python3.6.2用pyinstaller3.4報錯_OceanBase 2.2 版本體驗：用 BenchmarkSQL 跑 TPC-C

OB君：好消息！「 OceanBase 2.2 版本」正式上線官網啦！（點擊閱讀原文即可直接下載）OceanBase 2.2版本是成功支撐2019年天貓雙11大促的穩定版本，同時也是用于TPC-C測試且榮登TPC-C性能榜首的版本。我們將在接下來的時間里為大家持續推出 “OceanBase 2.2 手把手系列” ，將手把手帶大家一起體驗OceanBase 2.2的強大功能。歡迎持續關注！

引言

OceanBase 2.2版本近期已通過官網提供下載（https://oceanbase.alipay.com/download/resource），2.2支持Oracle租戶。OceanBase在2019年10月2日榮膺國際事務委員會(TPC)審計發布的TPC-C基準測試榜首，用的就是Oracle租戶。TPC-C測試使用了207多臺阿里云高配ECS服務器，是因為TPC-C標準對應用、數據庫等規范非常細致嚴格。一般來說普通企業或個人很難有那樣的條件去測試。

BenchmarkSQL是開源的TPC-C測試程序，它弱化了TPC-C的關鍵標準(數據分布和應用執行行為方面)，使得用幾臺服務器就可以跑TPC-C成為可能。當然這個結果不能跟官方TPC-C的結果相比較。不過，使用BenchmarkSQL來比較不同的數據庫的事務處理能力還是有一定參考意義的，尤其是相比Sysbench而言。

OceanBase測試租戶準備

1.sys租戶參數修改

BenchmarkSQL會加載大量數據，短時間內對OceanBase內存消耗速度會很快，因此需要針對內存凍結合并和限流參數做一些調優。在sys租戶執行：

ALTER SYSTEM SET enable_merge_by_turn=FALSE;
ALTER SYSTEM set minor_freeze_times=100;
ALTER SYSTEM set freeze_trigger_percentage=70;
ALTER SYSTEM set writing_throttling_trigger_percentage=70 tenant='obbmsql';
ALTER SYSTEM set writing_throttling_maximum_duration='10m' tenant='obbmsql';
show parameters where name  in ('minor_freeze_times','freeze_trigger_percentage');

注意：業務租戶限流參數的修改是在sys租戶里，需要指定相應的租戶名。然后查看確認需要到業務租戶里。
在業務租戶執行：

SHOW parameters WHERE name IN ('writing_throttling_trigger_percentage','writing_throttling_maximum_duration');

2. 業務租戶參數修改

OceanBase跟Oracle/MySQL相比，會有個默認SQL超時和事務超時機制。這個可能會導致后面查看修改數據的SQL報錯。所以先修改一下這些參數。

set global recyclebin=off;
set global ob_query_timeout=1000000000;
set global ob_trx_idle_timeout=1200000000;
set global ob_trx_timeout=1000000000;

此外，需要為bmsql準備一個單獨的schema(即用戶)。

drop user tpcc cascade;create user tpcc identified by 123456;
grant all privileges on tpcc.* to tpcc with grant option ;
grant create, drop on *.* to tpcc;

3. OBProxy配置修改

OBProxy是OceanBase的訪問代理，其內部一些參數也可能影響性能。如下面的壓縮參數對CPU有一定消耗，測試時可以關閉。

$ obclient -h127.1 -uroot@sys#obdemo -P2883 -p123456 -c -A oceanbasealter proxyconfig set enable_compression_protocol=False;
show proxyconfig like 'enable_compression_protocol';

該參數修改后，需要重啟obproxy進程

[admin@xxx /home/admin]
$kill -9 `pidof obproxy`[admin@h07d17167.sqa.eu95 /home/admin]
$cd /opt/taobao/install/obproxy[admin@xxx /opt/taobao/install/obproxy]
$bin/obproxy
bin/obproxy

BenchmarkSQL準備

BenchmarkSQL 官方下載地址是： https://sourceforge.net/projects/benchmarksql/ ，下載后請參考 HOW-TO-RUN.txt 中說明先編譯安裝BenchmarkSQL。然后按下面建議修改部分腳本增加對OceanBase支持。也可以直接下載我編譯修改好的BenchmarkSQL，地址是：https://github.com/obpilot/benchmarksql-5.0 。

1. 準備OceanBase驅動文件

BenchmarkSQL是通過jdbc連接各個數據庫的。此次OceanBase的測試租戶是Oracle類型，所以需要借用 lib/oracle 目錄，然后把相關jar包一并放入其中。其中 oceanbase-client-*.jar 是OceanBase提供的，其他jar包可以從互聯網獲取。

