达梦8数据库锁阻塞巡检SQL实现与优化

1. 项目背景与核心价值

在数据库运维领域，锁阻塞问题一直是影响系统性能的关键瓶颈。达梦8作为国产数据库的代表产品，其锁机制与Oracle等传统数据库存在显著差异。这套"最终锁阻塞巡检SQL"脚本的诞生，源于我在某大型金融系统迁移项目中遇到的真实痛点——当业务高峰期出现性能骤降时，传统监控工具往往只能告诉我们"有锁等待"，却无法快速定位到阻塞源头和受影响的关键业务。

这套脚本的价值在于：它不仅能实时捕获达梦8数据库中的锁阻塞链条，还能精确定位到阻塞会话的SQL语句、等待时长、涉及对象等关键信息。通过预设的阈值判断，运维人员可以快速区分"正常短时等待"和"异常长时阻塞"，把平均故障定位时间从小时级缩短到分钟级。特别是在达梦8的共享服务器架构下，一个长时间持有锁的会话可能拖垮整个实例，这种巡检工具的重要性更是不言而喻。

2. 锁阻塞原理与达梦8特性

2.1 达梦8锁机制解析

达梦8采用多粒度锁机制，包含表级锁（TM锁）和行级锁（TX锁）。与Oracle不同的是，其锁转换规则更为严格：

• 意向锁（IS/IX）到表锁（S/X）的升级需要额外检查
• 外键约束会自动加锁，容易形成意外阻塞
• DDL操作会获取高阶锁，可能阻塞并发DML

特别需要注意的是达梦8的"锁继承"特性：当会话A阻塞会话B，而会话B又阻塞会话C时，会形成级联阻塞链。传统查询只能看到A→B的关系，而我们的脚本通过递归CTE可以追踪整个链条。

2.2 阻塞检测关键技术点

核心数据源来自达梦8的V$LOCK和V$SESSION视图，但需要特别注意：

sql复制SELECT * FROM V$LOCK WHERE BLOCK = 1;  -- 仅显示主动阻塞的锁
SELECT * FROM V$SESSION WHERE BLOCKING_SESSION IS NOT NULL; -- 被阻塞会话

达梦8的锁等待超时参数（DM.INI中的LOCK_TIMEOUT）默认为-1（无限等待），这使长时阻塞更容易发生。我们的脚本会特别标注超过300秒的等待，这通常是需要立即干预的临界值。

3. 巡检SQL完整实现方案

3.1 核心查询结构

sql复制WITH BLOCK_CHAIN AS (
    -- 基础阻塞对查询
    SELECT 
        HOLDER.SID AS HOLDER_SID,
        WAITER.SID AS WAITER_SID,
        HOLDER.USERNAME AS HOLDER_USER,
        WAITER.USERNAME AS WAITER_USER,
        HOLDER.STATUS AS HOLDER_STATUS,
        WAITER.EVENT AS WAITER_EVENT,
        HOLDER.MACHINE AS HOLDER_MACHINE,
        WAITER.SECONDS_IN_WAIT
    FROM 
        V$SESSION HOLDER,
        V$SESSION WAITER
    WHERE 
        WAITER.BLOCKING_SESSION = HOLDER.SID
    UNION ALL
    -- 递归查询阻塞链
    SELECT 
        CHAIN.HOLDER_SID,
        WAITER.SID,
        CHAIN.HOLDER_USER,
        WAITER.USERNAME,
        CHAIN.HOLDER_STATUS,
        WAITER.EVENT,
        CHAIN.HOLDER_MACHINE,
        WAITER.SECONDS_IN_WAIT
    FROM 
        BLOCK_CHAIN CHAIN,
        V$SESSION WAITER
    WHERE 
        WAITER.BLOCKING_SESSION = CHAIN.WAITER_SID
)
SELECT * FROM BLOCK_CHAIN
ORDER BY HOLDER_SID, SECONDS_IN_WAIT DESC;

3.2 关键增强功能

1. SQL文本捕获：通过关联V$SQLAREA获取完整SQL

sql复制SELECT 
    CHAIN.*,
    HOLDER_SQL.SQL_TEXT AS HOLDER_SQL,
    WAITER_SQL.SQL_TEXT AS WAITER_SQL
FROM 
    BLOCK_CHAIN CHAIN
    LEFT JOIN V$SQLAREA HOLDER_SQL ON HOLDER_SQL.ADDRESS = (
        SELECT SQL_ADDRESS FROM V$SESSION WHERE SID = CHAIN.HOLDER_SID
    )
    LEFT JOIN V$SQLAREA WAITER_SQL ON WAITER_SQL.ADDRESS = (
        SELECT SQL_ADDRESS FROM V$SESSION WHERE SID = CHAIN.WAITER_SID
    )

2. 对象定位：通过V$LOCKED_OBJECT关联到具体表

sql复制SELECT 
    OBJECT_NAME, 
    SESSION_ID 
FROM 
    V$LOCKED_OBJECT LO, 
    DBA_OBJECTS OBJ 
WHERE 
    LO.OBJECT_ID = OBJ.OBJECT_ID

3. 等待时间分级：

sql复制CASE 
    WHEN SECONDS_IN_WAIT < 60 THEN '轻微(<1min)'
    WHEN SECONDS_IN_WAIT < 300 THEN '警告(1-5min)'
    ELSE '严重(>5min)' 
END AS WAIT_LEVEL

