MySQL MGR运维

1. 历史

 5.7.17+以后，出现的组复制技术。MySQL Group Replication。
 为了复制环境中，数据一致性考虑、多写的场景研发的技术。

2. 架构原理

 在2N+1个节点组成的单主模式组复制集群中，主库上一个事务提交时，会将事务修改记录相关的信息和
 事务产生的BINLOG事件打包生成一个写集(WRITE SET)，将写集发送给所有节点，并通过至少N个节
 点投票通过才能事务提交成功。
 ​
 在事务执行期间，会将：
 1、事务操作生成的map event/query event/dml event等写入BINLOG CACHE中(内存)
 2、将Write Set写入到Rpl_transaction_write_set_ctx中(内存)
 在事务提交时，具体在MYSQL_BIN_LOG::prepare之后，但是在
 MYSQL_BIN_LOG::ordered_commit之前，即事务相关的BINLOG Event还在BINLOG CACHE没有
 写入到BINLOG FILE前，将BINLOG CACHE中和Rpl_transaction_write_set_ctx中的数据进行
 处理并写入到transaction_msg中，由gcs_module负责发送transaction_msg到各个节点，等待
 各节点进行事务认证。
 由于transaction_msg中包含BINLOG信息，并在事务认证期间发送给MGR各节点，因此无需等待主节
 点的BINLOG落盘后再发送给备用节点。
 每个MGR群集中的节点上，都存在IO线程和SQL线程，IO线程会解析transaction_msg获取到BINLOG
 EVENT并保存到RELAY LOG中，再由SQL线程执行重放到辅助节点上。
 ​
 总结：
 single-primary:
 begin;
 update t1 set name=‘zs’ where id=10;
 commit;
 1. 主库发生新的事务，事务执行期间，会将事务binlog 刷到binlog cache。
 2. 生成write_set ,包含binlog日志+表、库hash值。写入
 Rpl_transaction_write_set_ctx。
 3. MYSQL_BIN_LOG::prepare之后，在MYSQL_BIN_LOG::ordered_commit之前
 通过gcs_module将所有的Rpl_transaction_write_set_ctx，以transaction_msg传输
 给各个节点。
 4. 通过certify验证（Paxos）,通过投票机制，判断此次事务是否满足半数以上节点通过（不
 冲突）。
 5. 主库binlog flush && sync disk
 6. 此时，各个从库开始回放relaylog。