[admin@xxx /home/admin/benchmarksql-5.0]$ll lib/oracle/
total 3728
-rwxr-xr-x 1 admin admin   52988 Jul 12  2019 commons-cli-1.3.1.jar
-rwxr-xr-x 1 admin admin  245274 Jul 12  2019 commons-lang-2.3.jar
-rwxr-xr-x 1 admin admin 2256213 Jul 12  2019 guava-18.0.jar
-rwxr-xr-x 1 admin admin   54495 Jul 12  2019 json-20160810.jar
-rwxr-xr-x 1 admin admin 1121698 Dec  3 15:04 oceanbase-client-1.0.8.jar
-rwxr-xr-x 1 admin admin     174 Jul 12  2019 README.txt
-rwxr-xr-x 1 admin admin   76997 Jul 12  2019 toolkit-common-logging-1.10.jar

2. 準備OB配置文件

$cat props.ob
db=oracle
driver=com.alipay.oceanbase.obproxy.mysql.jdbc.Driver
conn=jdbc:oceanbase://127.0.0.1:2883/tpcc?useUnicode=true&characterEncoding=utf-8
user=tpcc@obbmsql#obdemo
password=123456warehouses=10
loadWorkers=10
//fileLocation=/home/t4/tmpterminals=10
//To run specified transactions per terminal- runMins must equal zero
runTxnsPerTerminal=0
//To run for specified minutes- runTxnsPerTerminal must equal zero
runMins=10
//Number of total transactions per minute
limitTxnsPerMin=0//Set to true to run in 4.x compatible mode. Set to false to use the
//entire configured database evenly.
terminalWarehouseFixed=true//The following five values must add up to 100
newOrderWeight=45
paymentWeight=43
orderStatusWeight=4
deliveryWeight=4
stockLevelWeight=4// Directory name to create for collecting detailed result data.
// Comment this out to suppress.
resultDirectory=my_result_%tY-%tm-%td_%tH%tM%tS
osCollectorScript=./misc/os_collector_linux.py
osCollectorInterval=1
//osCollectorSSHAddr=user@dbhost
//osCollectorDevices=net_eth0 blk_sda

注意：a. 倉庫數(warehouses)決定了數據量。正式的壓測倉庫數一般在10000以上。b. loadworkers數決定了數據加載的性能。如果OceanBase租戶資源很小(尤其是內存資源)，那加載速度也不要太快；否則容易把租戶內存打爆。c. 并發數(terminals)是后期做TPC-C測試的客戶端并發數。這個每次測試都可以調整，以方便觀察不同壓力下的性能。

