背景:
MySQL 5.5开始新增一个数据库:PERFORMANCE_SCHEMA,主要用于收集数据库服务器性能参数。并且库里表的存储引擎均为PERFORMANCE_SCHEMA,而用户是不能创建存储引擎为PERFORMANCE_SCHEMA的表。MySQL5.5默认是关闭的,需要手动开启,在配置文件里添加:
[mysqld] performance_schema=ON
查看是否开启:
mysql>show variables like 'performance_schema'; +--------------------+-------+ | Variable_name | Value | +--------------------+-------+ | performance_schema | ON | +--------------------+-------+
从MySQL5.6开始,默认打开,本文就从MySQL5.6来说明,在数据库使用当中PERFORMANCE_SCHEMA的一些比较常用的功能。具体的信息可以查看官方文档。
相关表信息:
一:配置(setup)表:
zjy@performance_schema 10:16:56>show tables like '%setup%'; +----------------------------------------+ | Tables_in_performance_schema (%setup%) | +----------------------------------------+ | setup_actors | | setup_consumers | | setup_instruments | | setup_objects | | setup_timers | +----------------------------------------+
1,setup_actors:配置用户纬度的监控,默认监控所有用户。
zjy@performance_schema 10:19:11>select * from setup_actors; +------+------+------+ | HOST | USER | ROLE | +------+------+------+ | % | % | % | +------+------+------+
2,setup_consumers:配置events的消费者类型,即收集的events写入到哪些统计表中。
zjy@: performance_schema 10:23:35>select * from setup_consumers; +--------------------------------+---------+ | NAME | ENABLED | +--------------------------------+---------+ | events_stages_current | NO | | events_stages_history | NO | | events_stages_history_long | NO | | events_statements_current | YES | | events_statements_history | NO | | events_statements_history_long | NO | | events_waits_current | NO | | events_waits_history | NO | | events_waits_history_long | NO | | global_instrumentation | YES | | thread_instrumentation | YES | | statements_digest | YES | +--------------------------------+---------+
这里需要说明的是需要查看哪个就更新其ENABLED列为YES。如:
zjy@performance_schema 10:25:02>update setup_consumers set ENABLED='YES' where NAME in ('events_stages_current','events_waits_current'); Query OK, 2 rows affected (0.00 sec)
更新完后立即生效,但是服务器重启之后又会变回默认值,要永久生效需要在配置文件里添加:
[mysqld] #performance_schema performance_schema_consumer_events_waits_current=on performance_schema_consumer_events_stages_current=on performance_schema_consumer_events_statements_current=on performance_schema_consumer_events_waits_history=on performance_schema_consumer_events_stages_history=on performance_schema_consumer_events_statements_history=on
即在这些表的前面加上:performance_schema_consumer_xxx。表setup_consumers里面的值有个层级关系:
global_instrumentation > thread_instrumentation = statements_digest > events_stages_current = events_statements_current = events_waits_current > events_stages_history = events_statements_history = events_waits_history > events_stages_history_long = events_statements_history_long = events_waits_history_long
只有上一层次的为YES,才会继续检查该本层为YES or NO。global_instrumentation是最高级别consumer,如果它设置为NO,则所有的consumer都会忽略。其中history和history_long存的是current表的历史记录条数,history表记录了每个线程最近等待的10个事件,而history_long表则记录了最近所有线程产生的10000个事件,这里的10和10000都是可以配置的。这三个表表结构相同,history和history_long表数据都来源于current表。长度通过控制参数:
zjy@performance_schema 11:10:03>show variables like 'performance_schema%history%size'; +--------------------------------------------------------+-------+ | Variable_name | Value | +--------------------------------------------------------+-------+ | performance_schema_events_stages_history_long_size | 10000 | | performance_schema_events_stages_history_size | 10 | | performance_schema_events_statements_history_long_size | 10000 | | performance_schema_events_statements_history_size | 10 | | performance_schema_events_waits_history_long_size | 10000 | | performance_schema_events_waits_history_size | 10 | +--------------------------------------------------------+-------+
