Add Cluster singleton with global getter

Author: dhilst

include/cloysterhpc/functions.h

@@ -1,6 +1,7 @@
 #ifndef CLOYSTERHPC_FUNCTIONS_H_
 #define CLOYSTERHPC_FUNCTIONS_H_
 
+#include <cloysterhpc/models/cluster.h>
 #include <cloysterhpc/services/log.h>
 #include <boost/process/child.hpp>
 #include <boost/process/pipe.hpp>
@@ -19,6 +20,9 @@ namespace cloyster {
 extern bool dryRun;
 
 using OS = cloyster::models::OS;
+
+void initClusterSingleton(std::unique_ptr<models::Cluster> cluster);
+models::Cluster& getClusterSingleton();
 std::shared_ptr<cloyster::services::BaseRunner> getRunner();
 std::shared_ptr<cloyster::services::repos::RepoManager> getRepoManager(
     const OS& osinfo);

include/cloysterhpc/models/queuesystem.h

@@ -9,8 +9,6 @@
 #include <fmt/format.h>
 #include <string>
 
-#include <cloysterhpc/functions.h>
-
 // Forward declaration of Cluster
 namespace cloyster::models {
 

include/cloysterhpc/services/shell.h

@@ -23,7 +23,6 @@ using cloyster::models::Cluster;
  */
 class Shell final : public Execution {
 private:
-    const std::unique_ptr<Cluster> m_cluster;
 
     /**
      * @brief Configures SELinux mode.
@@ -171,20 +170,14 @@ class Shell final : public Execution {
     static void disableSELinux();
 
 public:
-    /**
-     * @brief Constructs a Shell object.
-     *
-     * Initializes the Shell object with a reference to a Cluster object.
-     *
-     * @param cluster A reference to a unique pointer managing a Cluster object.
-     */
-    explicit Shell(const std::unique_ptr<Cluster> cluster);
     /**
      * @brief Installs and configures the system.
      *
      * This function performs the installation and configuration processes.
      */
     void install() override;
+
+    Shell();
 };
 
 };

include/cloysterhpc/services/xcat.h

@@ -45,8 +45,6 @@ class XCAT : public Provisioner {
      */
     enum class NodeType { Compute, Service };
 
-    const std::unique_ptr<Cluster>& m_cluster;
-
     struct {
         std::vector<std::string_view> otherpkgs = {};
         // @TODO: We need to support more than one osimage (:
@@ -189,6 +187,9 @@ class XCAT : public Provisioner {
     static void configureEL9();
 
 public:
+    XCAT();
+
+
     /**
      * @brief Download the repositories
      */
@@ -262,8 +263,6 @@ class XCAT : public Provisioner {
      * This function resets the nodes.
      */
     static void resetNodes();
-
-    explicit XCAT(const std::unique_ptr<Cluster>& cluster);
 };
 
 };

src/functions.cpp

@@ -66,6 +66,19 @@ std::shared_ptr<BaseRunner> getRunner()
 using cloyster::models::Cluster;
 using cloyster::services::repos::RepoManager;
 
+static std::unique_ptr<Cluster> clusterSingleton;
+void initClusterSingleton(std::unique_ptr<Cluster> cluster)
+{
+    assert(!clusterSingleton);
+    clusterSingleton = std::move(cluster);
+}
+
+Cluster& getClusterSingleton()
+{
+    assert(clusterSingleton);
+    return *clusterSingleton;
+}
+
 std::shared_ptr<RepoManager> getRepoManager(const OS& osinfo)
 {
     static std::optional<std::shared_ptr<RepoManager>> repoManager

src/main.cpp

@@ -17,6 +17,7 @@
 #include <cloysterhpc/services/shell.h>
 #include <cloysterhpc/verification.h>
 #include <cloysterhpc/view/newt.h>
+#include <cloysterhpc/functions.h>
 #include <internal_use_only/config.hpp>
 #include <regex>
 
@@ -190,8 +191,10 @@ int main(int argc, const char** argv)
             model->dumpData(dumped_answerfile);
         }
 
+        cloyster::initClusterSingleton(std::move(model));
+
         std::unique_ptr<Execution> executionEngine
-            = std::make_unique<cloyster::services::Shell>(std::move(model));
+            = std::make_unique<cloyster::services::Shell>();
 
         executionEngine->install();
 

src/models/slurm.cpp

@@ -6,6 +6,7 @@
 #include <cloysterhpc/models/cluster.h>
 #include <cloysterhpc/models/slurm.h>
 #include <cloysterhpc/services/log.h>
+#include <cloysterhpc/functions.h>
 #include <filesystem>
 
 using cloyster::runCommand;

src/ofed.cpp

@@ -5,6 +5,7 @@
 
 #include <cloysterhpc/functions.h>
 #include <cloysterhpc/ofed.h>
+#include <utility>
 
 using cloyster::runCommand;
 
