前置条件
- 包含obd和obclient的中控机
- OceanBase 测试集群
- 独立的测试租户
- BenchmarkSQL 工具(可选)
为了能够方面的查看复杂SQL的执行计划,我们先用TPCC模拟一些数据库负载。
模拟数据库负载
obd里面已经集成了tpcc测试工具,需要联网更新一下插件即可。如果机器不具备外网环境,需要提前下载BenchmarkSQL上传到测试机中。
BenchmarkSQL下载地址:https://github.com/meiq4096/benchmarksql-5.0
这里直接用obd里的tpcc,操作比较简单:
[ob@localhost ~]$ sudo yum install -y yum-utils
[ob@localhost ~]$ sudo yum-config-manager --add-repo https://mirrors.aliyun.com/oceanbase/OceanBase.repo
[ob@localhost ~]$ sudo yum install obtpcc java
在前面新建的tt
租户下跑一个10仓的负载,时间是5分钟:
[ob@localhost ~]$ obd test tpcc obtest --tenant=tt --warehouses 10 --run-mins 5
Get local repositories and plugins ok
Open ssh connection ok
Cluster status check ok
Connect obproxy(x.x.x.222:2883) ok
Connect obproxy(x.x.x.222:2883) ok
Optimize for stage build ok
Connect to tenant tt ok
Server check ok
Merge ok
Starting BenchmarkSQL LoadData
driver=com.mysql.jdbc.Driver
conn=jdbc:mysql://x.x.x.222:2883/test?rewriteBatchedStatements=true&allowMultiQueries=true&useLocalSessionState=true&useUnicode=true&characterEncoding=utf-8&socketTimeout=30000000&useSSL=false
user=root@tt
password=***********
warehouses=10
loadWorkers=1
fileLocation (not defined)
csvNullValue (not defined - using default 'NULL')
Worker 000: Loading ITEM
Worker 000: Loading ITEM done
Worker 000: Loading Warehouse 1
Worker 000: Loading Warehouse 1 done
Worker 000: Loading Warehouse 2
Worker 000: Loading Warehouse 2 done
Worker 000: Loading Warehouse 3
Worker 000: Loading Warehouse 3 done
Worker 000: Loading Warehouse 4
Worker 000: Loading Warehouse 4 done
Worker 000: Loading Warehouse 5
Worker 000: Loading Warehouse 5 done
Worker 000: Loading Warehouse 6
Worker 000: Loading Warehouse 6 done
Worker 000: Loading Warehouse 7
Worker 000: Loading Warehouse 7 done
Worker 000: Loading Warehouse 8
Worker 000: Loading Warehouse 8 done
Worker 000: Loading Warehouse 9
Worker 000: Loading Warehouse 9 done
Worker 000: Loading Warehouse 10
Worker 000: Loading Warehouse 10 done
create index ok
finish build ok
check data ok
Optimize for stage test ok
Connect to tenant tt ok
Merge ok
14:20:27,866 [main] INFO jTPCC : Term-00,
14:20:27,869 [main] INFO jTPCC : Term-00, +-------------------------------------------------------------+
14:20:27,869 [main] INFO jTPCC : Term-00, BenchmarkSQL v5.0
14:20:27,869 [main] INFO jTPCC : Term-00, +-------------------------------------------------------------+
14:20:27,869 [main] INFO jTPCC : Term-00, (c) 2003, Raul Barbosa
14:20:27,869 [main] INFO jTPCC : Term-00, (c) 2004-2016, Denis Lussier
14:20:27,872 [main] INFO jTPCC : Term-00, (c) 2016, Jan Wieck
14:20:27,872 [main] INFO jTPCC : Term-00, +-------------------------------------------------------------+
14:20:27,872 [main] INFO jTPCC : Term-00,
14:20:27,872 [main] INFO jTPCC : Term-00, db=oceanbase
14:20:27,872 [main] INFO jTPCC : Term-00, driver=com.mysql.jdbc.Driver
14:20:27,872 [main] INFO jTPCC : Term-00, conn=jdbc:mysql://x.x.x.222:2883/test?rewriteBatchedStatements=true&allowMultiQueries=true&useLocalSessionState=true&useUnicode=true&characterEncoding=utf-8&socketTimeout=30000000&useSSL=false
14:20:27,872 [main] INFO jTPCC : Term-00, user=root@tt
