observer.backup 和 backup_clean 收集场景支持分租户

handler/gather/step/base.py

@@ -63,7 +63,7 @@ def execute(self):
                     handler = SshHandler(self.context, self.step, self.node, self.report_path, self.task_variable_dict)
                     handler.execute()
                 elif self.step["type"] == "sql" and (skip_type != "sql"):
-                    handler = StepSQLHandler(self.context, self.step, self.cluster, self.report_path, self.task_variable_dict)
+                    handler = StepSQLHandler(self.context, self.step, self.cluster, self.report_path, self.task_variable_dict, self.env)
                     handler.execute()
                 elif self.step["type"] == "log" and (skip_type != "ssh"):
                     if self.node.get("host_type") and self.node.get("host_type") == "OBSERVER":

handler/gather/step/sql.py

@@ -16,14 +16,15 @@
 @desc:
 """
 import os
+import re
 from stdio import SafeStdio
 from common.ob_connector import OBConnector
 from tabulate import tabulate
 from common.tool import StringUtils
 
 
 class StepSQLHandler(SafeStdio):
-    def __init__(self, context, step, ob_cluster, report_path, task_variable_dict):
+    def __init__(self, context, step, ob_cluster, report_path, task_variable_dict, env):
         self.context = context
         self.stdio = context.stdio
         try:
@@ -32,6 +33,7 @@ def __init__(self, context, step, ob_cluster, report_path, task_variable_dict):
             self.tenant_mode = None
             self.sys_database = None
             self.database = None
+            self.env = env
             self.ob_connector = OBConnector(ip=ob_cluster.get("db_host"), port=ob_cluster.get("db_port"), username=ob_cluster.get("tenant_sys").get("user"), password=ob_cluster.get("tenant_sys").get("password"), stdio=self.stdio, timeout=10000)
         except Exception as e:
             self.stdio.error("StepSQLHandler init fail. Please check the OBCLUSTER conf. OBCLUSTER: {0} Exception : {1} .".format(ob_cluster, e))
@@ -49,6 +51,13 @@ def execute(self):
                 self.stdio.error("StepSQLHandler execute sql is not set")
                 return
             sql = StringUtils.build_sql_on_expr_by_dict(self.step["sql"], self.task_variable_dict)
+            params = StringUtils.extract_parameters(sql)
+            for param in params:
+                values = self.env.get(param)
+                if values is None or len(values) == 0:
+                    self.stdio.error("the values of param %s is None", param)
+                    return
+            sql = StringUtils.replace_parameters(sql, self.env)
             self.stdio.verbose("StepSQLHandler execute: {0}".format(sql))
             columns, data = self.ob_connector.execute_sql_return_columns_and_data(sql)
             if data is None or len(data) == 0:
@@ -57,6 +66,23 @@ def execute(self):
         except Exception as e:
             self.stdio.error("StepSQLHandler execute Exception: {0}".format(e).strip())
 
+    @staticmethod
+    def extract_parameters(query_template):
+        pattern = re.compile(r'\$\{(\w+)\}')
+        parameters = pattern.findall(query_template)
+        return parameters
+
+    @staticmethod
+    def replace_parameters(query_template, params):
+        pattern = re.compile(r'\$\{(\w+)\}')
+
+        def replacer(match):
+            key = match.group(1)
+            return str(params.get(key, match.group(0)))
+
+        query = pattern.sub(replacer, query_template)
+        return query
+
     def update_step_variable_dict(self):
         return self.task_variable_dict
 

handler/gather/tasks/observer/backup.yaml

@@ -1,6 +1,6 @@
 info_en: "[backup problem]"
 info_cn: "[数据备份问题]"
-command: obdiag gather scene run --scene=observer.backup
+command: obdiag gather scene run --scene=observer.backup --env "{tenant_id=xxx}"
 task:
   - version: "[2.0.0.0, 4.0.0.0]"
     steps:
@@ -53,7 +53,7 @@ task:
         sql: "select * from oceanbase.__all_rootservice_event_history where gmt_create > ${from_time} and gmt_create < ${to_time} order by gmt_create desc;"
         global: true
       - type: sql
-        sql: "select b.* from oceanbase.__all_virtual_pg_backup_log_archive_status a,oceanbase.__all_virtual_pg_log_archive_stat b where a.table_id=b.table_id and a.partition_id=b.partition_id  order by log_archive_cur_ts limit 5;"
+        sql: "select b.* from oceanbase.__all_virtual_pg_backup_log_archive_status a,oceanbase.__all_virtual_pg_log_archive_stat b where a.table_id=b.table_id and a.partition_id=b.partition_id  order by log_archive_cur_ts limit 100;"
         global: true
       - type: log
         global: false
@@ -103,22 +103,22 @@ task:
         sql: "show parameters like '%ha_low_thread_score%';"
         global: true
       - type: sql
-        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_PARAMETER"
+        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_PARAMETER where tenant_id = ${tenant_id}"
         global: true
       - type: sql
-        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_JOBS limit 20;"
+        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_JOBS where tenant_id = ${tenant_id};"
         global: true
       - type: sql
         sql: "SELECT * FROM oceanbase.DBA_OB_ROOTSERVICE_EVENT_HISTORY WHERE module='backup_data' AND event ='start_backup_data';"
         global: true
       - type: sql
-        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_TASKS limit 20;"
+        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_TASKS where tenant_id = ${tenant_id};"
         global: true
       - type: sql
-        sql: "SELECT * FROM oceanbase.__all_virtual_backup_schedule_task limit 20"
+        sql: "SELECT * FROM oceanbase.__all_virtual_backup_schedule_task where tenant_id = ${tenant_id}"
         global: true
       - type: sql
-        sql: "SELECT * from oceanbase.CDB_OB_BACKUP_JOB_HISTORY where STATUS = 'FAILED' limit 20;"
+        sql: "SELECT * from oceanbase.CDB_OB_BACKUP_JOB_HISTORY where STATUS = 'FAILED' and tenant_id = ${tenant_id};"
         global: true
       - type: log
         global: false