4. 生产环境部署方案

4.1 定时巡检配置

建议通过达梦8的作业系统设置每5分钟执行一次：

sql复制BEGIN
    DBMS_JOB.SUBMIT(
        JOB => :job_no,
        WHAT => 'BEGIN 
            INSERT INTO BLOCK_HISTORY 
            SELECT *, SYSDATE FROM DM_BLOCK_CHECK_V; 
        END;',
        NEXT_DATE => SYSDATE,
        INTERVAL => 'SYSDATE + 5/1440'
    );
    COMMIT;
END;

历史记录表结构示例：

sql复制CREATE TABLE BLOCK_HISTORY (
    DETECT_TIME TIMESTAMP,
    HOLDER_SID NUMBER,
    WAITER_SID NUMBER,
    HOLDER_USER VARCHAR(30),
    WAITER_USER VARCHAR(30),
    HOLDER_SQL CLOB,
    WAITER_SQL CLOB,
    OBJECT_NAME VARCHAR(128),
    WAIT_SECONDS NUMBER,
    WAIT_LEVEL VARCHAR(20)
);

4.2 报警阈值设置

推荐报警规则组合：

1. 同一holder阻塞超过3个会话
2. 任何等待超过300秒的阻塞
3. 核心业务表出现TX锁等待

对应的报警SQL：

sql复制SELECT * FROM (
    SELECT 
        HOLDER_SID,
        COUNT(*) AS BLOCK_COUNT
    FROM 
        BLOCK_CHAIN
    GROUP BY 
        HOLDER_SID
) WHERE BLOCK_COUNT >= 3;

SELECT * FROM BLOCK_CHAIN WHERE SECONDS_IN_WAIT > 300;

SELECT * FROM BLOCK_CHAIN 
WHERE OBJECT_NAME IN ('ACCOUNT_INFO', 'TRADE_RECORD', 'PAYMENT_ORDER');

5. 典型问题排查手册

5.1 常见阻塞场景

1. DDL阻塞DML：

   • 现象：ALTER TABLE阻塞INSERT/UPDATE
   • 解决：检查是否有长时间运行的DDL，优先终止DDL会话

2. 外键锁升级：

   • 现象：无索引外键导致表锁
   • 解决：为外键添加索引，特别是ON DELETE CASCADE的情况

3. 事务未提交：

   • 现象：holder会话状态为"INACTIVE"但持有锁
   • 解决：联系应用开发者确认事务边界

5.2 性能优化技巧

1. 在查询V$SQLAREA时添加HINT避免硬解析：

sql复制SELECT /*+ FIRST_ROWS(10) */ SQL_TEXT FROM V$SQLAREA WHERE ROWNUM < 10

2. 对历史表创建时间分区提升查询效率：

sql复制CREATE TABLE BLOCK_HISTORY (
    DETECT_TIME TIMESTAMP,
    ...
) PARTITION BY RANGE (DETECT_TIME) (
    PARTITION P202301 VALUES LESS THAN (TO_DATE('2023-02-01','YYYY-MM-DD')),
    PARTITION P202302 VALUES LESS THAN (TO_DATE('2023-03-01','YYYY-MM-DD'))
);

3. 使用物化视图预聚合常见查询：

sql复制CREATE MATERIALIZED VIEW MV_BLOCK_STATS 
REFRESH COMPLETE ON DEMAND
AS 
SELECT 
    TRUNC(DETECT_TIME) AS DAY,
    HOLDER_USER,
    COUNT(*) AS TOTAL_BLOCKS
FROM 
    BLOCK_HISTORY
GROUP BY 
    TRUNC(DETECT_TIME), HOLDER_USER;

6. 进阶应用场景

6.1 与监控系统集成

通过达梦8的UTL_HTTP包直接将报警推送到监控平台：

sql复制BEGIN
    UTL_HTTP.SET_PROXY('proxy.example.com:8080');
    UTL_HTTP.REQUEST(
        url => 'http://monitor/api/alert',
        method => 'POST',
        body => JSON_OBJECT(
            'db_host' VALUE SYS_CONTEXT('USERENV','SERVER_HOST'),
            'block_sid' VALUE :holder_sid,
            'wait_seconds' VALUE :wait_seconds
        )
    );
END;

6.2 锁趋势分析

使用达梦8的分析函数计算阻塞趋势：

sql复制SELECT 
    TRUNC(DETECT_TIME, 'HH24') AS HOUR,
    COUNT(*) AS TOTAL_BLOCKS,
    ROUND(AVG(SECONDS_IN_WAIT)) AS AVG_WAIT,
    MAX(SECONDS_IN_WAIT) AS MAX_WAIT
FROM 
    BLOCK_HISTORY
WHERE 
    DETECT_TIME > SYSDATE - 7
GROUP BY 
    TRUNC(DETECT_TIME, 'HH24')
ORDER BY 
    HOUR DESC;

6.3 自动解锁机制

对于已知的安全会话，可以配置自动解锁脚本：

sql复制DECLARE
    CURSOR BLOCK_CURSOR IS
        SELECT HOLDER_SID 
        FROM BLOCK_CHAIN 
        WHERE SECONDS_IN_WAIT > 600
          AND HOLDER_USER = 'BATCH_USER';
BEGIN
    FOR R IN BLOCK_CURSOR LOOP
        EXECUTE IMMEDIATE 'ALTER SYSTEM KILL SESSION ''' || R.HOLDER_SID || ''' IMMEDIATE';
        INSERT INTO KILL_LOG VALUES(SYSDATE, R.HOLDER_SID, 'AUTO_KILL');
    END LOOP;
    COMMIT;
END;