d. 壓測時間(runMin)是每次測試時間，越長測試結果越好且穩定。因為有時候數據訪問有個預熱過程，效果會體現在內存命中率上。

3. 創建BenchmarkSQL相關表

1）建表腳本

該SQL腳本不需要直接執行。

create table bmsql_config (cfg_name    varchar2(30) primary key,cfg_value   varchar2(50)
);create tablegroup tpcc_group  partition by hash partitions 12;create table bmsql_warehouse (w_id        integer   not null,w_ytd       decimal(12,2),w_tax       decimal(4,4),w_name      varchar2(10),w_street_1  varchar2(20),w_street_2  varchar2(20),w_city      varchar2(20),w_state     char(2),w_zip       char(9),primary key(w_id)
)tablegroup='tpcc_group' partition by hash(w_id) partitions 12;create table bmsql_district (d_w_id       integer       not null,d_id         integer       not null,d_ytd        decimal(12,2),d_tax        decimal(4,4),d_next_o_id  integer,d_name       varchar2(10),d_street_1   varchar2(20),d_street_2   varchar2(20),d_city       varchar2(20),d_state      char(2),d_zip        char(9),PRIMARY KEY (d_w_id, d_id)
)tablegroup='tpcc_group' partition by hash(d_w_id) partitions 12;create table bmsql_customer (c_w_id         integer        not null,c_d_id         integer        not null,c_id           integer        not null,c_discount     decimal(4,4),c_credit       char(2),c_last         varchar2(16),c_first        varchar2(16),c_credit_lim   decimal(12,2),c_balance      decimal(12,2),c_ytd_payment  decimal(12,2),c_payment_cnt  integer,c_delivery_cnt integer,c_street_1     varchar2(20),c_street_2     varchar2(20),c_city         varchar2(20),c_state        char(2),c_zip          char(9),c_phone        char(16),c_since        timestamp,c_middle       char(2),c_data         varchar2(500),PRIMARY KEY (c_w_id, c_d_id, c_id)
)tablegroup='tpcc_group' use_bloom_filter=true compress partition by hash(c_w_id) partitions 12;create sequence bmsql_hist_id_seq;create table bmsql_history (hist_id  integer,h_c_id   integer,h_c_d_id integer,h_c_w_id integer,h_d_id   integer,h_w_id   integer,h_date   timestamp,h_amount decimal(6,2),h_data   varchar2(24)
)tablegroup='tpcc_group' use_bloom_filter=true compress partition by hash(h_w_id) partitions 12;create table bmsql_new_order (no_w_id  integer   not null ,no_d_id  integer   not null,no_o_id  integer   not null,PRIMARY KEY (no_w_id, no_d_id, no_o_id)
)tablegroup='tpcc_group' use_bloom_filter=true compress partition by hash(no_w_id) partitions 12;create table bmsql_oorder (o_w_id       integer      not null,o_d_id       integer      not null,o_id         integer      not null,o_c_id       integer,o_carrier_id integer,o_ol_cnt     integer,o_all_local  integer,o_entry_d    timestamp,PRIMARY KEY (o_w_id, o_d_id, o_id)
)tablegroup='tpcc_group' use_bloom_filter=true compress partition by hash(o_w_id) partitions 12;create table bmsql_order_line (ol_w_id         integer   not null,ol_d_id         integer   not null,ol_o_id         integer   not null,ol_number       integer   not null,ol_i_id         integer   not null,ol_delivery_d   timestamp,ol_amount       decimal(6,2),ol_supply_w_id  integer,ol_quantity     integer,ol_dist_info    char(24),PRIMARY KEY (ol_w_id, ol_d_id, ol_o_id, ol_number)
)tablegroup='tpcc_group' use_bloom_filter=true compress partition by hash(ol_w_id) partitions 12;create table bmsql_item (i_id     integer      not null,i_name   varchar2(24),i_price  decimal(5,2),i_data   varchar2(50),i_im_id  integer,PRIMARY KEY (i_id)
)use_bloom_filter=true compress locality='F,R{all_server}@zone1, F,R{all_server}@zone2, F,R{all_server}@zone3' primary_zone='zone1'  duplicate_scope='cluster';create table bmsql_stock (s_w_id       integer       not null,s_i_id       integer       not null,s_quantity   integer,s_ytd        integer,s_order_cnt  integer,s_remote_cnt integer,s_data       varchar2(50),s_dist_01    char(24),s_dist_02    char(24),s_dist_03    char(24),s_dist_04    char(24),s_dist_05    char(24),s_dist_06    char(24),s_dist_07    char(24),s_dist_08    char(24),s_dist_09    char(24),s_dist_10    char(24),PRIMARY KEY (s_w_id, s_i_id)
)tablegroup='tpcc_group' use_bloom_filter=true compress partition by hash(s_w_id) partitions 12;

注意：

a. 建表語句中的分區數目可以根據實際情況調整，跟集群節點數有關。如果集群是3臺(1-1-1)，建議是6個或6的倍數；如果集群是6臺(2-2-2)，建議是12個或12的倍數；如果集群是9臺(3-3-3)，建議是36個或36的倍數。這樣是方便后期彈性伸縮測試的時候能盡可能保證每個節點上的分區數均衡。

b. 上面bmsql_item使用了【復制表】功能，在租戶的所有節點上都會有一個副本。當然主副本始終只有一個。有關【復制表】功能介紹請參考《OceanBase事務引擎特性和應用實踐分享》。

c. 建表語句不包含非主鍵索引，是為了后面加載數據性能更快。

2）建表

./runSQL.sh props.ob ./sql.oceanbase/tableCreates.sql

建表后，可以查看主副本分布

SELECT  t1.tenant_id,t1.tenant_name,t2.database_name,t3.table_id,t3.table_Name,t3.tablegroup_id,t3.part_num,t4.partition_Id,t4.zone,t4.svr_ip,t4.role, round(t4.data_size/1024/1024) data_size_mb
from `gv$tenant` t1join `gv$database` t2 on (t1.tenant_id = t2.tenant_id)join gv$table t3 on (t2.tenant_id = t3.tenant_id    and t2.database_id = t3.database_id and t3.index_type = 0)left join `__all_virtual_meta_table` t4 on (t2.tenant_id = t4.tenant_id and ( t3.table_id = t4.table_id or t3.tablegroup_id = t4.table_id ) and t4.role in (1))
where t1.tenant_id = 1001
order by t3.tablegroup_id, t4.partition_Id, t3.table_name ;