3.安装搭建

3.1 主节点

 [client]
 port    = 3306
 socket    = /data/mysql/mysql_3306/tmp/mysql_3306.sock
 [mysql]
 prompt="\u@\h \R:\m:\s [\d]> "
 no-auto-rehash
 socket    = /data/mysql/mysql_3306/tmp/mysql_3306.sock
 ​
 [mysqld]
 user    = mysql
 port    = 3306
 admin_address = 127.0.0.1
 basedir    = /usr/local/mysql
 datadir    = /data/mysql/mysql_3306/data
 socket    = /data/mysql/mysql_3306/tmp/mysql_3306.sock
 pid-file = mysql_3306.pid
 character-set-server = utf8mb4
 secure-file-priv=/tmp
 skip_name_resolve = 1
 lower_case_table_names = 1
 report_host=10.0.0.57
 report_port=3306
 default_authentication_plugin=mysql_native_password
 binlog_checksum=NONE
 binlog_transaction_dependency_tracking = WRITESET
 transaction_write_set_extraction=XXHASH64
 loose-group_replication_group_name="ca842376-1c50-42ac-bb57-a5adc7da7a12"
 loose-group_replication_start_on_boot=OFF
 loose-group_replication_local_address= "10.0.0.57:33067"
 loose-group_replication_group_seeds="10.0.0.51:33067,10.0.0.52:33068,10.0.0.53:33069"
 loose-group_replication_bootstrap_group=OFF
 ​
 #replicate-wild-ignore-table=mysql.%
 replicate-wild-ignore-table=test.%
 replicate-wild-ignore-table=information_schema.%
 ​
 # Two-Master configure
 #server-1 
 #auto-increment-offset = 1
 #auto-increment-increment = 2 
 ​
 #server-2                          
 #auto-increment-offset = 2
 #auto-increment-increment = 2
 ​
 # semi sync replication settings #
 #plugin_dir = /usr/local/mysql/lib/mysql/plugin
 #plugin_load = "validate_password.so;rpl_semi_sync_master=semisync_master.so;rpl_semi_sync_slave=semisync_slave.so"
 plugin_dir = /usr/local/mysql/lib/plugin #官方版本的路径
 plugin_load = "rpl_semi_sync_master=semisync_master.so;rpl_semi_sync_slave=semisync_slave.so" #官方版本的路径
 ​
 slave_parallel_workers = 4
 slave_parallel_type = LOGICAL_CLOCK
 ​
 open_files_limit    = 65535
 back_log = 1024
 max_connections = 1024
 max_connect_errors = 1000000
 table_open_cache = 1024
 table_definition_cache = 1024
 table_open_cache_instances = 64
 thread_stack = 512K
 external-locking = FALSE
 max_allowed_packet = 32M
 sort_buffer_size = 4M
 join_buffer_size = 4M
 thread_cache_size = 1536
 interactive_timeout = 600
 wait_timeout = 600
 tmp_table_size = 32M
 max_heap_table_size = 32M
 slow_query_log = 1
 log_timestamps = SYSTEM
 slow_query_log_file = /data/mysql/mysql_3306/logs/slow.log
 log-error = /data/mysql/mysql_3306/logs/error.log
 long_query_time = 0.1
 log_queries_not_using_indexes =1
 log_throttle_queries_not_using_indexes = 60
 min_examined_row_limit = 100
 log_slow_admin_statements = 1
 log_slow_slave_statements = 1
 server-id = 1
 log-bin = /data/mysql/mysql_3306/logs/mysql-bin
 sync_binlog = 1
 binlog_cache_size = 4M
 max_binlog_cache_size = 2G
 max_binlog_size = 1G
 binlog_expire_logs_seconds=2592000 
 master_info_repository = TABLE
 relay_log_info_repository = TABLE
 gtid_mode = on
 enforce_gtid_consistency = 1
 log_slave_updates