14:20:27,872 [main] INFO jTPCC : Term-00,
14:20:27,872 [main] INFO jTPCC : Term-00, warehouses=10
14:20:27,873 [main] INFO jTPCC : Term-00, terminals=100
14:20:27,874 [main] INFO jTPCC : Term-00, runMins=5
14:20:27,874 [main] INFO jTPCC : Term-00, limitTxnsPerMin=0
14:20:27,874 [main] INFO jTPCC : Term-00, terminalWarehouseFixed=true
14:20:27,874 [main] INFO jTPCC : Term-00,
14:20:27,874 [main] INFO jTPCC : Term-00, newOrderWeight=45
14:20:27,875 [main] INFO jTPCC : Term-00, paymentWeight=43
14:20:27,875 [main] INFO jTPCC : Term-00, orderStatusWeight=4
14:20:27,875 [main] INFO jTPCC : Term-00, deliveryWeight=4
14:20:27,875 [main] INFO jTPCC : Term-00, stockLevelWeight=4
14:20:27,875 [main] INFO jTPCC : Term-00,
14:20:27,875 [main] INFO jTPCC : Term-00, resultDirectory=my_result_%tY-%tm-%td_%tH%tM%tS
14:20:27,875 [main] INFO jTPCC : Term-00, osCollectorScript=./misc/os_collector_linux.py
14:20:27,875 [main] INFO jTPCC : Term-00,
14:20:27,891 [main] INFO jTPCC : Term-00, copied /home/ob/tmp/props.oceanbase to my_result_2024-03-17_142027/run.properties
14:20:27,891 [main] INFO jTPCC : Term-00, created my_result_2024-03-17_142027/data/runInfo.csv for runID 1
14:20:27,891 [main] INFO jTPCC : Term-00, writing per transaction results to my_result_2024-03-17_142027/data/result.csv
14:20:27,892 [main] INFO jTPCC : Term-00, osCollectorScript=./misc/os_collector_linux.py
14:20:27,892 [main] INFO jTPCC : Term-00, osCollectorInterval=1
14:20:27,892 [main] INFO jTPCC : Term-00, osCollectorSSHAddr=null
14:20:27,892 [main] INFO jTPCC : Term-00, osCollectorDevices=null
14:20:27,953 [main] INFO jTPCC : Term-00,
14:20:28,239 [main] INFO jTPCC : Term-00, C value for C_LAST during load: 192 Term-00, 14:25:29,121 [Thread-5] INFO jTPCC : Term-00, mTOTAL: 1116984 Memory Usage: 95MB / 730MB 14:25:29,121 [Thread-5] INFO jTPCC : Term-00, 14:25:29,121 [Thread-5] INFO jTPCC : Term-00, Measured tpmC (NewOrders) = 15185.13
14:25:29,121 [Thread-5] INFO jTPCC : Term-00, Measured tpmTOTAL = 33746.42
14:25:29,121 [Thread-5] INFO jTPCC : Term-00, Session Start = 2024-03-17 14:20:28
14:25:29,121 [Thread-5] INFO jTPCC : Term-00, Session End = 2024-03-17 14:25:29
14:25:29,122 [Thread-5] INFO jTPCC : Term-00, Transaction Count = 168883
TPC-C Result
Measured tpmC (NewOrders) : 15185.13
Measured tpmTOTAL : 33746.42
Session Start : 2024-03-17 14:20:28
Session End : 2024-03-17 14:25:29
Transaction Count : 168883
Recover ok
Trace ID: eeb13702-e424-11ee-aaa8-1c697a639d50
If you want to view detailed obd logs, please run: obd display-trace eeb13702-e424-11ee-aaa8-1c697a639d50
手动部署 BenchmarkSQL 主要步骤:
git clone https://github.com/obpilot/benchmarksql-5.0.git
cd benchmarksql-5.0/run
vi props.ob # 修改成实际的连接信息,设置主要运行参数
sh runSQL.sh props.ob sql.oceanbase/tableCreates.sql # 创建表结构
sh runSQL.sh props.ob sql.oceanbase/indexCreates.sql # 创建索引
sh runLoader.sh props.ob # 装载数据
sh runBenchmark.sh props.ob # 执行测试
跑tpcc时数据库的负载情况如下:
查找 TOP SQL
查询某段时间内请求次数排在 TOP N 的 SQL:
SELECT/*+ PARALLEL(15)*/ SQL_ID,PLAN_ID, COUNT(*) AS QPS, AVG(t1.elapsed_time) RT
FROM oceanbase.GV$OB_SQL_AUDIT t1 WHERE tenant_id = 1002 AND
IS_EXECUTOR_RPC = 0 AND time_to_usec(DATE_SUB(current_timestamp, INTERVAL 10 MINUTE) ) AND