handler/gather/tasks/observer/backup_clean.yaml

@@ -1,6 +1,6 @@
 info_en: "[backup clean]"
 info_cn: "[备份清理问题]"
-command: obdiag gather scene run --scene=observer.backup_clean
+command: obdiag gather scene run --scene=observer.backup_clean --env "{tenant_id=xxx}"
 task:
   - version: "[2.0.0.0, 4.0.0.0]"
     steps:
@@ -50,7 +50,7 @@ task:
         sql: "select * from oceanbase.__all_virtual_sys_task_status where comment like '%backup%';"
         global: true
       - type: sql
-        sql: "select * from oceanbase.CDB_OB_BACKUP_SET_DETAILS order by START_TIME asc limit 1;"
+        sql: "select * from oceanbase.CDB_OB_BACKUP_SET_DETAILS order by START_TIME desc"
         global: true
       - type: sql
         sql: "select * from oceanbase.__all_backup_task_clean_history where gmt_create > ${from_time} and gmt_create < ${to_time} order by gmt_create desc;"
@@ -106,34 +106,34 @@ task:
         sql: "show parameters like '%backup%';"
         global: true
       - type: sql
-        sql: "select * from oceanbase.CDB_OB_BACKUP_JOB_HISTORY"
+        sql: "select * from oceanbase.CDB_OB_BACKUP_JOB_HISTORY where tenant_id=${tenant_id}"
         global: true
       - type: sql
         sql: "SELECT * FROM oceanbase.DBA_OB_ROOTSERVICE_EVENT_HISTORY WHERE module='backup_data' AND event ='start_backup_data';"
         global: true
       - type: sql
-        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_TASKS limit 20;"
+        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_TASKS where tenant_id=${tenant_id};"
         global: true
       - type: sql
-        sql: "SELECT * FROM oceanbase.__all_virtual_backup_schedule_task limit 20"
+        sql: "SELECT * FROM oceanbase.__all_virtual_backup_schedule_task where tenant_id=${tenant_id}"
         global: true
       - type: sql
-        sql: "SELECT * from oceanbase.CDB_OB_BACKUP_JOB_HISTORY where STATUS = 'FAILED' limit 20;"
+        sql: "SELECT * from oceanbase.CDB_OB_BACKUP_JOB_HISTORY where STATUS = 'FAILED' and tenant_id=${tenant_id};"
         global: true
       - type: sql
-        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_DELETE_POLICY;"
+        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_DELETE_POLICY where tenant_id=${tenant_id};"
         global: true
       - type: sql
-        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_DELETE_JOBS limit 20"
+        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_DELETE_JOBS where tenant_id=${tenant_id}"
         global: true
       - type: sql
-        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_DELETE_TASKS limit 20"
+        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_DELETE_TASKS where tenant_id=${tenant_id}"
         global: true
       - type: sql
-        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_DELETE_TASK_HISTORY limit 20"
+        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_DELETE_TASK_HISTORY where tenant_id=${tenant_id}"
         global: true
       - type: sql
-        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_DELETE_JOB_HISTORY limit 20"
+        sql: "SELECT * FROM oceanbase.CDB_OB_BACKUP_DELETE_JOB_HISTORY where tenant_id=${tenant_id}"
         global: true
       - type: log
         global: false

handler/gather/tasks/observer/io.yaml

@@ -82,7 +82,7 @@ task:
         sql: "show parameters like '%syslog_io_bandwidth_limit%';"
         global: true
       - type: sql
-        sql: "select * from oceanbase.__all_virtual_io_quota limit 20"
+        sql: "select * from oceanbase.__all_virtual_io_quota"
         global: true
       - type: ssh
         ssh: "df -h"

handler/gather/tasks/observer/rootservice_switch.yaml

@@ -121,13 +121,13 @@ task:
         sql: "select * from oceanbase.DBA_OB_SERVER_EVENT_HISTORY where event like '%migrat%' and name6 like '%fail%' and value6=1;"
         global: true
       - type: sql
-        sql: "select * from oceanbase.DBA_OB_SERVER_EVENT_HISTORY where module='FAILURE_DETECTOR' limit 10;"
+        sql: "select * from oceanbase.DBA_OB_SERVER_EVENT_HISTORY where module='FAILURE_DETECTOR';"
         global: true
       - type: sql
-        sql: "select * from oceanbase.DBA_OB_SERVER_EVENT_HISTORY where module like '%ELECTION%' limit 10;"
+        sql: "select * from oceanbase.DBA_OB_SERVER_EVENT_HISTORY where module like '%ELECTION%';"
         global: true
       - type: sql
-        sql: "select * from oceanbase.GV$OB_LOG_STAT where role='LEADER' limit 20;"
+        sql: "select * from oceanbase.GV$OB_LOG_STAT where role='LEADER';"
         global: true
       - type: sql
         sql: "SELECT TENANT_NAME, TENANT_ID, TENANT_ROLE, STATUS, SWITCHOVER_STATUS FROM oceanbase.DBA_OB_TENANTS"