 slave-rows-search-algorithms = 'INDEX_SCAN,HASH_SCAN'
 binlog_format = ROW
 binlog_row_image=FULL
 binlog_checksum = 1
 relay_log_recovery = 1
 relay-log-purge = 1
 key_buffer_size = 32M
 read_buffer_size = 8M
 read_rnd_buffer_size = 4M
 bulk_insert_buffer_size = 64M
 myisam_sort_buffer_size = 128M
 myisam_max_sort_file_size = 10G
 #myisam_repair_threads = 1
 lock_wait_timeout = 3600
 explicit_defaults_for_timestamp = 1
 innodb_thread_concurrency = 0
 innodb_sync_spin_loops = 100
 innodb_spin_wait_delay = 30
 ​
 #transaction_isolation = REPEATABLE-READ
 transaction_isolation = READ-COMMITTED
 #innodb_additional_mem_pool_size = 16M
 innodb_buffer_pool_size = 2867M
 innodb_buffer_pool_instances = 4
 innodb_buffer_pool_load_at_startup = 1
 innodb_buffer_pool_dump_at_shutdown = 1
 innodb_data_file_path = ibdata1:1G:autoextend
 innodb_flush_log_at_trx_commit = 1
 innodb_log_buffer_size = 32M
 innodb_log_file_size = 2G
 innodb_log_files_in_group = 3
 innodb_max_undo_log_size = 4G
 innodb_undo_directory = /data/mysql/mysql_3306/undolog
 innodb_undo_tablespaces = 95
 ​
 # 根据您的服务器IOPS能力适当调整
 # 一般配普通SSD盘的话，可以调整到 10000 - 20000
 # 配置高端PCIe SSD卡的话，则可以调整的更高，比如 50000 - 80000
 innodb_io_capacity = 4000
 innodb_io_capacity_max = 8000
 innodb_flush_sync = 0
 innodb_flush_neighbors = 0
 innodb_write_io_threads = 8
 innodb_read_io_threads = 8
 innodb_purge_threads = 4
 innodb_page_cleaners = 4
 innodb_open_files = 65535
 innodb_max_dirty_pages_pct = 50
 innodb_flush_method = O_DIRECT
 innodb_lru_scan_depth = 4000
 innodb_checksum_algorithm = crc32
 innodb_lock_wait_timeout = 10
 innodb_rollback_on_timeout = 1
 innodb_print_all_deadlocks = 1
 innodb_file_per_table = 1
 innodb_online_alter_log_max_size = 4G
 innodb_stats_on_metadata = 0
 ​
 # some var for MySQL 8
 log_error_verbosity = 3
 innodb_print_ddl_logs = 1
 binlog_expire_logs_seconds = 2592000
 #innodb_dedicated_server = 0
 ​
 innodb_status_file = 1
 # 注意: 开启 innodb_status_output & innodb_status_output_locks 后, 可能会导致log-error文件增长较快
 innodb_status_output = 0
 innodb_status_output_locks = 0
 ​
 #performance_schema
 performance_schema = 1
 performance_schema_instrument = '%memory%=on'
 performance_schema_instrument = '%lock%=on'
 ​
 #innodb monitor
 innodb_monitor_enable="module_innodb"
 innodb_monitor_enable="module_server"
 innodb_monitor_enable="module_dml"
 innodb_monitor_enable="module_ddl"
 innodb_monitor_enable="module_trx"
 innodb_monitor_enable="module_os"
 innodb_monitor_enable="module_purge"
 innodb_monitor_enable="module_log"
 innodb_monitor_enable="module_lock"
 innodb_monitor_enable="module_buffer"
 innodb_monitor_enable="module_index"
 innodb_monitor_enable="module_ibuf_system"
 innodb_monitor_enable="module_buffer_page"
 innodb_monitor_enable="module_adaptive_hash"
 ​
 [mysqldump]
 quick
 max_allowed_packet = 32M
 ​
 [mysqld_safe]
 #malloc-lib=/usr/local/mysql/lib/jmalloc.so 
 nice=-19
 open-files-limit=65535
 ​