4. 加載數據

1）開始加載數據

./runLoader.sh props.ob

2）觀察數據加載性能

為了對數據寫入速度進行觀察，可以在sys租戶下反復執行下面SQL，主要是觀察增量內存增速和增量內存總量，以及是否接近總增量內存限制。

SELECT tenant_id, ip, round(active/1024/1024) active_mb, round(total/1024/1024) total_mb, round(freeze_trigger/1024/1024) freeze_trg_mb, round(mem_limit/1024/1024) mem_limit_mb, freeze_cnt , round((active/freeze_trigger),2) freeze_pct, round(total/mem_limit, 2) mem_usage
FROM `gv$memstore`
WHERE tenant_id IN (1001)
ORDER BY tenant_id, ip;

當然，觀察數據加載另外一個方法就是使用監控。OCP的監控或者dooba腳本監控。

python dooba.py -h 127.1 -uroot@sys#obdemo -P2883 -p123456

dooba 進去后，默認是sys租戶。按字母小寫'c'，選擇業務租戶。按數字'1'查看幫助，數字'2'查看租戶總覽，數字'3'查看租戶的機器性能信息，按TAB切換當前焦點，按字母小寫'd' 刪除當前TAB，按字母大寫R 恢復所有TAB。總覽里的NET TAB沒有意義可以刪除以節省屏幕空間。

5. 建索引

索引很少，就2條。由于相關表是分區表，可以建全局索引或者本地索引。我們建本地索引。

$cat ./sql.oceanbase/indexCreates.sql
create index bmsql_customer_idx1on  bmsql_customer (c_w_id, c_d_id, c_last, c_first) local;
create  index bmsql_oorder_idx1on  bmsql_oorder (o_w_id, o_d_id, o_carrier_id, o_id) local;

開始建索引。OceanBase建索引很快就會返回，索引構建是異步的。

./runSQL.sh props.ob ./sql.oceanbase/indexCreates.sql

6. 數據校驗

檢查一下各個表的數據量

obclient> select /*+ parallel(16) read_consistency(weak) */ count(*) from TPCC.BMSQL_CONFIG;
*+ parallel(16) read_consistency(weak) */ count(*) from TPCC.BMSQL_STOCK;+----------+
| COUNT(*) |
+----------+
|        4 |
+----------+
1 row in set (0.06 sec)obclient> select /*+ parallel(16) read_consistency(weak) */ count(*) from TPCC.BMSQL_WAREHOUSE;
+----------+
| COUNT(*) |
+----------+
|       10 |
+----------+
1 row in set (0.06 sec)
obclient> select /*+ parallel(16) read_consistency(weak) */ count(*) from TPCC.BMSQL_DISTRICT;
+----------+
| COUNT(*) |
+----------+
|      100 |
+----------+
1 row in set (0.06 sec)obclient> select /*+ parallel(16) read_consistency(weak) */ count(*) from TPCC.BMSQL_CUSTOMER;
+----------+
| COUNT(*) |
+----------+
|   300000 |
+----------+
1 row in set (0.34 sec)obclient> select /*+ parallel(16) read_consistency(weak) */ count(*) from TPCC.BMSQL_HISTORY;
+----------+
| COUNT(*) |
+----------+
|   300000 |
+----------+
1 row in set (0.10 sec)obclient> select /*+ parallel(16) read_consistency(weak) */ count(*) from TPCC.BMSQL_NEW_ORDER;
+----------+
| COUNT(*) |
+----------+
|    90000 |
+----------+
1 row in set (0.07 sec)obclient> select /*+ parallel(16) read_consistency(weak) */ count(*) from TPCC.BMSQL_OORDER;
+----------+
| COUNT(*) |
+----------+
|   300000 |
+----------+
1 row in set (0.11 sec)obclient> select /*+ parallel(16) read_consistency(weak) */ count(*) from TPCC.BMSQL_ORDER_LINE;
+----------+
| COUNT(*) |
+----------+
|  3001782 |
+----------+
1 row in set (0.27 sec)obclient> select /*+ parallel(16) read_consistency(weak) */ count(*) from TPCC.BMSQL_ITEM;
+----------+
| COUNT(*) |
+----------+
|   100000 |
+----------+
1 row in set (0.08 sec)obclient> select /*+ parallel(16) read_consistency(weak) */ count(*) from TPCC.BMSQL_STOCK;
+----------+
| COUNT(*) |
+----------+
|  1000000 |
+----------+
1 row in set (0.63 sec)

