Allowing definition of k3s node name (#1517)

Author: jcchr

platform/services/installer/app/commands/upgrade.py

@@ -52,7 +52,7 @@
 from platform_stages.steps.install_system_packages import install_system_packages
 from platform_stages.upgrade import upgrade_platform
 from platform_utils.grafana import flush_ingesters as flush_lgtm_stack
-from platform_utils.k8s import check_if_telemetry_stack_is_installed
+from platform_utils.k8s import check_if_telemetry_stack_is_installed, get_node_name
 from platform_utils.management.data_folder import backup_data_folder, restore_data_folder
 from platform_utils.management.management import restore_platform, stop_platform
 from platform_utils.management.state import (
@@ -408,9 +408,16 @@ def perform_k3s_upgrade(config: UpgradeConfig) -> None:
     logger.info("Upgrading k3s.")
     click.echo(UpgradeCmdTexts.k3s_upgrade)
     logger.info(UpgradeCmdTexts.k3s_upgrade)
+
     try:
+        node_name = get_node_name()
+
+        if not node_name:
+            logger.error("Node name could not be determined.")
+            raise K3SInstallationError("Node name could not be determined.")
+
         with click_spinner.spinner():
-            install_k3s(setup_remote_kubeconfig=True)
+            install_k3s(setup_remote_kubeconfig=True, node_name=node_name)
     except K3SInstallationError:
         logger.exception("Error during k3s upgrade.")
         click.echo(UpgradeCmdTexts.k3s_upgrade_failed)

platform/services/installer/app/configuration_models/install_config.py

@@ -44,10 +44,11 @@ def custom_certificate(self) -> bool:
         """
         return bool(self.cert_file.value) or bool(self.key_file.value)
 
-    def __init__(self, interactive_mode: bool, install_telemetry_stack: bool) -> None:
+    def __init__(self, interactive_mode: bool, install_telemetry_stack: bool, node_name: str | None = None) -> None:
         self._interactive_mode = interactive_mode
 
         self.install_telemetry_stack = ConfigurationField(type=bool, required=False, value=install_telemetry_stack)
+        self.node_name = ConfigurationField(type=str, required=False, value=node_name)
         self.offer_k8s_option = ConfigurationField(
             type=bool, required=True, value=os.getenv("PLATFORM_K8S_OPTION") == "true"
         )

platform/services/installer/app/k3s/config.py

@@ -25,10 +25,12 @@ def __init__(  # noqa: PLR0913
         kubelet_args: dict[str, str] | None = None,
         service_node_port_range: str | None = None,
         node_ip: str | None = None,
+        node_name: str | None = None,
     ):
         self.version = version
         self.service_node_port_range = service_node_port_range
         self.node_ip = node_ip
+        self.node_name = node_name
         if disable_components is None:
             self.disable_components = []
         else:
@@ -68,6 +70,11 @@ def to_env_var_dict(self) -> dict:
         if self.node_ip:
             install_k3s_exec_big_str += f"--node-ip={self.node_ip} "
 
+        if self.node_name:
+            install_k3s_exec_big_str += f"--node-name={self.node_name} "
+        else:
+            install_k3s_exec_big_str += "--node-name=geti "
+
         flags_mapping = {
             "--kube-apiserver-arg": self.kube_apiserver_args,
             "--kube-controller-manager-arg": self.kube_controller_manager_args,

platform/services/installer/app/k3s/install.py

@@ -117,12 +117,16 @@ def _k3s_ready(time_wait: int = 300, delay: int = 10) -> bool:
     return False
 
 
-def _run_installer(k3s_script_path: str, logs_file_path: str):
+def _run_installer(k3s_script_path: str, logs_file_path: str, node_name: str | None = None):
     """
     Run K3S installer by running downloaded script in 'k3s_script_path' and writing
-    execution logs to 'logs_dir'.
+    execution logs to 'logs_dir'. node_name contains optional name of a node of k3s cluster
     """
+    if node_name:
+        k3s_configuration.node_name = node_name
     k3s_env_vars = k3s_configuration.to_env_var_dict()
+
+    logger.info(f"envs for k3s installation: {k3s_env_vars}")
     env = os.environ.copy()
     env.update(k3s_env_vars)
 
@@ -221,6 +225,7 @@ def _install_k3s_selinux_rpm() -> None:
 def install_k3s(  # noqa: ANN201
     logs_file_path: str = K3S_INSTALL_LOG_FILE_PATH,
     setup_remote_kubeconfig: bool = True,
+    node_name: str | None = None,
 ):
     """
     Install K3S to current system. Write installation logs to 'logs_dir'. Use optionally 'external_address' to adjust
@@ -230,7 +235,7 @@ def install_k3s(  # noqa: ANN201
         k3s_script_path = f"{K3S_OFFLINE_INSTALLATION_FILES_PATH}/install.sh"
         _prepare_k3s_files_structure()
         _install_k3s_selinux_rpm()
-        _run_installer(k3s_script_path=k3s_script_path, logs_file_path=logs_file_path)
+        _run_installer(k3s_script_path=k3s_script_path, logs_file_path=logs_file_path, node_name=node_name)
         _update_containerd_config(logs_file_path=logs_file_path)
         _mark_k3s_installation()
     except subprocess.CalledProcessError as ex:

platform/services/installer/app/platform_utils/k8s.py

@@ -280,3 +280,22 @@ def _deployment_exists(name: str, namespace: str) -> bool:
         if ex.status == 404:
             return False
         raise
+
+
+def get_node_name() -> str | None:
+    """
+    Get the name of the first node in the cluster.
+    """
+    KubernetesConfigHandler(kube_config=K3S_KUBECONFIG_PATH)
+    v1 = client.CoreV1Api()
+    try:
+        nodes = v1.list_node()
+
+        if not nodes.items:
+            logger.error(f"Missing node {nodes}")
+            return None
+    except ApiException as ex:
+        logger.error(f"Exception while getting the node name {ex}")
+        return None
+
+    return nodes.items[0].metadata.name

platform/services/installer/tests/unit/k3s/test_install.py

@@ -166,6 +166,7 @@ def test_k3s_configuration():
         "INSTALL_K3S_SKIP_DOWNLOAD": "true",
         "INSTALL_K3S_EXEC": "--disable traefik "
         "--disable kube-server "
+        "--node-name=geti "
         "--kube-apiserver-arg=enable-admission-plugins=NodeRestriction,"
         "PodSecurityPolicy,ServiceAccount "
         "--kube-controller-manager-arg=leader-elect-lease-duration=30s "