3.2 从1

 [client]
 port    = 3306
 socket    = /data/mysql/mysql_3306/tmp/mysql_3306.sock
 [mysql]
 prompt="\u@\h \R:\m:\s [\d]> "
 no-auto-rehash
 socket    = /data/mysql/mysql_3306/tmp/mysql_3306.sock
 ​
 [mysqld]
 user    = mysql
 port    = 3306
 admin_address = 127.0.0.1
 basedir    = /usr/local/mysql
 datadir    = /data/mysql/mysql_3306/data
 socket    = /data/mysql/mysql_3306/tmp/mysql_3306.sock
 pid-file = mysql_3306.pid
 character-set-server = utf8mb4
 skip_name_resolve = 1
 lower_case_table_names = 1
 secure-file-priv=/tmp
 report_host=10.0.0.58
 report_port=3306
 default_authentication_plugin=mysql_native_password
 binlog_checksum=NONE
 binlog_transaction_dependency_tracking = WRITESET
 transaction_write_set_extraction=XXHASH64
 loose-group_replication_group_name="ca842376-1c50-42ac-bb57-a5adc7da7a12"
 loose-group_replication_start_on_boot=OFF
 loose-group_replication_local_address= "10.0.0.58:33068"
 loose-group_replication_group_seeds="10.0.0.57:33067,10.0.0.58:33068,10.0.0.59:33069"
 loose-group_replication_bootstrap_group=OFF
 ​
 #replicate-wild-ignore-table=mysql.%
 replicate-wild-ignore-table=test.%
 replicate-wild-ignore-table=information_schema.%
 ​
 # Two-Master configure
 #server-1 
 #auto-increment-offset = 1
 #auto-increment-increment = 2 
 ​
 #server-2                          
 #auto-increment-offset = 2
 #auto-increment-increment = 2
 ​
 # semi sync replication settings #
 #plugin_dir = /usr/local/mysql/lib/mysql/plugin
 #plugin_load = "validate_password.so;rpl_semi_sync_master=semisync_master.so;rpl_semi_sync_slave=semisync_slave.so"
 plugin_dir = /usr/local/mysql/lib/plugin #官方版本的路径
 plugin_load = "rpl_semi_sync_master=semisync_master.so;rpl_semi_sync_slave=semisync_slave.so" #官方版本的路径
 ​
 slave_parallel_workers = 4
 slave_parallel_type = LOGICAL_CLOCK
 ​
 open_files_limit    = 65535
 back_log = 1024
 max_connections = 1024
 max_connect_errors = 1000000
 table_open_cache = 1024
 table_definition_cache = 1024
 table_open_cache_instances = 64
 thread_stack = 512K
 external-locking = FALSE
 max_allowed_packet = 32M
 sort_buffer_size = 4M
 join_buffer_size = 4M
 thread_cache_size = 1536
 interactive_timeout = 600
 wait_timeout = 600
 tmp_table_size = 32M
 max_heap_table_size = 32M
 slow_query_log = 1
 log_timestamps = SYSTEM
 slow_query_log_file = /data/mysql/mysql_3306/logs/slow.log
 log-error = /data/mysql/mysql_3306/logs/error.log
 long_query_time = 0.1
 log_queries_not_using_indexes =1
 log_throttle_queries_not_using_indexes = 60
 min_examined_row_limit = 100
 log_slow_admin_statements = 1
 log_slow_slave_statements = 1
 server-id = 2
 log-bin = /data/mysql/mysql_3306/logs/mysql-bin
 sync_binlog = 1
 binlog_cache_size = 4M
 max_binlog_cache_size = 2G
 max_binlog_size = 1G
 binlog_expire_logs_seconds=2592000 
 master_info_repository = TABLE
 relay_log_info_repository = TABLE
 gtid_mode = on
 enforce_gtid_consistency = 1
 log_slave_updates
 slave-rows-search-algorithms = 'INDEX_SCAN,HASH_SCAN'
 binlog_format = ROW
 binlog_row_image=FULL
 binlog_checksum = 1
 relay_log_recovery = 1
 relay-log-purge = 1
 key_buffer_size = 32M
 read_buffer_size = 8M
 read_rnd_buffer_size = 4M
 bulk_insert_buffer_size = 64M
 myisam_sort_buffer_size = 128M
 myisam_max_sort_file_size = 10G
 #myisam_repair_threads = 1
 lock_wait_timeout = 3600
 explicit_defaults_for_timestamp = 1
 innodb_thread_concurrency = 0
 innodb_sync_spin_loops = 100
 innodb_spin_wait_delay = 30
 ​
 #transaction_isolation = REPEATABLE-READ
 transaction_isolation = READ-COMMITTED
 #innodb_additional_mem_pool_size = 16M
 innodb_buffer_pool_size = 2867M
 innodb_buffer_pool_instances = 4
 innodb_buffer_pool_load_at_startup = 1
 innodb_buffer_pool_dump_at_shutdown = 1
 innodb_data_file_path = ibdata1:1G:autoextend
 innodb_flush_log_at_trx_commit = 1
 innodb_log_buffer_size = 32M
 innodb_log_file_size = 2G
 innodb_log_files_in_group = 3
 innodb_max_undo_log_size = 4G
 innodb_undo_directory = /data/mysql/mysql_3306/undolog
 innodb_undo_tablespaces = 95
 ​
 # 根据您的服务器IOPS能力适当调整
 # 一般配普通SSD盘的话，可以调整到 10000 - 20000
 # 配置高端PCIe SSD卡的话，则可以调整的更高，比如 50000 - 80000
 innodb_io_capacity = 4000
 innodb_io_capacity_max = 8000
 innodb_flush_sync = 0
 innodb_flush_neighbors = 0
 innodb_write_io_threads = 8
 innodb_read_io_threads = 8
 innodb_purge_threads = 4
 innodb_page_cleaners = 4
 innodb_open_files = 65535
 innodb_max_dirty_pages_pct = 50
 innodb_flush_method = O_DIRECT
 innodb_lru_scan_depth = 4000
 innodb_checksum_algorithm = crc32
 innodb_lock_wait_timeout = 10
 innodb_rollback_on_timeout = 1
 innodb_print_all_deadlocks = 1
 innodb_file_per_table = 1
 innodb_online_alter_log_max_size = 4G
 innodb_stats_on_metadata = 0
 ​
 # some var for MySQL 8
 log_error_verbosity = 3
 innodb_print_ddl_logs = 1
 binlog_expire_logs_seconds = 2592000
 #innodb_dedicated_server = 0
 ​
 innodb_status_file = 1
 # 注意: 开启 innodb_status_output & innodb_status_output_locks 后, 可能会导致log-error文件增长较快
 innodb_status_output = 0
 innodb_status_output_locks = 0
 ​
 #performance_schema
 performance_schema = 1
 performance_schema_instrument = '%memory%=on'
 performance_schema_instrument = '%lock%=on'
 ​
 #innodb monitor
 innodb_monitor_enable="module_innodb"
 innodb_monitor_enable="module_server"
 innodb_monitor_enable="module_dml"
 innodb_monitor_enable="module_ddl"
 innodb_monitor_enable="module_trx"
 innodb_monitor_enable="module_os"
 innodb_monitor_enable="module_purge"
 innodb_monitor_enable="module_log"
 innodb_monitor_enable="module_lock"
 innodb_monitor_enable="module_buffer"
 innodb_monitor_enable="module_index"
 innodb_monitor_enable="module_ibuf_system"
 innodb_monitor_enable="module_buffer_page"
 innodb_monitor_enable="module_adaptive_hash"
 ​
 [mysqldump]
 quick
 max_allowed_packet = 32M
 ​
 [mysqld_safe]
 #malloc-lib=/usr/local/mysql/lib/jmalloc.so 
 nice=-19
 open-files-limit=65535