3,setup_instruments:配置具体的instrument,主要包含4大类:idle、stage/xxx、statement/xxx、wait/xxx:
zjy@performance_schema 10:56:35>select name,count(*) from setup_instruments group by LEFT(name,5); +---------------------------------+----------+ | name | count(*) | +---------------------------------+----------+ | idle | 1 | | stage/sql/After create | 111 | | statement/sql/select | 179 | | wait/synch/mutex/sql/PAGE::lock | 296 | +---------------------------------+----------+
idle表示socket空闲的时间,stage类表示语句的每个执行阶段的统计,statement类统计语句维度的信息,wait类统计各种等待事件,比如IO,mutux,spin_lock,condition等。
4,setup_objects:配置监控对象,默认对mysql,performance_schema和information_schema中的表都不监控,而其它DB的所有表都监控。
zjy@performance_schema 11:00:18>select * from setup_objects; +-------------+--------------------+-------------+---------+-------+ | OBJECT_TYPE | OBJECT_SCHEMA | OBJECT_NAME | ENABLED | TIMED | +-------------+--------------------+-------------+---------+-------+ | TABLE | mysql | % | NO | NO | | TABLE | performance_schema | % | NO | NO | | TABLE | information_schema | % | NO | NO | | TABLE | % | % | YES | YES | +-------------+--------------------+-------------+---------+-------+
5,setup_timers:配置每种类型指令的统计时间单位。MICROSECOND表示统计单位是微妙,CYCLE表示统计单位是时钟周期,时间度量与CPU的主频有关,NANOSECOND表示统计单位是纳秒。但无论采用哪种度量单位,最终统计表中统计的时间都会装换到皮秒。(1秒=1000000000000皮秒)
zjy@performance_schema 11:05:12>select * from setup_timers; +-----------+-------------+ | NAME | TIMER_NAME | +-----------+-------------+ | idle | MICROSECOND | | wait | CYCLE | | stage | NANOSECOND | | statement | NANOSECOND | +-----------+-------------+
二:instance表
1,cond_instances:条件等待对象实例
表中记录了系统中使用的条件变量的对象,OBJECT_INSTANCE_BEGIN为对象的内存地址。
2,file_instances:文件实例
表中记录了系统中打开了文件的对象,包括ibdata文件,redo文件,binlog文件,用户的表文件等,open_count显示当前文件打开的数目,如果重来没有打开过,不会出现在表中。
zjy@performance_schema 11:20:04>select * from file_instances limit 2,5; +---------------------------------+--------------------------------------+------------+ | FILE_NAME | EVENT_NAME | OPEN_COUNT | +---------------------------------+--------------------------------------+------------+ | /var/lib/mysql/mysql/plugin.frm | wait/io/file/sql/FRM | 0 | | /var/lib/mysql/mysql/plugin.MYI | wait/io/file/myisam/kfile | 1 | | /var/lib/mysql/mysql/plugin.MYD | wait/io/file/myisam/dfile | 1 | | /var/lib/mysql/ibdata1 | wait/io/file/innodb/innodb_data_file | 2 | | /var/lib/mysql/ib_logfile0 | wait/io/file/innodb/innodb_log_file | 2 | +---------------------------------+--------------------------------------+------------+
3,mutex_instances:互斥同步对象实例
表中记录了系统中使用互斥量对象的所有记录,其中name为:wait/synch/mutex/*。LOCKED_BY_THREAD_ID显示哪个线程正持有mutex,若没有线程持有,则为NULL。
4,rwlock_instances: 读写锁同步对象实例
表中记录了系统中使用读写锁对象的所有记录,其中name为 wait/synch/rwlock/*。WRITE_LOCKED_BY_THREAD_ID为正在持有该对象的thread_id,若没有线程持有,则为NULL。READ_LOCKED_BY_COUNT为记录了同时有多少个读者持有读锁。(通过 events_waits_current 表可以知道,哪个线程在等待锁;通过rwlock_instances知道哪个线程持有锁。rwlock_instances的缺陷是,只能记录持有写锁的线程,对于读锁则无能为力)。
5,socket_instances:活跃会话对象实例
表中记录了thread_id,socket_id,ip和port,其它表可以通过thread_id与socket_instance进行关联,获取IP-PORT信息,能够与应用对接起来。
event_name主要包含3类:
wait/io/socket/sql/server_unix_socket,服务端unix监听socket
wait/io/socket/sql/server_tcpip_socket,服务端tcp监听socket
wait/io/socket/sql/client_connection,客户端socket
三:Wait表
1,events_waits_current:记录了当前线程等待的事件
2,events_waits_history:记录了每个线程最近等待的10个事件
3,events_waits_history_long:记录了最近所有线程产生的10000个事件
表结构定义如下:
CREATE TABLE `events_waits_current` ( `THREAD_ID` bigint(20) unsigned NOT NULL COMMENT '线程ID', `EVENT_ID` bigint(20) unsigned NOT NULL COMMENT '当前线程的事件ID,和THREAD_ID确定唯一', `END_EVENT_ID` bigint(20) unsigned DEFAULT NULL COMMENT '当事件开始时,这一列被设置为NULL。当事件结束时,再更新为当前的事件ID', `EVENT_NAME` varchar(128) NOT NULL COMMENT '事件名称', `SOURCE` varchar(64) DEFAULT NULL COMMENT '该事件产生时的源码文件', `TIMER_START` bigint(20) unsigned DEFAULT NULL COMMENT '事件开始时间(皮秒)', `TIMER_END` bigint(20) unsigned DEFAULT NULL COMMENT '事件结束结束时间(皮秒)', `TIMER_WAIT` bigint(20) unsigned DEFAULT NULL COMMENT '事件等待时间(皮秒)', `SPINS` int(10) unsigned DEFAULT NULL COMMENT '', `OBJECT_SCHEMA` varchar(64) DEFAULT NULL COMMENT '库名', `OBJECT_NAME` varchar(512) DEFAULT NULL COMMENT '文件名、表名、IP:SOCK值', `OBJECT_TYPE` varchar(64) DEFAULT NULL COMMENT 'FILE、TABLE、TEMPORARY TABLE', `INDEX_NAME` varchar(64) DEFAULT NULL COMMENT '索引名', `OBJECT_INSTANCE_BEGIN` bigint(20) unsigned NOT NULL COMMENT '内存地址', `NESTING_EVENT_ID` bigint(20) unsigned DEFAULT NULL COMMENT '该事件对应的父事件ID', `NESTING_EVENT_TYPE` enum('STATEMENT','STAGE','WAIT') DEFAULT NULL COMMENT '父事件类型(STATEMENT, STAGE, WAIT)', `OPERATION` varchar(32) NOT NULL COMMENT '操作类型(lock, read, write)', `NUMBER_OF_BYTES` bigint(20) DEFAULT NULL COMMENT '', `FLAGS` int(10) unsigned DEFAULT NULL COMMENT '标记' ) ENGINE=PERFORMANCE_SCHEMA DEFAULT CHARSET=utf8
四:Stage 表
1,events_stages_current:记录了当前线程所处的执行阶段
2,events_stages_history:记录了当前线程所处的执行阶段10条历史记录
3,events_stages_history_long:记录了当前线程所处的执行阶段10000条历史记录
表结构定义如下:
CREATE TABLE `events_stages_current` ( `THREAD_ID` bigint(20) unsigned NOT NULL COMMENT '线程ID', `EVENT_ID` bigint(20) unsigned NOT NULL COMMENT '事件ID', `END_EVENT_ID` bigint(20) unsigned DEFAULT NULL COMMENT '结束事件ID', `EVENT_NAME` varchar(128) NOT NULL COMMENT '事件名称', `SOURCE` varchar(64) DEFAULT NULL COMMENT '源码位置', `TIMER_START` bigint(20) unsigned DEFAULT NULL COMMENT '事件开始时间(皮秒)', `TIMER_END` bigint(20) unsigned DEFAULT NULL COMMENT '事件结束结束时间(皮秒)', `TIMER_WAIT` bigint(20) unsigned DEFAULT NULL COMMENT '事件等待时间(皮秒)', `NESTING_EVENT_ID` bigint(20) unsigned DEFAULT NULL COMMENT '该事件对应的父事件ID', `NESTING_EVENT_TYPE` enum('STATEMENT','STAGE','WAIT') DEFAULT NULL COMMENT '父事件类型(STATEMENT, STAGE, WAIT)' ) ENGINE=PERFORMANCE_SCHEMA DEFAULT CHARSET=utf8
五:Statement 表
1,events_statements_current:通过 thread_id+event_id可以唯一确定一条记录。Statments表只记录最顶层的请求,SQL语句或是COMMAND,每条语句一行。event_name形式为statement/sql/*,或statement/com/*
2,events_statements_history
3,events_statements_history_long
表结构定义如下:
CREATE TABLE `events_statements_current` ( `THREAD_ID` bigint(20) unsigned NOT NULL COMMENT '线程ID', `EVENT_ID` bigint(20) unsigned NOT NULL COMMENT '事件ID', `END_EVENT_ID` bigint(20) unsigned DEFAULT NULL COMMENT '结束事件ID', `EVENT_NAME` varchar(128) NOT NULL COMMENT '事件名称', `SOURCE` varchar(64) DEFAULT NULL COMMENT '源码位置', `TIMER_START` bigint(20) unsigned DEFAULT NULL COMMENT '事件开始时间(皮秒)', `TIMER_END` bigint(20) unsigned DEFAULT NULL COMMENT '事件结束结束时间(皮秒)', `TIMER_WAIT` bigint(20) unsigned DEFAULT NULL COMMENT '事件等待时间(皮秒)', `LOCK_TIME` bigint(20) unsigned NOT NULL COMMENT '锁时间', `SQL_TEXT` longtext COMMENT '记录SQL语句', `DIGEST` varchar(32) DEFAULT NULL COMMENT '对SQL_TEXT做MD5产生的32位字符串', `DIGEST_TEXT` longtext COMMENT '将语句中值部分用问号代替,用于SQL语句归类', `CURRENT_SCHEMA` varchar(64) DEFAULT NULL COMMENT '默认的数据库名', `OBJECT_TYPE` varchar(64) DEFAULT NULL COMMENT '保留字段', `OBJECT_SCHEMA` varchar(64) DEFAULT NULL COMMENT '保留字段', `OBJECT_NAME` varchar(64) DEFAULT NULL COMMENT '保留字段', `OBJECT_INSTANCE_BEGIN` bigint(20) unsigned DEFAULT NULL COMMENT '内存地址', `MYSQL_ERRNO` int(11) DEFAULT NULL COMMENT '', `RETURNED_SQLSTATE` varchar(5) DEFAULT NULL COMMENT '', `MESSAGE_TEXT` varchar(128) DEFAULT NULL COMMENT '信息', `ERRORS` bigint(20) unsigned NOT NULL COMMENT '错误数目', `WARNINGS` bigint(20) unsigned NOT NULL COMMENT '警告数目', `ROWS_AFFECTED` bigint(20) unsigned NOT NULL COMMENT '影响的数目', `ROWS_SENT` bigint(20) unsigned NOT NULL COMMENT '返回的记录数', `ROWS_EXAMINED` bigint(20) unsigned NOT NULL COMMENT '读取扫描的记录数目', `CREATED_TMP_DISK_TABLES` bigint(20) unsigned NOT NULL COMMENT '创建磁盘临时表数目', `CREATED_TMP_TABLES` bigint(20) unsigned NOT NULL COMMENT '创建临时表数目', `SELECT_FULL_JOIN` bigint(20) unsigned NOT NULL COMMENT 'join时,第一个表为全表扫描的数目', `SELECT_FULL_RANGE_JOIN` bigint(20) unsigned NOT NULL COMMENT '引用表采用range方式扫描的数目', `SELECT_RANGE` bigint(20) unsigned NOT NULL COMMENT 'join时,第一个表采用range方式扫描的数目', `SELECT_RANGE_CHECK` bigint(20) unsigned NOT NULL COMMENT '', `SELECT_SCAN` bigint(20) unsigned NOT NULL COMMENT 'join时,第一个表位全表扫描的数目', `SORT_MERGE_PASSES` bigint(20) unsigned NOT NULL COMMENT '', `SORT_RANGE` bigint(20) unsigned NOT NULL COMMENT '范围排序数目', `SORT_ROWS` bigint(20) unsigned NOT NULL COMMENT '排序的记录数目', `SORT_SCAN` bigint(20) unsigned NOT NULL COMMENT '全表排序数目', `NO_INDEX_USED` bigint(20) unsigned NOT NULL COMMENT '没有使用索引数目', `NO_GOOD_INDEX_USED` bigint(20) unsigned NOT NULL COMMENT '', `NESTING_EVENT_ID` bigint(20) unsigned DEFAULT NULL COMMENT '该事件对应的父事件ID', `NESTING_EVENT_TYPE` enum('STATEMENT','STAGE','WAIT') DEFAULT NULL COMMENT '父事件类型(STATEMENT, STAGE, WAIT)' ) ENGINE=PERFORMANCE_SCHEMA DEFAULT CHARSET=utf8
六:Connection 表
1,users:记录用户连接数信息
2,hosts:记录了主机连接数信息
3,accounts:记录了用户主机连接数信息
zjy@performance_schema 12:03:27>select * from users; +------------------+---------------------+-------------------+ | USER | CURRENT_CONNECTIONS | TOTAL_CONNECTIONS | +------------------+---------------------+-------------------+ | debian-sys-maint | 0 | 36 | | zjy | 1 | 22285 | | dchat_php | 0 | 37864 | | dxyslave | 2 | 9 | | nagios | 0 | 10770 | | dchat_data | 140 | 2233023 | | NULL | 0 | 15866 | | dchat_api | 160 | 2754212 | | mha_data | 1 | 36 | | backup | 0 | 15 | | cacti | 0 | 4312 | | kol | 10 | 172414 | +------------------+---------------------+-------------------+ 12 rows in set (0.00 sec) zjy@performance_schema 12:03:34>select * from hosts; +-----------------+---------------------+-------------------+ | HOST | CURRENT_CONNECTIONS | TOTAL_CONNECTIONS | +-----------------+---------------------+-------------------+ | 192.168.100.218 | 150 | 2499422 | | 192.168.100.240 | 10 | 172429 | | 192.168.100.139 | 0 | 698 | | 192.168.100.21 | 0 | 2 | | 192.168.100.220 | 150 | 2526136 | | 192.168.100.25 | 1 | 7 | | NULL | 0 | 15867 | | 192.168.100.241 | 0 | 21558 | | 192.168.100.191 | 1 | 34 | | localhost | 0 | 10807 | | 192.168.100.118 | 1 | 2 | | 192.168.100.251 | 0 | 4312 | | 192.168.100.23 | 1 | 31 | | 192.168.100.193 | 0 | 15 | +-----------------+---------------------+-------------------+ 14 rows in set (0.01 sec) zjy@performance_schema 12:05:21>select * from accounts; +------------------+-----------------+---------------------+-------------------+ | USER | HOST | CURRENT_CONNECTIONS | TOTAL_CONNECTIONS | +------------------+-----------------+---------------------+-------------------+ | cacti | 192.168.100.251 | 0 | 4313 | | debian-sys-maint | localhost | 0 | 36 | | backup | 192.168.100.193 | 0 | 15 | | dchat_api | 192.168.100.220 | 80 | 1382585 | | dchat_php | 192.168.100.220 | 0 | 20292 | | zjy | 192.168.100.139 | 0 | 698 | | zjy | 192.168.100.241 | 0 | 21558 | | mha_data | 192.168.100.191 | 1 | 34 | | dxyslave | 192.168.100.118 | 1 | 2 | | kol | 192.168.100.240 | 10 | 172431 | | dxyslave | 192.168.100.25 | 1 | 7 | | dchat_data | 192.168.100.218 | 70 | 1109974 | | zjy | 192.168.100.23 | 1 | 31 | | dchat_php | 192.168.100.218 | 0 | 17572 | | dchat_data | 192.168.100.220 | 70 | 1123306 | | NULL | NULL | 0 | 15868 | | mha_data | 192.168.100.21 | 0 | 2 | | dchat_api | 192.168.100.218 | 80 | 1371918 | | nagios | localhost | 0 | 10771 | +------------------+-----------------+---------------------+-------------------+
