mysql 简单更新 4000 万和 128GB RAM 花费太多时间

Question

我们在对单个表进行简单更新时遇到了需要很长时间的问题。该表包含约 4000 万行。

该作业每天都会运行，截断表并从该表中的其他源插入新数据。

这是表格:

 CREATE TABLE temp (
  NO int(4) NOT NULL AUTO_INCREMENT,
  DATE1 date DEFAULT NULL,
  CODE int(4) DEFAULT NULL,
  TYPE varchar(20) DEFAULT NULL,
  SCODE int(4) DEFAULT NULL,
  Nature varchar(25) DEFAULT NULL,
  UNITS decimal(19,4) DEFAULT NULL,
  BNITS decimal(19,4) DEFAULT NULL,
  DRECD double DEFAULT '0',
  FNO varchar(50) DEFAULT NULL,
  FLAG varchar(5) DEFAULT NULL,
  MBAL double DEFAULT NULL,
  PBAL double DEFAULT NULL,
  MTotalBal double DEFAULT NULL,
  PLNOT decimal(19,4) DEFAULT NULL,
  PLBOOK decimal(19,4) DEFAULT NULL,
  AGE int(4) DEFAULT NULL,
  RETABS decimal(19,4) DEFAULT NULL,
  RETAGR decimal(19,4) DEFAULT NULL,
  INDEX1 decimal(19,4) DEFAULT NULL,
  RETINDEXABS decimal(19,4) DEFAULT NULL,
  RetIndexCAGR decimal(19,4) DEFAULT NULL,
  CURRAMT decimal(19,4) DEFAULT NULL,
  GLOSSLT decimal(19,4) DEFAULT NULL,
  GLOSSST decimal(19,4) DEFAULT NULL,
  UNITSFORDIVID decimal(19,4) DEFAULT NULL,
  factor double DEFAULT NULL,
  LNav double DEFAULT '10',
  Date2 date DEFAULT NULL,
  IType int(4) DEFAULT NULL,
  Rate double DEFAULT NULL,
  CurrAmt double DEFAULT NULL,
  IndexVal double DEFAULT NULL,
  LatestIndexVal double DEFAULT NULL,
  Field int(4) DEFAULT NULL,
  C_Code int(4) DEFAULT NULL,
  B_Code int(4) DEFAULT NULL,
  Rm_Code int(4) DEFAULT NULL,
  Group_Name varchar(100) DEFAULT NULL,
  Type1 varchar(20) DEFAULT NULL,
  Type2 varchar(20) DEFAULT NULL,
  IsOnline tinyint(3) unsigned DEFAULT NULL,
  SFactor double DEFAULT NULL,
  OS_Code int(4) DEFAULT NULL,
  PRIMARY KEY (NO),
  KEY SCODE (SCODE),
  KEY C_Code (C_Code),
  KEY TYPE (TYPE),
  KEY OS_Code (OS_Code),
  KEY LNav (LNav),
  KEY IDX_1 (AGE,Type2),
  KEY DATE1 (DATE1)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci

注意:拥有这么多索引的原因是我们在 SP 中进行了许多选择，这将减少表扫描。

  UPDATE Temp 
      INNER JOIN SchDate ON Temp.Sch_Code = SchDate.Sch_Code
    SET LatestNav = NavRs, NavDate = LDate   ;

-- SchDate 表包含 41K 记录

 UPDATE Temp
    SET Age = DATEDIFF(NAVDATE, TR_DATE),
        CurrAmt = (LatestNav * Units),
        PL_Notional = (UNITS * (LatestNav - Rate)),
        Divd_Recd = 0;

这里是my.cnf供引用

[client]
port=3307
max_execution_time = 0
local_infile = 1

[mysql]
no-beep

[mysqld]

port=3307
#skip-locking
#skip-name-resolve
default_authentication_plugin=mysql_native_password
wait_timeout = 300
interactive_timeout = 300
default-storage-engine=INNODB
sql-mode="NO_ENGINE_SUBSTITUTION,ANSI_QUOTES"
max_execution_time = 0
innodb_autoinc_lock_mode = 0
group_concat_max_len=153600
skip-log-bin
log_bin_trust_function_creators = 1
#expire_logs_days = 3
local_infile = 1
skip-log-bin


### Cache/Buffer Related Parameters ###
table_open_cache=1024000
open_files_limit=2048000
key_buffer_size=2147483648

#myisam_max_sort_file_size=1G
#myisam_sort_buffer_size=512M
#myisam_repair_threads=1