3.3 从2

 [client]
 port    = 3306
 socket    = /data/mysql/mysql_3306/tmp/mysql_3306.sock
 [mysql]
 prompt="\u@\h \R:\m:\s [\d]> "
 no-auto-rehash
 socket    = /data/mysql/mysql_3306/tmp/mysql_3306.sock
 ​
 ​
 [mysqld]
 user    = mysql
 port    = 3306
 admin_address = 127.0.0.1
 basedir    = /usr/local/mysql
 datadir    = /data/mysql/mysql_3306/data
 socket    = /data/mysql/mysql_3306/tmp/mysql_3306.sock
 pid-file = mysql_3306.pid
 character-set-server = utf8mb4
 skip_name_resolve = 1
 lower_case_table_names = 1
 secure-file-priv=/tmp
 report_host=10.0.0.59
 report_port=3306
 default_authentication_plugin=mysql_native_password
 binlog_checksum=NONE
 binlog_transaction_dependency_tracking = WRITESET
 transaction_write_set_extraction=XXHASH64
 loose-group_replication_group_name="ca842376-1c50-42ac-bb57-a5adc7da7a12"
 loose-group_replication_start_on_boot=OFF
 loose-group_replication_local_address= "10.0.0.59:33069"
 loose-group_replication_group_seeds="10.0.0.57:33067,10.0.0.58:33068,10.0.0.59:33069"
 loose-group_replication_bootstrap_group=OFF
 ​
 ​
 #replicate-wild-ignore-table=mysql.%
 replicate-wild-ignore-table=test.%
 replicate-wild-ignore-table=information_schema.%
 ​
 # Two-Master configure
 #server-1 
 #auto-increment-offset = 1
 #auto-increment-increment = 2 
 ​
 #server-2                          
 #auto-increment-offset = 2
 #auto-increment-increment = 2
 ​
 # semi sync replication settings #
 #plugin_dir = /usr/local/mysql/lib/mysql/plugin
 #plugin_load = "validate_password.so;rpl_semi_sync_master=semisync_master.so;rpl_semi_sync_slave=semisync_slave.so"
 plugin_dir = /usr/local/mysql/lib/plugin #官方版本的路径
 plugin_load = "rpl_semi_sync_master=semisync_master.so;rpl_semi_sync_slave=semisync_slave.so" #官方版本的路径
 ​
 slave_parallel_workers = 4
 slave_parallel_type = LOGICAL_CLOCK
 ​
 open_files_limit    = 65535
 back_log = 1024
 max_connections = 1024
 max_connect_errors = 1000000
 table_open_cache = 1024
 table_definition_cache = 1024
 table_open_cache_instances = 64
 thread_stack = 512K
 external-locking = FALSE
 max_allowed_packet = 32M
 sort_buffer_size = 4M
 join_buffer_size = 4M
 thread_cache_size = 1536
 interactive_timeout = 600
 wait_timeout = 600
 tmp_table_size = 32M
 max_heap_table_size = 32M
 slow_query_log = 1
 log_timestamps = SYSTEM
 slow_query_log_file = /data/mysql/mysql_3306/logs/slow.log
 log-error = /data/mysql/mysql_3306/logs/error.log
 long_query_time = 0.1
 log_queries_not_using_indexes =1
 log_throttle_queries_not_using_indexes = 60
 min_examined_row_limit = 100
 log_slow_admin_statements = 1
 log_slow_slave_statements = 1
 server-id = 3306
 log-bin = /data/mysql/mysql_3306/logs/mysql-bin
 sync_binlog = 1
 binlog_cache_size = 4M
 max_binlog_cache_size = 2G
 max_binlog_size = 1G
 binlog_expire_logs_seconds=2592000 
 master_info_repository = TABLE
 relay_log_info_repository = TABLE
 gtid_mode = on
 enforce_gtid_consistency = 1
 log_slave_updates
 slave-rows-search-algorithms = 'INDEX_SCAN,HASH_SCAN'
 binlog_format = ROW
 binlog_row_image=FULL
 binlog_checksum = 1
 relay_log_recovery = 1
 relay-log-purge = 1
 key_buffer_size = 32M
 read_buffer_size = 8M
 read_rnd_buffer_size = 4M
 bulk_insert_buffer_size = 64M
 myisam_sort_buffer_size = 128M
 myisam_max_sort_file_size = 10G
 #myisam_repair_threads = 1
 lock_wait_timeout = 3600
 explicit_defaults_for_timestamp = 1
 innodb_thread_concurrency = 0
 innodb_sync_spin_loops = 100
 innodb_spin_wait_delay = 30
 ​
 #transaction_isolation = REPEATABLE-READ
 transaction_isolation = READ-COMMITTED
 #innodb_additional_mem_pool_size = 16M
 innodb_buffer_pool_size = 2867M
 innodb_buffer_pool_instances = 4
 innodb_buffer_pool_load_at_startup = 1
 innodb_buffer_pool_dump_at_shutdown = 1
 innodb_data_file_path = ibdata1:1G:autoextend
 innodb_flush_log_at_trx_commit = 1
 innodb_log_buffer_size = 32M
 innodb_log_file_size = 2G
 innodb_log_files_in_group = 3
 innodb_max_undo_log_size = 4G
 innodb_undo_directory = /data/mysql/mysql_3306/undolog
 innodb_undo_tablespaces = 95
 ​
 # 根据您的服务器IOPS能力适当调整
 # 一般配普通SSD盘的话，可以调整到 10000 - 20000
 # 配置高端PCIe SSD卡的话，则可以调整的更高，比如 50000 - 80000
 innodb_io_capacity = 4000
 innodb_io_capacity_max = 8000
 innodb_flush_sync = 0
 innodb_flush_neighbors = 0
 innodb_write_io_threads = 8
 innodb_read_io_threads = 8
 innodb_purge_threads = 4
 innodb_page_cleaners = 4
 innodb_open_files = 65535
 innodb_max_dirty_pages_pct = 50
 innodb_flush_method = O_DIRECT
 innodb_lru_scan_depth = 4000
 innodb_checksum_algorithm = crc32
 innodb_lock_wait_timeout = 10
 innodb_rollback_on_timeout = 1
 innodb_print_all_deadlocks = 1
 innodb_file_per_table = 1
 innodb_online_alter_log_max_size = 4G
 innodb_stats_on_metadata = 0
 ​
 # some var for MySQL 8
 log_error_verbosity = 3
 innodb_print_ddl_logs = 1
 binlog_expire_logs_seconds = 2592000
 #innodb_dedicated_server = 0
 ​
 innodb_status_file = 1
 # 注意: 开启 innodb_status_output & innodb_status_output_locks 后, 可能会导致log-error文件增长较快
 innodb_status_output = 0
 innodb_status_output_locks = 0
 ​
 #performance_schema
 performance_schema = 1
 performance_schema_instrument = '%memory%=on'
 performance_schema_instrument = '%lock%=on'
 ​
 #innodb monitor
 innodb_monitor_enable="module_innodb"
 innodb_monitor_enable="module_server"
 innodb_monitor_enable="module_dml"
 innodb_monitor_enable="module_ddl"
 innodb_monitor_enable="module_trx"
 innodb_monitor_enable="module_os"
 innodb_monitor_enable="module_purge"
 innodb_monitor_enable="module_log"
 innodb_monitor_enable="module_lock"
 innodb_monitor_enable="module_buffer"
 innodb_monitor_enable="module_index"
 innodb_monitor_enable="module_ibuf_system"
 innodb_monitor_enable="module_buffer_page"
 innodb_monitor_enable="module_adaptive_hash"
 ​
 [mysqldump]
 quick
 max_allowed_packet = 32M
 ​
 [mysqld_safe]
 #malloc-lib=/usr/local/mysql/lib/jmalloc.so 
 nice=-19
 open-files-limit=65535

