Skip to content

主从同步及稳定性报告

rockeet edited this page Aug 23, 2018 · 2 revisions

主从测试

测试环境

  • CPU: Intel(R) Xeon(R) CPU E5-2630 v3 @ 2.40GHz x2 (共16核32线程)
  • 内存: DDR4 1866 MHz(共192G)
  • SSD: INTEL 730 IOPS 89000
  • CentOS Linux release 7.3.1611 (Core)

测试程序及数据

使用的测试程序为经我们修改过的 YCSB ( rpl-test 分支),测试中使用的 workload 为 FileWorkload

测试程序所使用的配置见附录 1 。

表结构及索引如下:

CREATE TABLE `lineitem` (
  `YCSB_KEY` 		varchar(30) COLLATE utf8_bin NOT NULL DEFAULT '',
  `L_SUPPKEY` 		int(11) NOT NULL,
  `L_LINENUMBER`	int(11) NOT NULL,
  `L_QUANTITY`		decimal(15,2) NOT NULL,
  `L_EXTENDEDPRICE`	decimal(15,2) NOT NULL,
  `L_DISCOUNT` 		decimal(15,2) NOT NULL,
  `L_TAX`		decimal(15,2) NOT NULL,
  `L_RETURNFLAG` 	char(1) COLLATE utf8_bin NOT NULL,
  `L_LINESTATUS` 	char(1) COLLATE utf8_bin NOT NULL,
  `L_SHIPDATE` 		date NOT NULL,
  `L_COMMITDATE` 	date NOT NULL,
  `L_RECEIPTDATE` 	date NOT NULL,
  `L_SHIPINSTRUCT` 	char(25) COLLATE utf8_bin NOT NULL,
  `L_SHIPMODE` 		char(10) COLLATE utf8_bin NOT NULL,
  `L_COMMENT` 		varchar(1024) COLLATE utf8_bin NOT NULL,
  PRIMARY KEY (`YCSB_KEY`)
) DEFAULT CHARSET=utf8 COLLATE=utf8_bin

数据样本见附录 2 ,每条数据平均大小为:TBFilled

测试以插入为主,并有 5% 的几率立刻执行 update 操作将该条记录的 L_COMMENT 更新为随机生成的 36 个字节的 uuid(如:7f451daf-b512-439b-8996-c61df9d1017c),另有 1% 的几率执行 delete 操作将该条记录删除。

数据库配置

  • 原生 MySQL : 使用 InnoDB 为存储引擎,下记为 innodb;
  • 原生 MyRocks: 使用 RocksDB 为存储引擎,下记为 rocksdb;
  • TerarkSQL:使用 TerarkDB 为存储引擎,下记为 terarkdb。

各个主、从数据库配置见附录 3 。

如下为测试结果,所有的测试均进行了一个小时,每隔 5 分钟纪录一次数据,记录的数据有:主库的插入速度(记为 OPS),从库同步情况(即 Seconds_Behind_Master, 记为 SBM)。

主 innodb 从 innodb

5 min 10 min 15 min 20 min 25 min 30 min 35 min 40 min 45 min 50 min 55 min 60 min 平均 checksum
OPS 968.2 989.7 951.7 975.1 969.5 982.7 1000.3 996.4 949.1 988.6 993.4 978.2 975 883418646
SBM 0 0 0 0 0 0 0 0 0 0 0 0 N/A 883418646

主 innodb 从 terarkdb

5 min 10 min 15 min 20 min 25 min 30 min 35 min 40 min 45 min 50 min 55 min 60 min 平均 checksum
OPS 967.3 971.3 954.5 967.8 971.2 937.2 942.8 948.3 984.1 943.1 977 966.9 966.4 18798117
SBM 9 15 24 32 43 51 56 61 64 65 66 69 N/A 18798117

在 terarkdb options 配置中设置将 __system__ 加入到 blackList(TerarkZipTable_blackListColumnFamily=__system__) 以减少 __system__ column family 在后台频繁 compact 对从库同步的影响。

5 min 10 min 15 min 20 min 25 min 30 min 35 min 40 min 45 min 50 min 55 min 60 min 平均 checksum
OPS 989 978.3 970.5 972.9 997.2 990.8 991.5 973.1 980.6 945.2 979.6 982.9 973.8 3284333877
SBM 3 4 10 10 18 25 31 39 48 43 50 55 N/A 3284333877