request_time < time_to_usec (now())
GROUP BY t1.sql_id ORDER BY QPS DESC LIMIT 10;
+----------------------------------+---------+------+----------+
| SQL_ID | PLAN_ID | QPS | RT |
+----------------------------------+---------+------+----------+
| A460265EC2F0763A15DD27CE9E4E2200 | 4518 | 4771 | 59.6087 |
| 7229213613983BC5FDA15AD11EC70D01 | 3206 | 1440 | 340.2958 |
| E1F2BDA1D7391B757859ED3704E5AFB7 | 3213 | 1440 | 125.9278 |
| 2B5697196EDFE9E97FA3384F581178F5 | 3214 | 1440 | 87.7139 |
| FFFCA4D67EA0A788813031B8BBC3B329 | 0 | 301 | 34.6412 |
| 54B5A5861DAF78F52D9ADFBEE83D35B5 | 3202 | 152 | 94.4276 |
| 7F4A525CC8F849F6527F4911CC4BC348 | 3189 | 152 | 121.2303 |
| BB01ECB7605AE31CBC65E6949D84B235 | 3195 | 152 | 220.9079 |
| E86A0CA8BE3F21A2FBC9F1F9855075A1 | 3184 | 152 | 824.8158 |
| FC3FED8CCB2946DE54F1C5BA3656023C | 3181 | 152 | 501.9671 |
+----------------------------------+---------+------+----------+
10 rows in set (0.012 sec)
查询某段时间内平均 执行时间 排在 TOP N 的 SQL(可根据执行时间定位慢SQL):
SELECT/*+ PARALLEL(15)*/ SQL_ID,PLAN_ID, COUNT(*) AS QPS, AVG(t1.elapsed_time) RT
FROM oceanbase.GV$OB_SQL_AUDIT t1
WHERE tenant_id = 1002 AND IS_EXECUTOR_RPC = 0
AND request_time > time_to_usec(DATE_SUB(current_timestamp, INTERVAL 10 MINUTE) )
GROUP BY t1.sql_id ORDER BY RT DESC LIMIT 10;
根据sql_id
找到对应的SQL文本:
obclient [(none)]> select query_sql from oceanbase.GV$OB_PLAN_CACHE_PLAN_STAT where sql_id='A460265EC2F0763A15DD27CE9E4E2200';
+---------------------------------------------------------------------------+
| query_sql |
+---------------------------------------------------------------------------+
| SELECT i_price, i_name, i_data FROM bmsql_item WHERE i_id = 14972 |
+---------------------------------------------------------------------------+
1 row in set (0.005 sec)
分析执行计划
我们获取平均执行时间最长的三条SQL,分析他们的实际执行计划和解析执行计划。
先找出TOP SQL:
obclient [(none)]> SELECT/*+ PARALLEL(15)*/ SQL_ID,PLAN_ID, COUNT(*) AS QPS, AVG(t1.elapsed_time) RT
-> FROM oceanbase.GV$OB_SQL_AUDIT t1
-> WHERE tenant_id = 1002 AND IS_EXECUTOR_RPC = 0
-> AND request_time > time_to_usec(DATE_SUB(current_timestamp, INTERVAL 10 MINUTE) )
-> GROUP BY t1.sql_id ORDER BY RT DESC LIMIT 3;
+----------------------------------+---------+------+------------+
| SQL_ID | PLAN_ID | QPS | RT |
+----------------------------------+---------+------+------------+
| 89B757D18634B472A21A0EB8EB83ECE3 | 4035 | 1 | 11747.0000 |
| F59A700FA168324279B0DBC25E19760F | 3217 | 8 | 5686.6250 |
| C9FA41444B84AF63AD4FB24B9252F1A6 | 4025 | 7 | 2694.7143 |
+----------------------------------+---------+------+------------+
3 rows in set (0.012 sec)
第一条sql_id
找出对应的SQL文本,再手动获取执行计划:
obclient [tpcc]> explain SELECT count(*) AS low_stock FROM ( SELECT s_w_id, s_i_id, s_quantity FROM bmsql_stock WHERE s_w_id = 4 AND s_quantity < 17 AND s_i_id IN ( SELECT ol_i_id FROM bmsql_district JOIN bmsql_order_line ON ol_w_id = d_w_id AND ol_d_id = d_id AND ol_o_id >= d_next_o_id - 20 AND ol_o_id < d_next_o_id WHERE d_w_id = 4 AND d_id = 9 ) ) ;
+----------------------------------------------------------------------------------------------------------------------------------------------------------+