3.3 实现单主模式

 初始化
 /usr/local/mysql/bin/mysqld --defaults-file=/data/mysql/mysql_3306/my_3306.cnf --initialize-insecure --lower-case-table-names=1 --user=mysql
 ​
 启动
 /usr/local/mysql/bin/mysqld_safe --defaults-file=/data/mysql/mysql_3306/my_3306.cnf &
 ​
 ​
 设置本地root用户密码和密码插件（所有节点）
 mysql> ALTER USER 'root'@'localhost' IDENTIFIED
     -> WITH mysql_native_password BY '123';
     
     
 安装MGR插件（所有节点）
 INSTALL PLUGIN   group_replication SONAME 'group_replication.so';
 ​
 ​
 设置复制账号(所有节点)
 SET SQL_LOG_BIN=0;
 CREATE USER repl@'%' IDENTIFIED BY '123';
 CREATE USER repl@'localhost' IDENTIFIED BY '123';
 CREATE USER repl@'127.0.0.1' IDENTIFIED BY '123';
 GRANT REPLICATION SLAVE,replication client ON *.* TO repl@'%';
 grant replication slave,replication client on *.* to repl@'localhost' ;
 grant replication slave,replication client on *.* to repl@'127.0.0.1' ;
 FLUSH PRIVILEGES;
 SET SQL_LOG_BIN=1;
 ​
 ​
 ​
 j. 启动MGR单主模式
 # 启动MGR，在主库(10.0.0.57)上执行
 ​
 CHANGE MASTER TO MASTER_USER='repl', MASTER_PASSWORD='123' FOR CHANNEL
 'group_replication_recovery';
 SET GLOBAL group_replication_bootstrap_group=ON;
 START GROUP_REPLICATION;
 SET GLOBAL group_replication_bootstrap_group=OFF;
 SELECT * FROM performance_schema.replication_group_members;
 ​
 ​
 ​
 其他节点加入MGR，在从库上执行
 reset master;
 CHANGE MASTER TO MASTER_USER='repl', MASTER_PASSWORD='123' FOR CHANNEL 'group_replication_recovery';
 START GROUP_REPLICATION;
 ​
 查看MGR组信息（主库查看）
 SELECT * FROM performance_schema.replication_group_members;
 ​
 ​
 可以看到，3个节点状态为online，并且主节点为192.168.56.101，只有主节点可以写入，其
 他节点只读，MGR单主模式搭建成功。
 ​
 ​
 ​
 如果需要重置，那么需要执行如下命令：
 STOP GROUP_REPLICATION;
 reset master;
 SET SQL_LOG_BIN=1;
 CHANGE MASTER TO MASTER_USER='repl', MASTER_PASSWORD='123' FOR CHANNEL  'group_replication_recovery';
 start GROUP_REPLICATION；