為了避免產生的數據不符合規范(如中間報錯導致有事務失敗)，運行下面校驗腳本

#!/usr/bin/shcc1="
SELECT /*+ no_use_px parallel(8) */ * FROM(SELECT w.w_id, w.w_ytd, d.sum_d_ytdFROM bmsql_warehouse w,(SELECT /*+ no_use_px parallel(8) */ d_w_id, sum(d_ytd) sum_d_ytd FROM bmsql_district GROUP BY d_w_id) dWHERE w.w_id= d.d_w_id
) x
WHERE w_ytd != sum_d_ytd;
"
cc2="
SELECT /*+ no_use_px parallel(8) */ * FROM(SELECT d.d_w_id, d.d_id, d.d_next_o_id, o.max_o_id, no.max_no_o_idFROM bmsql_district d,(SELECT /*+ no_use_px parallel(8) */ o_w_id, o_d_id, MAX(o_id) max_o_id FROM bmsql_oorder GROUP BY o_w_id, o_d_id) o,(SELECT /*+ no_use_px parallel(8) */ no_w_id, no_d_id, MAX(no_o_id) max_no_o_id FROM bmsql_new_order GROUP BY no_w_id, no_d_id) noWHERE d.d_w_id= o.o_w_id AND d.d_w_id= no.no_w_id AND d.d_id= o.o_d_id AND d.d_id= no.no_d_id
) x
WHERE d_next_o_id - 1!= max_o_id OR d_next_o_id - 1!= max_no_o_id;
"cc3="
SELECT /*+ no_use_px paratLel(8) */ * FROM(SELECT /*+ no_use_px parallel(8) */ no_w_id, no_d_id, MAX(no_o_id) max_no_o_id, MIN(no_o_id) min_no_o_id, COUNT(*) count_noFROM bmsql_new_orderGROUP BY no_w_id, no_d_Id
) x
WHERE max_no_o_id - min_no_o_id+ 1!= count_no;
"cc4="
SELECT /*+ no_use_px parallel(8) */ * FROM (SELECT o.o_w_id, o.o_d_id, o.sum_o_ol_cnt, ol.count_olFROM (SELECT /*+ no_use_px parallel(8) */ o_w_id, o_d_id, SUM(o_ol_cnt) sum_o_ol_cnt FROM bmsql_oorder GROUP BY o_w_id, o_d_id) o,(SELECT /*+ no_use_px parallel(8) */ ol_w_id, ol_d_id, COUNT(*) count_ol FROM bmsql_order_line GROUP BY ol_w_id, ol_d_id) olWHERE o.o_w_id = ol.ol_w_id AND o.o_d_id = ol.ol_d_id
) x
WHERE sum_o_ol_cnt != count_ol;
"cc5="
SELECT /*+ no_use_px parallel(8) */ * FROM (SELECT o.o_w_id, o.o_d_id, o.o_id, o.o_carrier_id, no.count_noFROM bmsql_oorder o,(SELECT /*+ no_use_px parallels) */ no_w_id, no_d_id, no_o_id, COUNT(*) count_no FROM bmsql_new_order GROUP BY no_w_id, no_d_id, no_o_id) noWHERE o.o_w_id = no.no_w_id AND o.o_d_id = no.no_d_id AND o.o_id = no.no_o_id
) x
WHERE (o_carrier_id IS NULL AND count_no = 0) OR (o_carrier_id IS NOT NULL AND count_no != 0);
"cc6="
SELECT /*+ no_use_px parallel(8) */ * FROM (SELECT o.o_w_id, o.o_d_id, o.o_id, o.o_ol_cnt, ol.count_olFROM bmsql_oorder o,(SELECT /*+ no_use_px parallel(8) */ ol_w_id, ol_d_id, ol_o_id, COUNT(*) count_ol FROM bmsql_order_line GROUP BY ol_w_id, ol_d_id, ol_o_id) olWHERE o.o_w_id = ol.ol_w_id AND o.o_d_id = ol.ol_d_id AND o.o_id = ol.ol_o_id
) x
WHERE o_ol_cnt != count_ol;
"
cc7="
SELECT /*+ no_use_px parallel(8) */ * FROM (
SELECT /*+ no_use_px parallel(8) */ * FROM (SELECT o.o_w_id, o.o_d_id, o.o_id, o.o_ol_cnt, ol.count_olFROM bmsql_oorder o,(SELECT /*+ no_use_px parallel(8) */ ol_w_id, ol_d_id, ol_o_id, COUNT(*) count_ol FROM bmsql_order_line GROUP BY ol_w_id, ol_d_id, ol_o_id) olWHERE o.o_w_id = ol.ol_w_id AND o.o_d_id = ol.ol_d_id AND o.o_id = ol.ol_o_id
) x
WHERE o_ol_cnt != count_ol;
"cc7="
SELECT /*+ no_use_px parallel(8) */ * FROM (SELECT /*+ no_use_px parallel(8) */ ol.ol_w_id, ol.ol_d_id, ol.ol_o_id, ol.ol_delivery_d, o.o_carrier_idFROM bmsql_order_line ol, bmsql_oorder oWHERE ol.ol_w_id = o.o_w_id ANDol.ol_d_id = o.o_d_id ANDol.ol_o_id = o.o_id
) x
WHERE (ol_delivery_d IS NULL AND o_carrier_id IS NOT NULL) OR(ol_delivery_d IS NOT NULL AND o_carrier_id IS NULL);
"cc8="
SELECT /*+ no_use_px parallel(8) */ * FROM (SELECT w.w_id, w.w_ytd, h.sum_h_amountFROM bmsql_warehouse w,(SELECT /*+ no_use_px parallel(8) */ h_w_id, SUM(h_amount) sum_h_amount FROM bmsql_history GROUP BY h_w_id) hWHERE w.w_id = h.h_w_id) x
WHERE w_ytd != sum_h_amount;
"cc9="
SELECT /*+ no_use_px parallel(8) */ * FROM (SELECT d.d_w_id, d.d_id, d.d_ytd, h.sum_h_amountFROM bmsql_district d,(SELECT /*+ no_use_px parallel(8) */ h_w_id, h_d_id, SUM(h_amount) sum_h_amount FROM bmsql_history GROUP BY h_w_id, h_d_id) hWHERE d.d_w_id = h.h_w_id AND d.d_id = h.h_d_id
) x
WHERE d_ytd != sum_h_amount;
"cc_list="$cc1|$cc2|$cc3|$cc4|$cc5|$cc6|$cc7|$cc8|$cc9"
oldIFS=$IFS
IFS="|"counter=0
for sql in $cc_list
dolet counter++echo `date '+%F %X'`" cc$counter start"obclient -Dtpcc -h127.1 -P2883  -utpcc@obbmsql#obdemo -p123456 -A -c -e "$sql"#echo $?if [[ $? -ne 0 ]];thenIFS=$oldIFSecho `date '+%F %X'`" cc$counter failed"exit 1fiecho `date '+%F %X'`" cc$counter finished"
done
IFS=$oldIFS