| Query Plan |
+----------------------------------------------------------------------------------------------------------------------------------------------------------+
| ================================================================================ |
| |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| |
| -------------------------------------------------------------------------------- |
| |0 |SCALAR GROUP BY | |1 |269 | |
| |1 |└─HASH RIGHT SEMI JOIN | |15 |269 | |
| |2 | ├─SUBPLAN SCAN |VIEW1 |15 |24 | |
| |3 | │ └─NESTED-LOOP JOIN | |15 |24 | |
| |4 | │ ├─TABLE GET |bmsql_district |1 |3 | |
| |5 | │ └─DISTRIBUTED TABLE RANGE SCAN|bmsql_order_line|1 |21 | |
| |6 | └─TABLE RANGE SCAN |bmsql_stock |1035 |156 | |
| ================================================================================ |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([T_FUN_COUNT(*)]), filter(nil), rowset=256 |
| group(nil), agg_func([T_FUN_COUNT(*)]) |
| 1 - output(nil), filter(nil), rowset=256 |
| equal_conds([bmsql_stock.s_i_id = VIEW1.ol_i_id]), other_conds(nil) |
| 2 - output([VIEW1.ol_i_id]), filter(nil), rowset=256 |
| access([VIEW1.ol_i_id]) |
| 3 - output([bmsql_order_line.ol_i_id]), filter(nil), rowset=256 |
| conds(nil), nl_params_([bmsql_district.d_next_o_id(:0)]), use_batch=false |
| 4 - output([bmsql_district.d_next_o_id]), filter([bmsql_district.d_next_o_id > bmsql_district.d_next_o_id - 20]), rowset=256 |
| access([bmsql_district.d_next_o_id]), partitions(p0) |
| is_index_back=false, is_global_index=false, filter_before_indexback[false], |
| range_key([bmsql_district.d_w_id], [bmsql_district.d_id]), range[4,9 ; 4,9], |
| range_cond([bmsql_district.d_w_id = 4], [bmsql_district.d_id = 9]) |
| 5 - output([bmsql_order_line.ol_i_id]), filter(nil), rowset=256 |
| access([bmsql_order_line.ol_i_id]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([bmsql_order_line.ol_w_id], [bmsql_order_line.ol_d_id], [bmsql_order_line.ol_o_id], [bmsql_order_line.ol_number]), range(MIN ; MAX), |
| range_cond([bmsql_order_line.ol_w_id = 4], [bmsql_order_line.ol_d_id = 9], [bmsql_order_line.ol_o_id >= :0 - 20], [bmsql_order_line.ol_o_id < :0]) |
| 6 - output([bmsql_stock.s_i_id]), filter([bmsql_stock.s_quantity < 17]), rowset=256 |
| access([bmsql_stock.s_i_id], [bmsql_stock.s_quantity]), partitions(p0) |
| is_index_back=false, is_global_index=false, filter_before_indexback[false], |
| range_key([bmsql_stock.s_w_id], [bmsql_stock.s_i_id]), range(4,MIN ; 4,MAX), |
| range_cond([bmsql_stock.s_w_id = 4]) |
+----------------------------------------------------------------------------------------------------------------------------------------------------------+
36 rows in set (0.088 sec)
它的真实执行计划:
select * from oceanbase.V$OB_PLAN_CACHE_PLAN_EXPLAIN where tenant_id=1002 and plan_id=3217 ;
执行计划重要信息:
- OPERATOR,算子的名称,比如全表扫描TABLE SCAN,真实执行计划会加上
PHY_
前缀 - NAME,算子操作的对象名称,比如表名、索引名等
- ROWS,真实执行计划里面代表实际数据行数,explain里面代表估算的行数(EST.ROWS)
- COST,算子执行时间,单位是微秒,explain里面的EST.TIME为估算时间
- PROPERTY,算子执行的详细过程
这里在查询执行计划时有一个坑要注意,oceanbase.V$OB_PLAN_CACHE_PLAN_EXPLAIN
视图只能在SQL执行的observer节点上查询,因为PLAN CACHE是按节点缓存的,通过obproxy登录的话往往查不到想要的数据,要直连observer。