七:Summary 表: Summary表聚集了各个维度的统计信息包括表维度,索引维度,会话维度,语句维度和锁维度的统计信息
1,events_waits_summary_global_by_event_name:按等待事件类型聚合,每个事件一条记录
CREATE TABLE `events_waits_summary_global_by_event_name` ( `EVENT_NAME` varchar(128) NOT NULL COMMENT '事件名称', `COUNT_STAR` bigint(20) unsigned NOT NULL COMMENT '事件计数', `SUM_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '总的等待时间', `MIN_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '最小等待时间', `AVG_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '平均等待时间', `MAX_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '最大等待时间' ) ENGINE=PERFORMANCE_SCHEMA DEFAULT CHARSET=utf8
2,events_waits_summary_by_instance:按等待事件对象聚合,同一种等待事件,可能有多个实例,每个实例有不同的内存地址,因此
event_name+object_instance_begin唯一确定一条记录。
CREATE TABLE `events_waits_summary_by_instance` ( `EVENT_NAME` varchar(128) NOT NULL COMMENT '事件名称', `OBJECT_INSTANCE_BEGIN` bigint(20) unsigned NOT NULL COMMENT '内存地址', `COUNT_STAR` bigint(20) unsigned NOT NULL COMMENT '事件计数', `SUM_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '总的等待时间', `MIN_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '最小等待时间', `AVG_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '平均等待时间', `MAX_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '最大等待时间' ) ENGINE=PERFORMANCE_SCHEMA DEFAULT CHARSET=utf8
3,events_waits_summary_by_thread_by_event_name:按每个线程和事件来统计,thread_id+event_name唯一确定一条记录。
CREATE TABLE `events_waits_summary_by_thread_by_event_name` ( `THREAD_ID` bigint(20) unsigned NOT NULL COMMENT '线程ID', `EVENT_NAME` varchar(128) NOT NULL COMMENT '事件名称', `COUNT_STAR` bigint(20) unsigned NOT NULL COMMENT '事件计数', `SUM_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '总的等待时间', `MIN_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '最小等待时间', `AVG_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '平均等待时间', `MAX_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '最大等待时间' ) ENGINE=PERFORMANCE_SCHEMA DEFAULT CHARSET=utf8
4,events_stages_summary_global_by_event_name:按事件阶段类型聚合,每个事件一条记录,表结构同上。
5,events_stages_summary_by_thread_by_event_name:按每个线程和事件来阶段统计,表结构同上。
6,events_statements_summary_by_digest:按照事件的语句进行聚合。
CREATE TABLE `events_statements_summary_by_digest` ( `SCHEMA_NAME` varchar(64) DEFAULT NULL COMMENT '库名', `DIGEST` varchar(32) DEFAULT NULL COMMENT '对SQL_TEXT做MD5产生的32位字符串。如果为consumer表中没有打开statement_digest选项,则为NULL', `DIGEST_TEXT` longtext COMMENT '将语句中值部分用问号代替,用于SQL语句归类。如果为consumer表中没有打开statement_digest选项,则为NULL。', `COUNT_STAR` bigint(20) unsigned NOT NULL COMMENT '事件计数', `SUM_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '总的等待时间', `MIN_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '最小等待时间', `AVG_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '平均等待时间', `MAX_TIMER_WAIT` bigint(20) unsigned NOT NULL COMMENT '最大等待时间', `SUM_LOCK_TIME` bigint(20) unsigned NOT NULL COMMENT '锁时间总时长', `SUM_ERRORS` bigint(20) unsigned NOT NULL COMMENT '错误数的总', `SUM_WARNINGS` bigint(20) unsigned NOT NULL COMMENT '警告的总数', `SUM_ROWS_AFFECTED` bigint(20) unsigned NOT NULL COMMENT '影响的总数目', `SUM_ROWS_SENT` bigint(20) unsigned NOT NULL COMMENT '返回总数目', `SUM_ROWS_EXAMINED` bigint(20) unsigned NOT NULL COMMENT '总的扫描的数目', `SUM_CREATED_TMP_DISK_TABLES` bigint(20) unsigned NOT NULL COMMENT '创建磁盘临时表的总数目', `SUM_CREATED_TMP_TABLES` bigint(20) unsigned NOT NULL COMMENT '创建临时表的总数目', `SUM_SELECT_FULL_JOIN` bigint(20) unsigned NOT NULL COMMENT '第一个表全表扫描的总数目', `SUM_SELECT_FULL_RANGE_JOIN` bigint(20) unsigned NOT NULL