add check task

plugins/check/tasks/observer/cluster/ls_nu.yaml

@@ -0,0 +1,45 @@
+info: 'Check ls id is not_enough_replica'
+task:
+  - version: "[4.0.0.0,*]"
+    steps:
+    - type: sql
+      sql: "
+WITH LeaderInfo AS (
+    SELECT 
+        tenant_id, 
+        ls_id, 
+        paxos_replica_num
+    FROM 
+        oceanbase.__all_virtual_log_stat
+    WHERE 
+        role = 'LEADER'
+),
+RowCounts AS (
+    SELECT 
+        tenant_id, 
+        ls_id, 
+        COUNT(*) as row_count
+    FROM 
+        oceanbase.__all_virtual_log_stat
+    GROUP BY 
+        tenant_id, 
+        ls_id
+)
+SELECT 
+	GROUP_CONCAT(DISTINCT L.tenant_id)
+FROM 
+    LeaderInfo L
+JOIN 
+    RowCounts R
+ON 
+    L.tenant_id = R.tenant_id AND L.ls_id = R.ls_id
+WHERE 
+    R.row_count < L.paxos_replica_num;"
+      result:
+        set_value: not_enough_replica
+        verify: '[ -z "${not_enough_replica}" ]'
+        err_msg: 'There is not_enough_replica tenant_id: #{not_enough_replica} not_enough_replica, please check as soon as possible.'
+
+
+
+

plugins/check/tasks/observer/cluster/no_leader.yaml

@@ -0,0 +1,12 @@
+info: 'Check cluster tenant ls leader.'
+task:
+  - version: "[4.0.0.0,*]"
+    steps:
+    - type: sql
+      sql: "SELECT GROUP_CONCAT(DISTINCT TENANT_ID) 
+FROM oceanbase.GV$OB_LOG_STAT
+HAVING COUNT(CASE WHEN ROLE = 'LEADER' THEN 1 END) = 0;"
+      result:
+        set_value: no_leader_tenant_id
+        verify: '[ -z "${no_leader_tenant_id}" ]'
+        err_msg: 'there is no leader tenant, please check the cluster. tenant_id: #{no_leader_tenant_id}. You can use "obdiag rca run --scene=log_error" to get more information.'

plugins/check/tasks/observer/cluster/zone_not_active.yaml

@@ -0,0 +1,14 @@
+info: 'Check whether there is any zone not in the ACTIVE state.'
+task:
+  - version: "[4.0.0.0,*]"
+    steps:
+    - type: sql
+      sql: 'select GROUP_CONCAT(DISTINCT ZONE) from oceanbase.dba_ob_zones where STATUS<>"ACTIVE";'
+      result:
+        set_value: not_ACTIVE
+        verify: '[ -z "${not_ACTIVE}" ]'
+        err_msg: 'There is #{not_ACTIVE} not_ACTIVE zone, please check as soon as possible.'
+
+
+
+

plugins/check/tasks/observer/network/network_offset.yaml

@@ -0,0 +1,38 @@
+info: 'Check cluster info about network clockdiff offset'
+task:
+  - version: "[4.0.0.0,*]"
+    steps:
+    - type: local_ssh
+      ssh: 'clockdiff -o #{remote_ip}'
+      result:
+        set_value: clockdiff
+        verify: '[[ ! $clockdiff == *"is down"* ]]'
+        #report_type: warning
+        err_msg: "node: #{remote_ip} can not get clock offset by 'clockdiff -o #{remote_ip}', doc: https://www.oceanbase.com/knowledge-base/ocp-ee-1000000000346970?back=kb"
+    - type: local_ssh
+      ssh: "echo \"#{clockdiff}\" | awk '{print $2}'"
+      result:
+        set_value: offset
+        verify_type: max
+        verify: 50
+        err_msg: "node: #{remote_ip} clock offset is #{offset}, it is over 50ms, issue: https://github.com/oceanbase/obdiag/issues/701"
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+