3.4 切换到多主模式

 MGR切换模式需要重新启动组复制，因些需要在所有节点上先关闭组复制，设置
 group_replication_single_primary_mode=OFF 等参数，再启动组复制。
 ​
 1、该模式启用需设置两个参数
 group_replication_single_primary_mode=0 #这个参数很好理解，就是关闭单master模式
 group_replication_enforce_update_everywhere_checks=1 #这个参数设置多主模式下各
 个节点严格一致性检查
 ================
 db02 [(none)]>stop GROUP_REPLICATION;
 db02 [(none)]>set global group_replication_single_primary_mode=OFF;
 db02 [(none)]>set global  group_replication_enforce_update_everywhere_checks=1;
 db03 [(none)]>stop GROUP_REPLICATION;
 db03 [(none)]>set global group_replication_single_primary_mode=OFF;
 db03 [(none)]>set global  group_replication_enforce_update_everywhere_checks=1;
 select    @@group_replication_single_primary_mode,@@group_replication_enforce_update_everywhere_checks;
 ===============
 ​
 2、 默认启动的都是单master模式，其他节点都设置了read_only、super_read_only这两个参
 数，需要修改这两个配置
 ===========
 db03 [(none)]>set global read_only=0;
 db03 [(none)]>set global super_read_only=0;
 ===========
 ​
 3、 完成上面的配置后就可以执行多点写入了，多点写入会存在冲突检查，这耗损性能挺大的，官方
 建议采用网络分区功能，在程序端把相同的业务定位到同一节点，尽量减少冲突发生几率。
 # 停止组复制(所有节点执行)：
 stop group_replication;
 set global group_replication_single_primary_mode=OFF;
 set global group_replication_enforce_update_everywhere_checks=ON;
 # 随便选择某个节点执行
 SET GLOBAL group_replication_bootstrap_group=ON;
 START GROUP_REPLICATION;
 SET GLOBAL group_replication_bootstrap_group=OFF;
 # 其他节点执行
 START GROUP_REPLICATION;
 # 查看组信息，所有节点的 MEMBER_ROLE 都为 PRIMARY，
 SELECT * FROM performance_schema.replication_group_members;
 ​
 ​
 ​
 ​
 切回单主模式
 # 所有节点执行
 stop group_replication;
 set global group_replication_enforce_update_everywhere_checks=OFF;
 set global group_replication_single_primary_mode=ON;
 ​
 # 主节点执行
 SET GLOBAL group_replication_bootstrap_group=ON;
 START GROUP_REPLICATION;
 SET GLOBAL group_replication_bootstrap_group=OFF;
 # 从节点执行
 START GROUP_REPLICATION;
 # 查看MGR组信息，SELECT * FROM performance_schema.replication_group_members;

4.运维

4.1 MGR的监控

 SELECT * FROM performance_schema.replication_group_members;

4.2 MGR故障处理

1. 宕掉主节点测试。-->自动切换，类似、MHA ，切换过程中所有节点为只读
db02 [(none)]>SELECT * FROM performance_schema.replication_group_members;
+---------------------------+--------------------------------------+-------------+-------------+--------------+-------------+----------------+----------------------------+
| CHANNEL_NAME              | MEMBER_ID                            | MEMBER_HOST | MEMBER_PORT | MEMBER_STATE | MEMBER_ROLE | MEMBER_VERSION | MEMBER_COMMUNICATION_STACK |
+---------------------------+--------------------------------------+-------------+-------------+--------------+-------------+----------------+----------------------------+
| group_replication_applier | 0a10985b-4183-11ee-a843-000c2917e898 | 10.0.0.58   |        3306 | ONLINE       | PRIMARY     | 8.0.30         | XCom                       |
| group_replication_applier | fe9e2ab7-4183-11ee-a8d7-000c29f7dccf | 10.0.0.59   |        3306 | ONLINE       | SECONDARY   | 8.0.30         | XCom                       |
+---------------------------+--------------------------------------+-------------+-------------+--------------+-------------+----------------+----------------------------+
2 rows in set (0.00 sec)


2. 启动主节点-->不会自动加入到集群中
db02 [(none)]>SELECT * FROM performance_schema.replication_group_members;
+---------------------------+-----------+-------------+-------------+--------------+-------------+----------------+----------------------------+
| CHANNEL_NAME              | MEMBER_ID | MEMBER_HOST | MEMBER_PORT | MEMBER_STATE | MEMBER_ROLE | MEMBER_VERSION | MEMBER_COMMUNICATION_STACK |
+---------------------------+-----------+-------------+-------------+--------------+-------------+----------------+----------------------------+
| group_replication_applier |           |             |        NULL | OFFLINE      |             |                |                            |
+---------------------------+-----------+-------------+-------------+--------------+-------------+----------------+----------------------------+
1 row in set (0.00 sec)