COMMENT '总的采用range方式扫描的数目', `SUM_SELECT_RANGE` bigint(20) unsigned NOT NULL COMMENT '第一个表采用range方式扫描的总数目', `SUM_SELECT_RANGE_CHECK` bigint(20) unsigned NOT NULL COMMENT '', `SUM_SELECT_SCAN` bigint(20) unsigned NOT NULL COMMENT '第一个表位全表扫描的总数目', `SUM_SORT_MERGE_PASSES` bigint(20) unsigned NOT NULL COMMENT '', `SUM_SORT_RANGE` bigint(20) unsigned NOT NULL COMMENT '范围排序总数', `SUM_SORT_ROWS` bigint(20) unsigned NOT NULL COMMENT '排序的记录总数目', `SUM_SORT_SCAN` bigint(20) unsigned NOT NULL COMMENT '第一个表排序扫描总数目', `SUM_NO_INDEX_USED` bigint(20) unsigned NOT NULL COMMENT '没有使用索引总数', `SUM_NO_GOOD_INDEX_USED` bigint(20) unsigned NOT NULL COMMENT '', `FIRST_SEEN` timestamp NOT NULL DEFAULT '0000-00-00 00:00:00' COMMENT '第一次执行时间', `LAST_SEEN` timestamp NOT NULL DEFAULT '0000-00-00 00:00:00' COMMENT '最后一次执行时间' ) ENGINE=PERFORMANCE_SCHEMA DEFAULT CHARSET=utf8
7,events_statements_summary_global_by_event_name:按照事件的语句进行聚合。表结构同上。
8,events_statements_summary_by_thread_by_event_name:按照线程和事件的语句进行聚合,表结构同上。
9,file_summary_by_instance:按事件类型统计(物理IO维度)
10,file_summary_by_event_name:具体文件统计(物理IO维度)
9和10一起说明:
统计IO操作:COUNT_STAR,SUM_TIMER_WAIT,MIN_TIMER_WAIT,AVG_TIMER_WAIT,MAX_TIMER_WAIT
统计读 :COUNT_READ,SUM_TIMER_READ,MIN_TIMER_READ,AVG_TIMER_READ,MAX_TIMER_READ, SUM_NUMBER_OF_BYTES_READ
统计写 :COUNT_WRITE,SUM_TIMER_WRITE,MIN_TIMER_WRITE,AVG_TIMER_WRITE,MAX_TIMER_WRITE, SUM_NUMBER_OF_BYTES_WRITE
统计其他IO事件,比如create,delete,open,close等:COUNT_MISC,SUM_TIMER_MISC,MIN_TIMER_MISC,AVG_TIMER_MISC,MAX_TIMER_MISC
11,table_io_waits_summary_by_table:根据wait/io/table/sql/handler,聚合每个表的I/O操作(逻辑IO纬度)
统计IO操作:COUNT_STAR,SUM_TIMER_WAIT,MIN_TIMER_WAIT,AVG_TIMER_WAIT,MAX_TIMER_WAIT
统计读 :COUNT_READ,SUM_TIMER_READ,MIN_TIMER_READ,AVG_TIMER_READ,MAX_TIMER_READ
:COUNT_FETCH,SUM_TIMER_FETCH,MIN_TIMER_FETCH,AVG_TIMER_FETCH, MAX_TIMER_FETCH
统计写 :COUNT_WRITE,SUM_TIMER_WRITE,MIN_TIMER_WRITE,AVG_TIMER_WRITE,MAX_TIMER_WRITE
INSERT统计,相应的还有DELETE和UPDATE统计:COUNT_INSERT,SUM_TIMER_INSERT,MIN_TIMER_INSERT,AVG_TIMER_INSERT,MAX_TIMER_INSERT
12,table_io_waits_summary_by_index_usage:与table_io_waits_summary_by_table类似,按索引维度统计
13,table_lock_waits_summary_by_table:聚合了表锁等待事件,包括internal lock 和 external lock
internal lock通过SQL层函数thr_lock调用,OPERATION值为:
read normal、read with shared locks、read high priority、read no insert、write allow write、write concurrent insert、write delayed、write low priority、write normal
external lock则通过接口函数handler::external_lock调用存储引擎层,OPERATION列的值为:read external、write external
14,Connection Summaries表:account、user、host
events_waits_summary_by_account_by_event_name
events_waits_summary_by_user_by_event_name
events_waits_summary_by_host_by_event_name
events_stages_summary_by_account_by_event_name
events_stages_summary_by_user_by_event_name
events_stages_summary_by_host_by_event_name
events_statements_summary_by_account_by_event_name
events_statements_summary_by_user_by_event_name
events_statements_summary_by_host_by_event_name
15,socket_summary_by_instance、socket_summary_by_event_name:socket聚合统计表。
八:其他相关表
1,performance_timers:系统支持的统计时间单位
2,threads:监视服务端的当前运行的线程
统计应用:
关于SQL维度的统计信息主要集中在events_statements_summary_by_digest表中,通过将SQL语句抽象出digest,可以统计某类SQL语句在各个维度的统计信息
1,哪个SQL执行最多:
zjy@performance_schema 11:36:22>SELECT SCHEMA_NAME,DIGEST_TEXT,COUNT_STAR,SUM_ROWS_SENT,SUM_ROWS_EXAMINED,FIRST_SEEN,LAST_SEEN FROM events_statements_summary_by_digest ORDER BY COUNT_STAR desc LIMIT 1G *************************** 1. row *************************** SCHEMA_NAME: dchat DIGEST_TEXT: SELECT ... COUNT_STAR: 1161210102 SUM_ROWS_SENT: 1161207842 SUM_ROWS_EXAMINED: 0 FIRST_SEEN: 2016-02-17 00:36:46 LAST_SEEN: 2016-03-07 11:36:29
