 0
+#   devices:
+#   - iface: eth0
+#     domain: eth0