|
| 1 | +# Copyright (c) Microsoft Corporation. |
| 2 | +# Licensed under the MIT License. |
| 3 | + |
| 4 | +""" |
| 5 | +Base class for cluster status checking implementations. |
| 6 | +""" |
| 7 | + |
| 8 | +import logging |
| 9 | +import xml.etree.ElementTree as ET |
| 10 | +from datetime import datetime |
| 11 | +from typing import Dict, Any |
| 12 | + |
| 13 | +try: |
| 14 | + from ansible.module_utils.sap_automation_qa import SapAutomationQA, TestStatus |
| 15 | + from ansible.module_utils.commands import ( |
| 16 | + STONITH_ACTION, |
| 17 | + PACEMAKER_STATUS, |
| 18 | + CLUSTER_STATUS, |
| 19 | + ) |
| 20 | +except ImportError: |
| 21 | + from src.module_utils.sap_automation_qa import SapAutomationQA, TestStatus |
| 22 | + from src.module_utils.commands import ( |
| 23 | + STONITH_ACTION, |
| 24 | + PACEMAKER_STATUS, |
| 25 | + CLUSTER_STATUS, |
| 26 | + ) |
| 27 | + |
| 28 | + |
| 29 | +class BaseClusterStatusChecker(SapAutomationQA): |
| 30 | + """ |
| 31 | + Base class to check the status of a pacemaker cluster. |
| 32 | + """ |
| 33 | + |
| 34 | + def __init__(self, ansible_os_family: str = ""): |
| 35 | + super().__init__() |
| 36 | + self.ansible_os_family = ansible_os_family |
| 37 | + self.result.update( |
| 38 | + { |
| 39 | + "cluster_status": "", |
| 40 | + "start": datetime.now(), |
| 41 | + "end": None, |
| 42 | + "pacemaker_status": "", |
| 43 | + "stonith_action": "", |
| 44 | + } |
| 45 | + ) |
| 46 | + |
| 47 | + def _get_stonith_action(self) -> None: |
| 48 | + """ |
| 49 | + Retrieves the stonith action from the system. |
| 50 | + """ |
| 51 | + try: |
| 52 | + stonith_action = self.execute_command_subprocess(STONITH_ACTION[self.ansible_os_family]) |
| 53 | + stonith_action = ( |
| 54 | + stonith_action.split("stonith-action:")[-1] |
| 55 | + if self.ansible_os_family == "REDHAT" |
| 56 | + else stonith_action |
| 57 | + ) |
| 58 | + self.result["stonith_action"] = stonith_action.strip() |
| 59 | + except Exception: |
| 60 | + self.result["stonith_action"] = "reboot" |
| 61 | + |
| 62 | + def _validate_cluster_basic_status(self, cluster_status_xml: ET.Element) -> bool: |
| 63 | + """ |
| 64 | + Validate the basic status of the cluster. |
| 65 | +
|
| 66 | + :param cluster_status_xml: XML element containing cluster status. |
| 67 | + :type cluster_status_xml: ET.Element |
| 68 | + :return: True if the cluster is stable, False otherwise. |
| 69 | + :rtype: bool |
| 70 | + """ |
| 71 | + # Check pacemaker status |
| 72 | + if self.execute_command_subprocess(PACEMAKER_STATUS).strip() == "active": |
| 73 | + self.result["pacemaker_status"] = "running" |
| 74 | + else: |
| 75 | + self.result["pacemaker_status"] = "stopped" |
| 76 | + self.log(logging.INFO, f"Pacemaker status: {self.result['pacemaker_status']}") |
| 77 | + |
| 78 | + # Check node count |
| 79 | + if int(cluster_status_xml.find("summary").find("nodes_configured").attrib["number"]) < 2: |
| 80 | + self.result["message"] = "Pacemaker cluster isn't stable (insufficient nodes)" |
| 81 | + self.log(logging.WARNING, self.result["message"]) |
| 82 | + return False |
| 83 | + |
| 84 | + # Check if all nodes are online |
| 85 | + nodes = cluster_status_xml.find("nodes") |
| 86 | + for node in nodes: |
| 87 | + if node.attrib["online"] != "true": |
| 88 | + self.result["message"] = f"Node {node.attrib['name']} is not online" |
| 89 | + self.log(logging.WARNING, self.result["message"]) |
| 90 | + return False |
| 91 | + |
| 92 | + return True |
| 93 | + |
| 94 | + def _process_node_attributes(self, node_attributes: ET.Element) -> Dict[str, Any]: |
| 95 | + """ |
| 96 | + Abstract method to process node attributes. |
| 97 | +
|
| 98 | + :param node_attributes: XML element containing node attributes. |
| 99 | + :type node_attributes: ET.Element |
| 100 | + :raises NotImplementedError: If the method is not implemented in a child class. |
| 101 | + :return: Dictionary with node attributes. |
| 102 | + :rtype: Dict[str, Any] |
| 103 | + """ |
| 104 | + raise NotImplementedError("Child classes must implement this method") |
| 105 | + |
| 106 | + def run(self) -> Dict[str, str]: |
| 107 | + """ |
| 108 | + Run the cluster status check. |
| 109 | +
|
| 110 | + :return: Result of the cluster status check. |
| 111 | + :rtype: Dict[str, str] |
| 112 | + """ |
| 113 | + self.log(logging.INFO, "Starting cluster status check") |
| 114 | + self._get_stonith_action() |
| 115 | + |
| 116 | + try: |
| 117 | + # Implementing a common loop structure |
| 118 | + while not self._is_cluster_ready(): |
| 119 | + self.result["cluster_status"] = self.execute_command_subprocess(CLUSTER_STATUS) |
| 120 | + cluster_status_xml = ET.fromstring(self.result["cluster_status"]) |
| 121 | + self.log(logging.INFO, "Cluster status retrieved") |
| 122 | + |
| 123 | + if self._validate_cluster_basic_status(cluster_status_xml): |
| 124 | + self._process_node_attributes(cluster_status_xml.find("node_attributes")) |
| 125 | + |
| 126 | + if not self._is_cluster_stable(): |
| 127 | + self.result["message"] = "Pacemaker cluster isn't stable" |
| 128 | + self.log(logging.WARNING, self.result["message"]) |
| 129 | + |
| 130 | + except Exception as e: |
| 131 | + self.handle_error(e) |
| 132 | + |
| 133 | + self.result["end"] = datetime.now() |
| 134 | + self.result["status"] = TestStatus.SUCCESS.value |
| 135 | + self.log(logging.INFO, "Cluster status check completed") |
| 136 | + return self.result |
| 137 | + |
| 138 | + def _is_cluster_ready(self) -> bool: |
| 139 | + """ |
| 140 | + Abstract method to check if the cluster is ready. |
| 141 | + To be implemented by child classes. |
| 142 | +
|
| 143 | + :raises NotImplementedError: If the method is not implemented in a child class. |
| 144 | + :return: True if the cluster is ready, False otherwise. |
| 145 | + :rtype: bool |
| 146 | + """ |
| 147 | + raise NotImplementedError("Child classes must implement this method") |
| 148 | + |
| 149 | + def _is_cluster_stable(self) -> bool: |
| 150 | + """ |
| 151 | + Abstract method to check if the cluster is in a stable state. |
| 152 | + To be implemented by child classes. |
| 153 | +
|
| 154 | + :raises NotImplementedError: If the method is not implemented in a child class. |
| 155 | + :return: True if the cluster is ready, False otherwise. |
| 156 | + :rtype: bool |
| 157 | + """ |
| 158 | + raise NotImplementedError("Child classes must implement this method") |
0 commit comments