Refactor SCS cluster status checker to enhance node attribute processing and improve stability checks

Author: devanshjainms

src/modules/get_cluster_status_scs.py

@@ -2,154 +2,76 @@
 # Licensed under the MIT License.
 
 """
-Python script to get and validate the status of a cluster.
-This script uses the `crm_mon` command-line tool to retrieve the status of a cluster and performs
-various validations on the cluster status.
-
-Methods:
-    check_node(node, sap_sid)
-    run_module()
-    main()
+Python script to get and validate the status of an SCS cluster.
 """
 
 import logging
 import xml.etree.ElementTree as ET
-from datetime import datetime
 from typing import Dict, Any
 from ansible.module_utils.basic import AnsibleModule
 
 try:
-    from ansible.module_utils.sap_automation_qa import SapAutomationQA, TestStatus
-    from ansible.module_utils.commands import (
-        STONITH_ACTION,
-        PACEMAKER_STATUS,
-        CLUSTER_STATUS,
-    )
+    from ansible.module_utils.get_cluster_status import BaseClusterStatusChecker
 except ImportError:
-    from src.module_utils.sap_automation_qa import SapAutomationQA, TestStatus
-    from src.module_utils.commands import (
-        STONITH_ACTION,
-        PACEMAKER_STATUS,
-        CLUSTER_STATUS,
-    )
+    from src.module_utils.get_cluster_status import BaseClusterStatusChecker
 
 
-class ClusterStatusChecker(SapAutomationQA):
+class SCSClusterStatusChecker(BaseClusterStatusChecker):
     """
-    Class to check the status of a pacemaker cluster in a SAP HANA environment.
+    Class to check the status of a pacemaker cluster in an SAP SCS environment.
     """
 
     def __init__(self, sap_sid: str, ansible_os_family: str = ""):
-        super().__init__()
+        super().__init__(ansible_os_family)
         self.sap_sid = sap_sid
-        self.ansible_os_family = ansible_os_family
         self.result.update(
             {
                 "ascs_node": "",
                 "ers_node": "",
-                "cluster_status": "",
-                "start": datetime.now(),
-                "end": None,
-                "pacemaker_status": "",
-                "stonith_action": "",
             }
         )
 
-    def _get_stonith_action(self) -> None:
-        """
-        Retrieves the STONITH action from the crm_attribute.
-        """
-        try:
-            stonith_action = self.execute_command_subprocess(STONITH_ACTION[self.ansible_os_family])
-            stonith_action = (
-                stonith_action.split("stonith-action:")[-1]
-                if self.ansible_os_family == "REDHAT"
-                else stonith_action
-            )
-            self.result["stonith_action"] = stonith_action.strip()
-        except Exception:
-            self.result["stonith_action"] = "reboot"
-
     def _process_node_attributes(self, node_attributes: ET.Element) -> Dict[str, Any]:
         """
-        Processes node attributes and identifies primary/secondary nodes.
+        Processes node attributes and identifies ASCS and ERS nodes.
 
-        :param node_attributes: The XML element containing node attributes.
+        :param node_attributes: XML element containing node attributes.
         :type node_attributes: ET.Element
-        :return: A dictionary containing the primary and secondary node information, plus cluster status.
+        :return: Dictionary with ASCS and ERS node information.
         :rtype: Dict[str, Any]
         """
-        attribute_name = f"runs_ers_{self.sap_sid.upper()}"
+        all_nodes = [node.attrib.get("name") for node in node_attributes]
         for node in node_attributes:
+            node_name = node.attrib["name"]
             for attribute in node:
-                if attribute.attrib["name"] == attribute_name:
+                if attribute.attrib["name"] == f"runs_ers_{self.sap_sid.upper()}":
                     if attribute.attrib["value"] == "1":
-                        self.result["ers_node"] = node.attrib["name"]
+                        self.result["ers_node"] = node_name
                     else:
-                        self.result["ascs_node"] = node.attrib["name"]
-                else:
-                    continue
+                        self.result["ascs_node"] = node_name
+
+        if self.result["ascs_node"] == "" and self.result["ers_node"] != "":
+            self.result["ascs_node"] = next(
+                (n for n in all_nodes if n != self.result["ers_node"]), ""
+            )
 
-    def run(self) -> Dict[str, str]:
+    def _is_cluster_ready(self) -> bool:
         """
-        Main function that runs the Ansible module and performs the cluster status checks.
+        Check if the cluster is ready by verifying the ASCS node.
 
-        This function retrieves operation step from module arguments and performs following checks:
-        - Checks the status of the cluster using the `crm_mon` command.
-        - Validates the cluster status and checks if pacemakerd is running.
-        - Checks if the minimum required number of nodes are configured in the cluster.
-        - Checks if all nodes in the cluster are online.
-        - Checks the attributes of each node in the cluster.
+        :return: True if the cluster is ready, False otherwise.
+        :rtype: bool
+        """
+        return self.result["ascs_node"] != ""
+
+    def _is_cluster_stable(self) -> bool:
+        """
+        Check if the cluster is stable by verifying both ASCS and ERS nodes.
 
-        :return: A dictionary containing the result of the cluster status checks.
-        :rtype: Dict[str, str]
+        :return: True if the cluster is stable, False otherwise.
+        :rtype: bool
         """
-        self.log(logging.INFO, "Starting cluster status check")
-
-        self._get_stonith_action()
-
-        try:
-            while self.result["ascs_node"] == "":
-                self.result["cluster_status"] = self.execute_command_subprocess(CLUSTER_STATUS)
-                cluster_status_xml = ET.fromstring(self.result["cluster_status"])
-                self.log(logging.INFO, "Cluster status retrieved")
-
-                if self.execute_command_subprocess(PACEMAKER_STATUS).strip() == "active":
-                    self.result["pacemaker_status"] = "running"
-                else:
-                    self.result["pacemaker_status"] = "stopped"
-                self.log(logging.INFO, f"Pacemaker status: {self.result['pacemaker_status']}")
-
-                if (
-                    int(
-                        cluster_status_xml.find("summary").find("nodes_configured").attrib["number"]
-                    )
-                    < 2
-                ):
-                    self.result["message"] = (
-                        "Pacemaker cluster isn't stable and does not have primary or secondary node"
-                    )
-                    self.log(logging.WARNING, self.result["message"])
-
-                nodes = cluster_status_xml.find("nodes")
-                for node in nodes:
-                    if node.attrib["online"] != "true":
-                        self.result["message"] = f"Node {node.attrib['name']} is not online"
-                        self.log(logging.WARNING, self.result["message"])
-
-                self._process_node_attributes(cluster_status_xml.find("node_attributes"))
-
-            if self.result["ascs_node"] == "" or self.result["ers_node"] == "":
-                self.result["message"] = (
-                    "Pacemaker cluster isn't stable and does not have primary or secondary node"
-                )
-                self.log(logging.WARNING, self.result["message"])
-
-        except Exception as e:
-            self.handle_error(e)
-        self.result["end"] = datetime.now()
-        self.result["status"] = TestStatus.SUCCESS.value
-        self.log(logging.INFO, "Cluster status check completed")
+        return self.result["ascs_node"] != "" and self.result["ers_node"] != ""
 
 
 def run_module() -> None:
@@ -163,7 +85,7 @@ def run_module() -> None:
 
     module = AnsibleModule(argument_spec=module_args, supports_check_mode=True)
 
-    checker = ClusterStatusChecker(
+    checker = SCSClusterStatusChecker(
         sap_sid=module.params["sap_sid"],
         ansible_os_family=module.params["ansible_os_family"],
     )