@@ -24,11 +25,39 @@ gpgkey=https://linux.mellanox.com/public/repo/doca/{0}/GPG-KEY-Mellanox
     return data;
 }
 
+// Return distro to match repositories at
+// https://linux.mellanox.com/public/repo/doca/latest/
+std::string headnodeDistroName()
+{
+    const auto& cluster = cloyster::getClusterSingleton();
+    switch (cluster.getHeadnode().getOS().getDistro()) {
+        // Assuming we'll be using the last distro version
+        case cloyster::OS::Distro::RHEL:
+            return "rhel9.5";
+        case cloyster::OS::Distro::OL:
+            return "ol9.4";
+        case cloyster::OS::Distro::Rocky:
+            return "rockylinux9.2";
+        case cloyster::OS::Distro::AlmaLinux:
+            return "alinux3.2";
+        default:
+            std::unreachable();
+    };
+
+    std::unreachable();
+}
+
 void installMellanoxDoca(cloyster::services::repos::RepoManager& repoManager, const OFED& ofed)
 {
     auto runner = cloyster::getRunner();
+
+    if (runner->executeCommand("modprobe mlx5_core") == 0) {
+        LOG_WARN("mlx5_core module loaded, skiping DOCA setup");
+        return;
+    }
+
     // @FIMXE deduce "rockylinux9.2"
-    auto repoData = docaRepoTemplate(ofed.getVersion(), "rockylinux9.2");
+    auto repoData = docaRepoTemplate(ofed.getVersion(), headnodeDistroName());
     std::filesystem::path path = "/etc/yum.repos.d/mlx-doca.repo";
 
     // Install the repository and enable it

src/services/shell.cpp

@@ -25,8 +25,10 @@
 #include <cloysterhpc/dbus_client.h>
 #include <ranges>
 
+using cloyster::getClusterSingleton;
 using cloyster::OS;
 using cloyster::runCommand;
+using cloyster::models::Cluster;
 
 namespace {
 
@@ -49,12 +51,12 @@ auto getToEnableRepoNames(const OS& osinfo)
     }
 }
 
+Cluster& cluster() { return getClusterSingleton(); }
 }
 
 namespace cloyster::services {
 
-Shell::Shell(std::unique_ptr<Cluster> cluster)
-    : m_cluster(std::move(cluster))
+Shell::Shell()
 {
     // Initialize directory tree
     cloyster::createDirectory(installPath);
@@ -79,7 +81,7 @@ void Shell::configureSELinuxMode()
 {
     LOG_INFO("Setting up SELinux")
 
-    switch (m_cluster->getSELinux()) {
+    switch (cluster().getSELinux()) {
         case Cluster::SELinuxMode::Permissive:
             runCommand("setenforce 0");
             /* Permissive mode */
@@ -104,22 +106,22 @@ void Shell::configureFirewall()
 {
     LOG_INFO("Setting up firewall")
 
-    if (m_cluster->isFirewall()) {
+    if (cluster().isFirewall()) {
         runCommand("systemctl enable --now firewalld");
 
         // Add the management interface as trusted
         runCommand(fmt::format(
             "firewall-cmd --permanent --zone=trusted --change-interface={}",
-            m_cluster->getHeadnode()
+            cluster().getHeadnode()
                 .getConnection(Network::Profile::Management)
                 .getInterface()
                 .value()));
 
         // If we have IB, also add its interface as trusted
-        if (m_cluster->getOFED())
+        if (cluster().getOFED())
             runCommand(fmt::format(
                 "firewall-cmd --permanent --zone=trusted --change-interface={}",
-                m_cluster->getHeadnode()
+                cluster().getHeadnode()
                     .getConnection(Network::Profile::Application)
                     .getInterface()
                     .value()));
@@ -137,15 +139,15 @@ void Shell::configureFQDN()
     LOG_INFO("Setting up hostname")
 
     runCommand(fmt::format(
-        "hostnamectl set-hostname {}", m_cluster->getHeadnode().getFQDN()));
+        "hostnamectl set-hostname {}", cluster().getHeadnode().getFQDN()));
 }
 
 // TODO: Proper file parsing
 void Shell::configureHostsFile()
 {
     LOG_INFO("Setting up additional entries on hosts file")
 