src/handler/checker/step/local_ssh.py

@@ -0,0 +1,64 @@
+#!/usr/bin/env python
+# -*- coding: UTF-8 -*
+# Copyright (c) 2022 OceanBase
+# OceanBase Diagnostic Tool is licensed under Mulan PSL v2.
+# You can use this software according to the terms and conditions of the Mulan PSL v2.
+# You may obtain a copy of Mulan PSL v2 at:
+#          http://license.coscl.org.cn/MulanPSL2
+# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
+# EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
+# MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
+# See the Mulan PSL v2 for more details.
+"""
+@time: 2025/01/21
+@file: local_ssh.py
+@desc:
+"""
+from src.common.ssh_client.local_client import LocalClient
+from src.handler.checker.check_exception import StepExecuteFailException
+from src.handler.checker.check_report import TaskReport
+from src.common.tool import StringUtils
+from src.common.tool import Util
+
+
+class StepLocalHandler:
+    def __init__(self, context, step, node, task_variable_dict):
+        self.context = context
+        self.stdio = context.stdio
+        self.ssh_report_value = None
+        self.parameters = None
+        self.step = step
+        self.node = node
+        try:
+            self.ssh_client = LocalClient(self.context, node={"ssh_type": "local"})
+        except Exception as e:
+            self.stdio.error("StepLocalHandler init fail. Please check the NODES conf. node: {0}. Exception : {1} .".format(node, e))
+            raise Exception("StepLocalHandler init fail. Please check the NODES conf node: {0}  Exception : {1} .".format(node, e))
+        self.task_variable_dict = task_variable_dict
+        self.parameter = []
+        self.report = TaskReport
+
+    def execute(self):
+        try:
+            if "ssh" not in self.step:
+                raise StepExecuteFailException("StepLocalHandler execute ssh is not set")
+            ssh_cmd = StringUtils.build_str_on_expr_by_dict(self.step["ssh"], self.task_variable_dict)
+            self.stdio.verbose("step StepLocalHandler execute :{0} ".format(ssh_cmd))
+            ssh_report_value = self.ssh_client.exec_cmd(ssh_cmd)
+            if ssh_report_value is None:
+                ssh_report_value = ""
+            if len(ssh_report_value) > 0:
+                ssh_report_value = ssh_report_value.strip()
+            self.stdio.verbose("ssh result:{0}".format(Util.convert_to_number(ssh_report_value)))
+            if "result" in self.step and "set_value" in self.step["result"]:
+                self.stdio.verbose("ssh result set {0}".format(self.step["result"]["set_value"], Util.convert_to_number(ssh_report_value)))
+                self.task_variable_dict[self.step["result"]["set_value"]] = Util.convert_to_number(ssh_report_value)
+        except Exception as e:
+            self.stdio.error("ssh execute Exception:{0}".format(e).strip())
+            raise StepExecuteFailException("ssh execute Exception:{0}".format(e).strip())
+        finally:
+            self.ssh_client.ssh_close()
+        self.stdio.verbose("step StepLocalHandler ssh_report_value:{0}".format(ssh_report_value))
+
+    def update_step_variable_dict(self):
+        return self.task_variable_dict

src/handler/checker/step/stepbase.py

@@ -20,6 +20,7 @@
 from src.handler.checker.step.data_size import DataSizeHandler
 from src.handler.checker.step.get_system_parameter import GetSystemParameterHandler
 from src.handler.checker.result.result import CheckResult
+from src.handler.checker.step.local_ssh import StepLocalHandler
 from src.handler.checker.step.ssh import SshHandler
 from src.handler.checker.step.sql import StepSQLHandler
 from src.common.ssh_client.ssh import SshClient
@@ -58,6 +59,8 @@ def execute(self, report):
                 handler = StepSQLHandler(self.context, self.step, task_variable_dict=self.task_variable_dict)
             elif self.step["type"] == "data_size":
                 handler = DataSizeHandler(self.context, self.step, self.cluster, self.task_variable_dict)
+            elif self.step["type"] == "local_ssh":
+                handler = StepLocalHandler(self.context, self.step, self.cluster, self.task_variable_dict)
             else:
                 raise StepExecuteFailException("the type not support: {0}".format(self.step["type"]))
             self.stdio.verbose("task execute and result")