主 innodb 从 rocksdb

5 min 10 min 15 min 20 min 25 min 30 min 35 min 40 min 45 min 50 min 55 min 60 min 平均 checksum
OPS 1005.2 996 1017.9 980.7 964.1 1003.9 983 996.9 990.2 965.6 996.6 939.3 986.6 3842942686
SBM 11 26 34 45 57 68 81 91 100 114 123 135 N/A 3842942686

主 terarkdb 从 innodb

5 min 10 min 15 min 20 min 25 min 30 min 35 min 40 min 45 min 50 min 55 min 60 min 平均 checksum
OPS 943.8 904.7 909 896.6 954.9 954.3 948.8 889.9 951.1 923.3 906.8 931.4 925 1088205640
SBM 0 0 0 0 0 0 0 0 0 0 0 0 N/A 1088205640

适当增大 ops,可以看到半小时后从库出现延迟(?)

5 min 10 min 15 min 20 min 25 min 30 min 35 min 40 min 45 min 50 min 55 min 60 min 平均 checksum
OPS 982.5 976.8 985.6 976.9 961.6 973.7 975.4 984.8 971.9 993.1 999.7 991.7 982.8 3662635349
SBM 0 0 0 0 0 3 5 4 4 7 9 10 N/A 3662635349

主 terarkdb 从 terarkdb

5 min 10 min 15 min 20 min 25 min 30 min 35 min 40 min 45 min 50 min 55 min 60 min 平均 checksum
OPS 931.1 904 933 934.2 970.1 957.4 875 899.5 930.9 906.7 917.7 904.8 925.3 4273250936
SBM 1 0 0 0 1 2 0 0 2 0 0 0 N/A 4273250936

测试分析

从上述测试可以看出,主 innodb 从 terarkdb 时且 OPS 大于 ~950 后,会出现同步延迟。

我们当前正进行改进以解决延迟问题,改进包括有 使用 RBTree 替换原生的 SkipList,Rocksdb 更新,MyRocks 的更新等。

稳定性

稳定性方面的测试,我们构建了一个测试用程序,使用多个线程随机向 MyTerark 里发送包含建表、删表、改表、插入、删除等操作。持续跑了10个小时,MyTerark 依然可以正常运行。程序如下:

for i = 1 to N:
	start_thread(executor)
	
func executor:
	while (true):
		op = rand()
		switch op:
		case CreateTable: 	CreateTable(); break;
		case DropTable: 	DropTable(); break;
		case AlterTable: 	AlterTable(); break;
		case Insert:		Insert(); break;
		case Delete:		Delete(); break;
		case Query:			Query(); break;

// terark_N is a random selected table between terark_[1, 100]
func CreateTable() 	// create table terark_N, with 9 indexes, from tpcl
func DropTable()		// drop table terark_N
func AlterTable()	// use table terark_N, drop 2 indexes, create 2 indexes
func Insert()			// use table terark_N, insert aound 10k rows from tpch
func Delete()			// use table terark_N, delete 10k rows

附录

附录 1

测试时所使用的配置如下:

workload=com.yahoo.ycsb.workloads.FileWorkload
 
recordcount=40000000
operationcount=1500000
 
fieldnames=L_ORDERKEY,L_PARTKEY,L_SUPPKEY,L_LINENUMBER,L_QUANTITY,L_EXTENDEDPRICE,L_DISCOUNT,L_TAX,L_RETURNFLAG,L_LINESTATUS,L_SHIPDATE,L_COMMITDATE,L_RECEIPTDATE,L_SHIPINSTRUCT,L_SHIPMODE,L_COMMENT
usecustomkey=true
keyfield=0,1
fieldnum=16
 
datafile=/data/lineitem_512b.tbl
table=lineitem 
delimiter=|

threadcount=16
 
jdbc.driver=com.mysql.jdbc.Driver
db.url=jdbc:mysql://127.0.0.1:3336/ycsb
db.user=root
db.passwd=
mysql.upsert=true

maxexecutiontime=3720

附录 2

数据样本如下:

1|16605260|822776|1|17|19795.31|0.04|0.02|N|O|1996-03-13|1996-02-12|1996-03-22|DELIVER IN PERSON|TRUCK|fluffily ironic pinto beans sleep daringly pending instructions. fluffily permanent foxes mold along the furiously express ideas. ironic pinto beans cajole fluffily unusual theodolites. carefully regular theodolites across the pending pinto beans haggle even, ironic deposits. quickly special packages above the furiously regular deposits wake carefully pending requests. carefully regular courts above the pending, even accounts haggle blithely even |
1|7202072|782073|2|36|35053.56|0.09|0.06|N|O|1996-04-12|1996-02-28|1996-04-20|TAKE BACK RETURN|MAIL|n excuses. fluffily final requests wake quickly against the carefully unusual courts. packages cajole. unusual ideas nag quickly about the slyly pending warhorses! carefully ironic escapades are carefully. requests integrate blithely. slyly regular patterns alongside of the blithely ironic packages sleep carefully blithely final ideas. blithely dogged accounts according to the frays would use fluffily across the blithely final dependencies. carefully blithe instructions according to the blithely bold packages are busily carefully final dependencies: slyly bold requests across the fluffily pending pinto beans wake furiously slyly regular ide|
1|6815877|395878|3|8|14340.24|0.10|0.02|N|O|1996-01-29|1996-03-05|1996-01-31|TAKE BACK RETURN|REG AIR|y final asymptotes are unusual, special warhorses. silent foxes sleep carefully regular requests! ironic orbits haggle furiously. bold instructions sleep ironically fluffily silent dependencies. slyly special accounts solve slyly special grouches. silent dependencies sleep furiously. regular, express deposits wake. carefully final theodolites haggle blithely. quickly even foxes across the fluffily silent hockey players nag slyly pending requests. slyly ironic requests at the bold accounts wake carefully regular packages. deposits snooze blithely special reque|

附录 3

innodb(主)配置如下:

[mysqld]

character-set-server=utf8
collation-server=utf8_bin

back_log = 600
max_connections = 6000

binlog-format=ROW
secure_file_priv=""

server-id = 1

innodb(从)配置如下:

[mysqld]

character-set-server=utf8
collation-server=utf8_bin

back_log = 600
max_connections = 6000

binlog-format=ROW
secure_file_priv=""

server-id = 2
log_slave_updates = 1
read_only         = 1
replicate-ignore-db = mysql

rocksdb(从)配置如下:

[mysqld]

rocksdb
default-storage-engine=rocksdb
skip-innodb
default-tmp-storage-engine=MyISAM
character-set-server=utf8
collation-server=utf8_bin

binlog-format=ROW
secure_file_priv=""
rocksdb_commit_in_the_middle=ON
rocksdb_bulk_load_size=1000
rocksdb_strict_collation_check=OFF

server-id = 2
log_slave_updates = 1
read_only         = 1

replicate-ignore-db = mysql

terarkdb(主)配置如下:

[mysqld]

rocksdb
default-storage-engine=rocksdb
skip-innodb
default-tmp-storage-engine=MyISAM
character-set-server=utf8
collation-server=utf8_bin

back_log = 600
max_connections = 6000

binlog-format=ROW
secure_file_priv=""
rocksdb_commit_in_the_middle=ON
rocksdb_bulk_load_size=1000
rocksdb_strict_collation_check=OFF

server-id = 1

terarkdb(从)配置如下:

[mysqld]

rocksdb
default-storage-engine=rocksdb
skip-innodb
default-tmp-storage-engine=MyISAM
character-set-server=utf8
collation-server=utf8_bin

back_log = 600
max_connections = 6000

binlog-format=ROW
secure_file_priv=""
rocksdb_commit_in_the_middle=ON
rocksdb_bulk_load_size=1000
rocksdb_strict_collation_check=OFF

server-id = 2
log_slave_updates = 1
read_only         = 1

replicate-ignore-db = mysql

terarkdb options 配置如下:

TerarkZipTable_localTempDir=$PWD/terark-temp
TerarkZipTable_keyPrefixLen=4
TerarkZipTable_offsetArrayBlockUnits=128
TerarkZipTable_indexCacheRatio=0.001
TerarkZipTable_extendedConfigFile=$PWD/license
TerarkUseDivSufSort=1