BenchmarkSQL TPC-C場景分析

1. E-R模型

2. 場景SQL

TPC-C 系統需要處理的交易有以下五種：場景名場景描述交易占比New-Order客戶輸入一筆新的訂貨交易45%Payment更新客戶賬戶余額以反應其支付狀況43%Delivery發貨（批處理交易）4%Order-Status查詢客戶最近交易的狀態4%Stock-Level查詢倉庫庫存狀況，以便能夠及時補貨。4%場景的比例是在數據庫配置文件中定義的。這里是默認值。對于前四種類型的交易，要求響應時間在 5 秒以內；對于庫存狀況的查詢交易，要求響應時間在 20 秒以內。這五種交易作用在圖 1 所示的九張表上，事務操作類型包括更新，插入，刪除和取消操作。

下面是我事先通過OceanBase的全量SQL審計抓出的TPCC的事務SQL(做過去重,但可能不全)。

1）場景1：New-Order

SELECT d_tax, d_next_o_id FROM bmsql_district WHERE d_w_id = 778 AND d_id = 5 FOR UPDATE;
SELECT c_discount, c_last, c_credit, w_tax FROM bmsql_customer JOIN bmsql_warehouse ON (w_id = c_w_id) WHERE c_w_id = 778 AND c_d_id = 5 AND c_id = 2699;
UPDATE bmsql_district SET d_next_o_id = d_next_o_id + 1 WHERE d_w_id = 778 AND d_id = 5 ;
INSERT INTO bmsql_oorder ( o_id, o_d_id, o_w_id, o_c_id, o_entry_d, o_ol_cnt, o_all_local) VALUES (5686, 5, 778, 2699, timestamp '2020-01-04 13:49:34.137', 8, 1);
INSERT INTO bmsql_new_order ( no_o_id, no_d_id, no_w_id) VALUES (5686, 5, 778);
SELECT i_price, i_name, i_data FROM bmsql_item WHERE i_id = 7752 ;   -- 循環8次
SELECT s_quantity, s_data, s_dist_01, s_dist_02, s_dist_03, s_dist_04, s_dist_05, s_dist_06, s_dist_07, s_dist_08, s_dist_09, s_dist_10 FROM bmsql_stock WHERE s_w_id = 778 AND s_i_id = 7752 FOR UPDATE;  -- 循環8次
SHOW VARIABLES WHERE Variable_name = 'tx_read_only';
UPDATE bmsql_stock SET s_quantity = 47, s_ytd = s_ytd + 8, s_order_cnt = s_order_cnt + 1, s_remote_cnt = s_remote_cnt + 0 WHERE s_w_id = 778 AND s_i_id = 7752;  -- 循環8次
SHOW VARIABLES WHERE Variable_name = 'tx_read_only';
INSERT INTO bmsql_order_line ( ol_o_id, ol_d_id, ol_w_id, ol_number, ol_i_id, ol_supply_w_id, ol_quantity, ol_amount, ol_dist_info) VALUES (5686, 5, 778, 1, 7752, 778, 8, 589.36, 'lYvcNHkOvt3iNoBb5W29umGO');  -- 循環8次
COMMIT;