3. 宕机后添加到集群
3.1 clone plugin 克隆目前主库数据
主库：
[root@db01 ~]# mysql -uroot -p123  -S  /data/mysql/mysql_3306/tmp/mysql_3306.sock -e "INSTALL PLUGIN clone SONAME 'mysql_clone.so';create user test@'%' identified by '123';grant backup_admin on *.* to test@'%';"
mysql: [Warning] Using a password on the command line interface can be insecure


宕掉的从库
mysql -uroot -p123 -S  /data/mysql/mysql_3306/tmp/mysql_3306.sock -e "INSTALL PLUGIN clone  SONAME 'mysql_clone.so';create user test1@'%' identified by '123';grant  clone_admin on *.* to test1@'%';SET GLOBAL  clone_valid_donor_list='10.0.0.58:3306';"

mysql -utest1 -p123 -h10.0.0.57 -P3306 -e "CLONE INSTANCE FROM test@'10.0.0.58':3306 IDENTIFIED BY '123';"



宕机的库加入集群
CHANGE MASTER TO MASTER_USER='repl', MASTER_PASSWORD='123' FOR CHANNEL 'group_replication_recovery';
START GROUP_REPLICATION;

4.3 MGR的限制

1. 仅支持innodb存储引擎
MGR集群中，只支持innodb存储引擎，能够创建非innodb引擎的表，但是无法写入数据，向非innodb表写数据直接报错。
2. 表必须有主键，或者非Null的唯一键
MGR集群中，只支持innodb引擎的表，并且该表必须有显式的主键，或者非Null的唯一键，否则即使能够创建表，也无法向表中写入数据。
3. 网络限制
MGR 组通信引擎目前仅支持IPv4网络，并且对节点间的网络性能要求较高，低延迟、高带宽的网络是
部署MGR集群的基础。
4. MGR忽略表锁和命名锁，在MGR中lock tables、unlock tables、get_lock、release_lock等这些表锁和命名锁将被忽略。
5. MGR多主模式中，默认不支持 SERIALIZABLE 隔离级别。建议使用RC。
6. 多主模式下，对同一个对象进行并发的有冲突的ddl和dml操作导致这种冲突在部分成员节点中无法检测到，最终可能导致数据不一致。
7. 多主模式下，不支持级联约束的外键，可能造成有冲突的操作无法检测。
8. 不支持超大事务。
9. 多主模式下可能导致死锁，比如select ...for update在不同节点执行，由于多节点锁无法共享，很容易导致死锁。
10.不支持复制过滤，如果有节点设置了复制过滤，将影响节点间决议的达成。
11. MGR最多支持9个节点，大于9个节点，将拒绝新节点的加入。


MGR在8.0版本读写一致性保证
MGR相对于半同步复制，在relay log前增加了冲突检查协调，但是binlog回放仍然可能延时，也就是跟我们熟悉的半同步复制存在SQL线程的回放延迟情况类似。当然关于SQL线程回放慢的原因，跟半同步
也类似，比如大事务！！
所以MGR并不是全同步方案，关于如何处理一致性读写的问题，MySQL 在8.0.14版本中加入了“读写一致性”特性，并引入了参数：group_replication_consistency，下面将对读写一致性的相关参数及
不同应用场景进行详细说明。


MGR读写一致性的优缺点
官方引入的MGR读写一致性既有它自身的天然优势，也不可避免的存在相应的不足，其优缺点如下：
优点：MGR配合中间件，比如读写分离功能的中间件，在MGR单主模式下，可以根据业务场景进行读写分离，不用担心会产生延迟，充分利用了MGR主节点以外的节点。
缺点：使用读写一致性会对性能有极大影响，尤其是网络环境不稳定的场景下。
在实际应用中需要大家因地制宜，根据实际情况选择最适配的方案


MGR读写一致性的方案
针对不同应用场景应当如何选择MGR读写一致性的相关方式，官方提供了几个参数以及与其相对应的应
用场景：

AFTER
适用场景1：写少读多的场景进行读写分离，担心读取到过期事务，可选择AFTER。
适用场景2：只读为主的集群，有RW的事务需要保证提交的事务能被其他后序事务读到最新读数据，可
选择AFTER。

BEFORE
适用场景1：应用大量写入数据，偶尔进行读取一致性数据，应当选择BEFORE。
适用场景2：有特定事务需要读写一致性，以便对敏感数据操作时，始终读取最新的数据；应当选择
BEFORE。

BEFORE_AND_AFTER
适用场景：有一个读为主的集群，有RW的事务既要保证读到最新的数据，又要保证这个事务提交后，被
其他后序事务读到；在这种情况下可选择BEFORE_AND_AFTER。

4.4 故障案例

MGR环境下要求每个表都需要有主键，没有主键就会报错。
MGR 在生成binlog前，先去从节点做校验，能不能执行，或者提交
binlog 刷盘之前，对别的节点进行校验