-    const auto& headnode = m_cluster->getHeadnode();
+    const auto& headnode = cluster().getHeadnode();
 
     const auto& ip = headnode.getConnection(Network::Profile::Management)
                          .getAddress()
@@ -165,15 +167,15 @@ void Shell::configureTimezone()
     LOG_INFO("Setting up timezone")
 
     runCommand(fmt::format(
-        "timedatectl set-timezone {}", m_cluster->getTimezone().getTimezone()));
+        "timedatectl set-timezone {}", cluster().getTimezone().getTimezone()));
 }
 
 void Shell::configureLocale()
 {
     LOG_INFO("Setting up locale")
 
     runCommand(fmt::format(
-        "localectl set-locale {}", m_cluster->getLocale().getLocale()));
+        "localectl set-locale {}", cluster().getLocale().getLocale()));
 }
 
 void Shell::disableNetworkManagerDNSOverride()
@@ -271,7 +273,7 @@ void Shell::configureNetworks(const std::list<Connection>& connections)
 
 void Shell::runSystemUpdate()
 {
-    if (m_cluster->isUpdateSystem()) {
+    if (cluster().isUpdateSystem()) {
         LOG_INFO("Checking if system updates are available")
         runCommand("dnf -y update");
     }
@@ -339,7 +341,7 @@ void Shell::configureQueueSystem()
 {
     LOG_INFO("Setting up the queue system")
 
-    if (const auto& queue = m_cluster->getQueueSystem()) {
+    if (const auto& queue = cluster().getQueueSystem()) {
         switch (queue.value()->getKind()) {
             case QueueSystem::Kind::None: {
                 __builtin_unreachable();
@@ -375,14 +377,14 @@ void Shell::configureMailSystem()
 {
     LOG_INFO("Setting up the mail system");
 
-    m_cluster->getMailSystem()->setup();
+    cluster().getMailSystem()->setup();
 }
 
 void Shell::configureInfiniband()
 {
-    const auto& osinfo = m_cluster->getHeadnode().getOS();
+    const auto& osinfo = cluster().getHeadnode().getOS();
     auto repos = cloyster::getRepoManager(osinfo);
-    if (const auto& ofed = m_cluster->getOFED()) {
+    if (const auto& ofed = cluster().getOFED()) {
         LOG_INFO("Setting up Infiniband support")
         ofed->install(*repos); // shared pointer
     }
@@ -421,7 +423,7 @@ void Shell::installDevelopmentComponents()
 
 void Shell::configureRepositories()
 {
-    const auto& osinfo = m_cluster->getHeadnode().getOS();
+    const auto& osinfo = cluster().getHeadnode().getOS();
     auto repos = cloyster::getRepoManager(osinfo);
     // 1. Install files into /etc, these files are the templates
     //    at include/cloysterhpc/repos/el*/*.repo
@@ -437,7 +439,7 @@ void Shell::configureRepositories()
  */
 void Shell::install()
 {
-    auto systemdBus = m_cluster->getDaemonBus();
+    auto systemdBus = cluster().getDaemonBus();
 
     configureSELinuxMode();
     configureFirewall();
@@ -448,17 +450,17 @@ void Shell::install()
     configureTimezone();
     configureLocale();
 
-    configureNetworks(m_cluster->getHeadnode().getConnections());
+    configureNetworks(cluster().getHeadnode().getConnections());
     runSystemUpdate();
-    configureTimeService(m_cluster->getHeadnode().getConnections());
+    configureTimeService(cluster().getHeadnode().getConnections());
     installRequiredPackages();
     configureRepositories();
     installOpenHPCBase();
     configureInfiniband();
 
     // BUG: Broken. Compute nodes does not mount anything.
     NFS networkFileSystem = NFS(systemdBus, "pub", "/opt/ohpc",
-        m_cluster->getHeadnode()
+        cluster().getHeadnode()
             .getConnection(Network::Profile::Management)
             .getAddress(),
         "ro,no_subtree_check");
@@ -467,21 +469,22 @@ void Shell::install()
     networkFileSystem.start();
 
     configureQueueSystem();
-    if (m_cluster->getMailSystem().has_value())
+    if (cluster().getMailSystem().has_value()) {
         configureMailSystem();
+    }
     removeMemlockLimits();
 
     installDevelopmentComponents();
 
     const auto& provisionerName { magic_enum::enum_name(
-        m_cluster->getProvisioner()) };
+        cluster().getProvisioner()) };
 
     LOG_DEBUG("Setting up the provisioner: {}", provisionerName)
     // std::unique_ptr<Provisioner> provisioner;
     std::unique_ptr<XCAT> provisioner;
-    switch (m_cluster->getProvisioner()) {
+    switch (cluster().getProvisioner()) {
         case Cluster::Provisioner::xCAT:
-            provisioner = std::make_unique<XCAT>(m_cluster);
+            provisioner = std::make_unique<XCAT>();
             break;
     }