2）場景2：New-Order

SELECT c_id FROM bmsql_customer WHERE c_w_id = 778 AND c_d_id = 2 AND c_last = 'PRICALLYPRES' ORDER BY c_first;
SELECT c_first, c_middle, c_last, c_balance FROM bmsql_customer WHERE c_w_id = 778 AND c_d_id = 2 AND c_id = 2694;
SELECT o_id, o_entry_d, o_carrier_id FROM bmsql_oorder WHERE o_w_id = 778 AND o_d_id = 2 AND o_c_id = 2694 AND o_id = ( SELECT max(o_id) FROM bmsql_oorder WHERE o_w_id = 778 AND o_d_id = 2 AND o_c_id = 2694 );
SELECT ol_i_id, ol_supply_w_id, ol_quantity, ol_amount, ol_delivery_d FROM bmsql_order_line WHERE ol_w_id = 778 AND ol_d_id = 2 AND ol_o_id = 4494 ORDER BY ol_w_id, ol_d_id, ol_o_id, ol_number;
ROLLBACK;

3）場景3：Payment

UPDATE bmsql_district SET d_ytd = d_ytd + 4806.11 WHERE d_w_id = 778 AND d_id = 10;
SELECT d_name, d_street_1, d_street_2, d_city, d_state, d_zip FROM bmsql_district WHERE d_w_id = 778 AND d_id = 10;
UPDATE bmsql_warehouse SET w_ytd = w_ytd + 4806.11 WHERE w_id = 778;
SELECT w_name, w_street_1, w_street_2, w_city, w_state, w_zip FROM bmsql_warehouse WHERE w_id = 778 ;
SELECT c_id FROM bmsql_customer WHERE c_w_id = 778 AND c_d_id = 10 AND c_last = 'ESEBAROUGHT' ORDER BY c_first;
SELECT c_first, c_middle, c_last, c_street_1, c_street_2, c_city, c_state, c_zip, c_phone, c_since, c_credit, c_credit_lim, c_discount, c_balance FROM bmsql_customer WHERE c_w_id = 778 AND c_d_id = 10 AND c_id = 502 FOR UPDATE;
UPDATE bmsql_customer SET c_balance = c_balance - 4806.11, c_ytd_payment = c_ytd_payment + 4806.11, c_payment_cnt = c_payment_cnt + 1 WHERE c_w_id = 778 AND c_d_id = 10 AND c_id = 502;
INSERT INTO bmsql_history ( h_c_id, h_c_d_id, h_c_w_id, h_d_id, h_w_id, h_date, h_amount, h_data) VALUES (502, 10, 778, 10, 778, timestamp '2020-01-04 13:49:34.148', 4806.11, 'HfYovpM6 b6aJtf2Xk6');
COMMIT;

4）場景4：

SELECT count(*) AS low_stock FROM ( SELECT s_w_id, s_i_id, s_quantity FROM bmsql_stock WHERE s_w_id = 778 AND s_quantity < 10 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 = 778 AND d_id = 1 ) );
ROLLBACK;

5）場景5：

SELECT no_o_id FROM bmsql_new_order WHERE no_w_id = 778 AND no_d_id = 1 ORDER BY no_o_id ASC;
DELETE FROM bmsql_new_order WHERE no_w_id = 778 AND no_d_id = 1 AND no_o_id = 4488;
UPDATE bmsql_oorder SET o_carrier_id = 2 WHERE o_w_id = 778 AND o_d_id = 1 AND o_id = 4488;
SELECT o_c_id FROM bmsql_oorder WHERE o_w_id = 778 AND o_d_id = 1 AND o_id = 4488;
UPDATE bmsql_order_line SET ol_delivery_d = timestamp '2020-01-04 13:49:34.181' WHERE ol_w_id = 778 AND ol_d_id = 1 AND ol_o_id = 4488;
SELECT sum(ol_amount) AS sum_ol_amount FROM bmsql_order_line WHERE ol_w_id = 778 AND ol_d_id = 1 AND ol_o_id = 4488;
UPDATE bmsql_customer SET c_balance = c_balance + 3733.14, c_delivery_cnt = c_delivery_cnt + 1 WHERE c_w_id = 778 AND c_d_id = 1 AND c_id = 1260;
<---循環8次--->
commit