# General and Slow logging.
log-output=FILE
#general-log=0
#general_log_file="E:\Mysql\MySQL Server 8.0\Data\2016SERVER.log"
#slow-query-log=1
#slow_query_log_file="E:\Mysql\MySQL Server 8.0\Data\2016SERVER-slow.log"
long_query_time=100


# Thread Specific Values
sort_buffer_size=2147483648
read_buffer_size=2147483648
read_rnd_buffer_size=1073741824
join_buffer_size=1073741824
thread_cache_size=600
bulk_insert_buffer_size=4294967296

### Mysql Directory & Tables ###
datadir="E:\Mysql\Data\Data\"
tmp_table_size=17179869184

max_heap_table_size=8589934592



### Innodb Related Parameters ###
#innodb_force_recovery=3

## Innodb startup-shutdown related parameter
innodb_max_dirty_pages_pct = 0
innodb_buffer_pool_dump_pct = 100
innodb_buffer_pool_dump_at_shutdown = 1
innodb_buffer_pool_load_at_startup = 1

innodb_change_buffer_max_size = 50
innodb_file_per_table = 1
innodb_log_file_size = 10G
innodb_log_buffer_size = 4294967295
innodb_log_files_in_group = 10
#innodb_buffer_pool_chunk_size = 1024M
innodb_buffer_pool_size = 96636764160
###innodb_buffer_pool_size=90G
innodb_buffer_pool_instances = 50
#innodb_flush_method=O_DIRECT
innodb_flush_log_at_trx_commit = 1
innodb_lock_wait_timeout = 100
innodb_write_io_threads = 64
innodb_read_io_threads = 64

# Binary Logging.
#log-bin="E:\Mysql\Data\Data\2016SERVER-bin"

# Error Logging.
log-error="E:\Mysql\Data\Data\2016SERVER.err"

# Server-Id.
server-id=2

lower_case_table_names=1

# Secure File Priv.
secure-file-priv="E:\Mysql\Uploads"
max_connections=500
#innodb_thread_concurrency=9

innodb_thread_concurrency=0
innodb_adaptive_max_sleep_delay=150000
innodb_autoextend_increment = 2048
#innodb_concurrency_tickets=5000
#innodb_old_blocks_time=1000
innodb_open_files=1500
innodb_stats_on_metadata=0
innodb_checksum_algorithm=0
#back_log=80
#flush_time=0
max_allowed_packet=512M
table_definition_cache=1400
binlog_row_event_max_size=8K
#sync_master_info=10000
#sync_relay_log=10000
#sync_relay_log_info=10000
loose_mysqlx_port=33060

innodb_flush_method = unbuffered
###innodb_flush_method = async_unbuffered
default-time-zone = +05:30
tmpdir = "C:/TEMP"
innodb_io_capacity = 1000
plugin_dir = "C:/Program Files/MySQL/MySQL Server 8/lib/plugin"
innodb_log_write_ahead_size = 16394
mysqlx_max_connections = 500
innodb_random_read_ahead = 1

第一次更新需要 30 到 35 分钟，第二次更新需要 15 分钟。

这里解释一下更新1的计划

1   SIMPLE  SchDate     index   PRIMARY,Sch_Code,IDX_1  Sch_Code    4       39064   100 Using index
1   SIMPLE  temp        ref SCH_Code    SCH_Code    9   SchDate.Sch_Code    1   100 Using index condition

我正在 Windows 10 上运行此查询。有什么办法可以提高 UPDATE 查询的速度吗？任何与配置相关的更改都会有帮助吗？

Answer 1

table_open_cache=1024000

不!那是表，而不是字节。将其更改为 2000。

key_buffer_size=2147483648

假设您使用的是 InnoDB，而不是 MyISAM:

key_buffer_size = 50M
innodb_buffer_pool_size is fine at 96G (for 128GB of RAM)

这几乎没用，改为5:

long_query_time=100

还有...

# Thread Specific Values
sort_buffer_size=2147483648
read_buffer_size=2147483648
read_rnd_buffer_size=1073741824
join_buffer_size=1073741824
bulk_insert_buffer_size=4294967296
tmp_table_size=17179869184
max_heap_table_size=8589934592

阅读该评论! 每个连接可以分配这些大小!你将耗尽内存。交换速度很慢，否则会崩溃!即使您有大量 RAM，这些数字也过多。

innodb_flush_method = unbuffered

手册说:

unbuffered: ... is used for internal performance testing and is currently unsupported. Use at your own risk.

innodb_random_read_ahead = 1

来自手册:

Because this feature can improve performance in some cases and reduce performance in others, before relying on this setting, benchmark both with and without the setting enabled.

底线:撤消除 buffer_pool 之外的所有配置更改。