注意：可能還有事務SQL沒有找到。

3. TPC-C輸出指標

TPC-C 的測試結果主要有兩個指標：

流量指標（tpmC）：描述了系統在執行 Payment，Order-Status，Delivery，Stock-level 這四種交易的同時，每分鐘可以處理的 New-Order交易的數量。流量指標值越大越好。
tpm 是 transactions per minute 的簡稱；C 指 TPC 中的 C 基準程序。它的定義是每分鐘內系統處理的新訂單個數。要注意的是，在處理新訂單的同時，系統還要按圖 1 的要求處理其它 4 類事務請求。從圖 1 可以看出，新訂單請求不可能超出全部事務請求的 45％，因此，當一個系統的性能為 1000tpmC 時，它每分鐘實際處理的請求數是 2000 多個。
性價比（Price/tpmC）：測試系統價格與流量指標的比值。性價比越小越好。

運行BenchmarkSQL TPC-C測試

1. OceanBase內存凍結與合并

前面加載了大量數據，OceanBase的增量都在內存中，需要做一次major freeze以釋放增量內存。這個事件分兩步。一是凍結操作，這個很快。二是合并操作，這個跟增量數據量有關，通常要幾分鐘或者幾十分鐘。每次重復測試的時候都建議做一次major freeze事件以釋放內存，弊端就是隨后測試中內存數據訪問又需要一個預熱過程。

1）觀察內存增量使用情況

select tenant_id, ip, round(active/1024/1024) active_mb, round(total/1024/1024) total_mb, round(freeze_trigger/1024/1024) freeze_trg_mb, round(mem_limit/1024/1024) mem_limit_mb, freeze_cnt, round(total/mem_limit,2) total_pct
from `gv$memstore` where tenant_id>1001 order by tenant_id;

2）發起內存major freeze事件

ALTER SYSTEM major freeze;

3）觀察合并進度

觀察合并事件

SELECT DATE_FORMAT(gmt_create, '%b%d %H:%i:%s') gmt_create_ , module, event, name1, value1, name2, value2, rs_svr_ip
FROM __all_rootservice_event_history
WHERE 1 = 1 AND module IN ('daily_merge')
ORDER BY gmt_create DESC
LIMIT 100;

觀察合并進度

select ZONE,svr_ip,major_version,ss_store_count ss_sc, merged_ss_store_count merged_ss_sc, modified_ss_store_count modified_ss_sc, date_format(merge_start_time, "%h:%i:%s") merge_st, date_format(merge_finish_time,"%h:%i:%s") merge_ft, merge_process
from `__all_virtual_partition_sstable_image_info` s
order by major_version, zone, svr_ip ;

2. 跑TPC-C測試

1）運行測試程序

$./runBenchmark.sh props.ob

2）性能監控

注意：這個監控界面重點關注QPS/TPS、以及相應的RT、增量內存的增量和總量占比等。此外還能看出測試過程中還是有不少物理讀IO。

注意：

這個監控界面里的重點看各個節點的QPS和TPS分布，以及遠程SQL的數量占總QPS的比例(SRC/SLC)。TPC-C業務定義會有約1%的遠程倉庫交易事務，在OceanBase里這個交易又有一定概率是分布式事務。

3）TPC-C報告

運行結束后會生成結果。

從圖上看，tpmC結果是48204。這個業務租戶總資源是20C25G*3。倉庫數只有10倉太少了，如果機器好一點，并做10000倉，這個結果應該會更高。

運行同時還生成了一個文件夾

$ll my_result_2020-01-13_175531/
total 16
drwxrwxr-x 2 admin admin 4096 Jan 13 17:55 data
-rw-rw-r-- 1 admin admin 5130 Jan 13 18:10 report.html
-rwxr-xr-x 1 admin admin 1050 Jan 13 17:55 run.properties

以上就是通過BenchmarkSQL跑TPC-C測試程序的完整過程，感興趣的同學也可以按照上述步驟體驗。有更多反饋歡迎在文章評論區留言。