各个字段的注释可以看上面的表结构说明:从2月17号到3月7号该SQL执行了1161210102次。
2,哪个SQL平均响应时间最多:
zjy@performance_schema 11:36:28>SELECT SCHEMA_NAME,DIGEST_TEXT,COUNT_STAR,AVG_TIMER_WAIT,SUM_ROWS_SENT,SUM_ROWS_EXAMINED,FIRST_SEEN,LAST_SEEN FROM events_statements_summary_by_digest ORDER BY AVG_TIMER_WAIT desc LIMIT 1G *************************** 1. row *************************** SCHEMA_NAME: dchat DIGEST_TEXT: SELECT ... COUNT_STAR: 1 AVG_TIMER_WAIT: 273238183964000 SUM_ROWS_SENT: 50208 SUM_ROWS_EXAMINED: 5565651 FIRST_SEEN: 2016-02-22 13:27:33 LAST_SEEN: 2016-02-22 13:27:33
各个字段的注释可以看上面的表结构说明:从2月17号到3月7号该SQL平均响应时间273238183964000皮秒(1000000000000皮秒=1秒)
3,哪个SQL扫描的行数最多:
SUM_ROWS_EXAMINED
4,哪个SQL使用的临时表最多:
SUM_CREATED_TMP_DISK_TABLES、SUM_CREATED_TMP_TABLES
5,哪个SQL返回的结果集最多:
SUM_ROWS_SENT
6,哪个SQL排序数最多:
SUM_SORT_ROWS
通过上述指标我们可以间接获得某类SQL的逻辑IO(SUM_ROWS_EXAMINED),CPU消耗(SUM_SORT_ROWS),网络带宽(SUM_ROWS_SENT)的对比。
通过file_summary_by_instance表,可以获得系统运行到现在,哪个文件(表)物理IO最多,这可能意味着这个表经常需要访问磁盘IO。
7,哪个表、文件逻辑IO最多(热数据):
zjy@performance_schema 12:16:18>SELECT FILE_NAME,EVENT_NAME,COUNT_READ,SUM_NUMBER_OF_BYTES_READ,COUNT_WRITE,SUM_NUMBER_OF_BYTES_WRITE FROM file_summary_by_instance ORDER BY SUM_NUMBER_OF_BYTES_READ+SUM_NUMBER_OF_BYTES_WRITE DESC LIMIT 2G *************************** 1. row *************************** FILE_NAME: /var/lib/mysql/ibdata1 #文件 EVENT_NAME: wait/io/file/innodb/innodb_data_file COUNT_READ: 544 SUM_NUMBER_OF_BYTES_READ: 10977280 COUNT_WRITE: 3700729 SUM_NUMBER_OF_BYTES_WRITE: 1433734217728 *************************** 2. row *************************** FILE_NAME: /var/lib/mysql/dchat/fans.ibd #表 EVENT_NAME: wait/io/file/innodb/innodb_data_file COUNT_READ: 9370680 SUM_NUMBER_OF_BYTES_READ: 153529188352 COUNT_WRITE: 67576376 SUM_NUMBER_OF_BYTES_WRITE: 1107815432192
8,哪个索引使用最多:
zjy@performance_schema 12:18:42>SELECT OBJECT_NAME, INDEX_NAME, COUNT_FETCH, COUNT_INSERT, COUNT_UPDATE, COUNT_DELETE FROM table_io_waits_summary_by_index_usage ORDER BY SUM_TIMER_WAIT DESC limit 1; +-------------+------------+-------------+--------------+--------------+--------------+ | OBJECT_NAME | INDEX_NAME | COUNT_FETCH | COUNT_INSERT | COUNT_UPDATE | COUNT_DELETE | +-------------+------------+-------------+--------------+--------------+--------------+ | fans | PRIMARY | 29002695158 | 0 | 296373434 | 0 | +-------------+------------+-------------+--------------+--------------+--------------+ 1 row in set (0.29 sec)
通过table_io_waits_summary_by_index_usage表,可以获得系统运行到现在,哪个表的具体哪个索引(包括主键索引,二级索引)使用最多。
9,哪个索引没有使用过:
zjy@performance_schema 12:23:22>SELECT OBJECT_SCHEMA, OBJECT_NAME, INDEX_NAME FROM table_io_waits_summary_by_index_usage WHERE INDEX_NAME IS NOT NULL AND COUNT_STAR = 0 AND OBJECT_SCHEMA <> 'mysql' ORDER BY OBJECT_SCHEMA,OBJECT_NAME;
10,哪个等待事件消耗的时间最多:
zjy@performance_schema 12:25:22>SELECT EVENT_NAME, COUNT_STAR, SUM_TIMER_WAIT, AVG_TIMER_WAIT FROM events_waits_summary_global_by_event_name WHERE event_name != 'idle' ORDER BY SUM_TIMER_WAIT DESC LIMIT 1;
11,类似profiling功能:
分析具体某条SQL,该SQL在执行各个阶段的时间消耗,通过events_statements_xxx表和events_stages_xxx表,就可以达到目的。两个表通过event_id与nesting_event_id关联,stages表的nesting_event_id为对应statements表的event_id;针对每个stage可能出现的锁等待,一个stage会对应一个或多个wait,通过stage_xxx表的event_id字段与waits_xxx表的nesting_event_id进行关联。如:
比如分析包含count(*)的某条SQL语句,具体如下: SELECT EVENT_ID, sql_text FROM events_statements_history WHERE sql_text LIKE '%count(*)%'; +----------+--------------------------------------+ | EVENT_ID | sql_text | +----------+--------------------------------------+ | 1690 | select count(*) from chuck.test_slow | +----------+--------------------------------------+ 首先得到了语句的event_id为1690,通过查找events_stages_xxx中nesting_event_id为1690的记录,可以达到目的。 a.查看每个阶段的时间消耗: SELECT event_id, EVENT_NAME, SOURCE, TIMER_END - TIMER_START FROM events_stages_history_long WHERE NESTING_EVENT_ID = 1690; +----------+--------------------------------+----------------------+-----------------------+ | event_id | EVENT_NAME | SOURCE | TIMER_END-TIMER_START | +----------+--------------------------------+----------------------+-----------------------+ | 1691 | stage/sql/init | mysqld.cc:990 | 316945000 | | 1693 | stage/sql/checking permissions | sql_parse.cc:5776 | 26774000 | | 1695 | stage/sql/Opening tables | sql_base.cc:4970 | 41436934000 | | 2638 | stage/sql/init | sql_select.cc:1050 | 85757000 | | 2639 | stage/sql/System lock | lock.cc:303 | 40017000 | | 2643 | stage/sql/optimizing | sql_optimizer.cc:138 | 38562000 | | 2644 | stage/sql/statistics | sql_optimizer.cc:362 | 52845000 | | 2645 | stage/sql/preparing | sql_optimizer.cc:485 | 53196000 | | 2646 | stage/sql/executing | sql_executor.cc:112 | 3153000 | | 2647 | stage/sql/Sending data | sql_executor.cc:192 | 7369072089000 | | 4304138 | stage/sql/end | sql_select.cc:1105 | 19920000 | | 4304139 | stage/sql/query end | sql_parse.cc:5463 | 44721000 | | 4304145 | stage/sql/closing tables | sql_parse.cc:5524 | 61723000 | | 4304152 | stage/sql/freeing items | sql_parse.cc:6838 | 455678000 | | 4304155 | stage/sql/logging slow query | sql_parse.cc:2258 | 83348000 | | 4304159 | stage/sql/cleaning up | sql_parse.cc:2163 | 4433000 | +----------+--------------------------------+----------------------+-----------------------+ 通过间接关联,我们能分析得到SQL语句在每个阶段的时间消耗,时间单位以皮秒表示。这里展示的结果很类似profiling功能,有了performance schema,就不再需要profiling这个功能了。另外需要注意的是,由于默认情况下events_stages_history表中只为每个连接记录了最近10条记录,为了确保获取所有记录,需要访问events_stages_history_long表 b.查看某个阶段的锁等待情况 针对每个stage可能出现的锁等待,一个stage会对应一个或多个wait,events_waits_history_long这个表容易爆满[默认阀值10000]。由于select count(*)需要IO(逻辑IO或者物理IO),所以在stage/sql/Sending data阶段会有io等待的统计。通过stage_xxx表的event_id字段与waits_xxx表的nesting_event_id进行关联。 SELECT event_id, event_name, source, timer_wait, object_name, index_name, operation, nesting_event_id FROM events_waits_history_long WHERE nesting_event_id = 2647; +----------+---------------------------+-----------------+------------+-------------+------------+-----------+------------------+ | event_id | event_name | source | timer_wait | object_name | index_name | operation | nesting_event_id | +----------+---------------------------+-----------------+------------+-------------+------------+-----------+------------------+ | 190607 | wait/io/table/sql/handler | handler.cc:2842 | 1845888 | test_slow | idx_c1 | fetch | 2647 | | 190608 | wait/io/table/sql/handler | handler.cc:2842 | 1955328 | test_slow | idx_c1 | fetch | 2647 | | 190609 | wait/io/table/sql/handler | handler.cc:2842 | 1929792 | test_slow | idx_c1 | fetch | 2647 | | 190610 | wait/io/table/sql/handler | handler.cc:2842 | 1869600 | test_slow | idx_c1 | fetch | 2647 | | 190611 | wait/io/table/sql/handler | handler.cc:2842 | 1922496 | test_slow | idx_c1 | fetch | 2647 | +----------+---------------------------+-----------------+------------+-------------+------------+-----------+------------------+ 通过上面的实验,我们知道了statement,stage,wait的三级结构,通过nesting_event_id进行关联,它表示某个事件的父event_id。 (2).模拟innodb行锁等待的例子 会话A执行语句update test_icp set y=y+1 where x=1(x为primary key),不commit;会话B执行同样的语句update test_icp set y=y+1 where x=1,会话B堵塞,并最终报错。通过连接连接查询events_statements_history_long和events_stages_history_long,可以看到在updating阶段花了大约60s的时间。这主要因为实例上的innodb_lock_wait_timeout设置为60,等待60s后超时报错了。 SELECT statement.EVENT_ID, stages.event_id, statement.sql_text, stages.event_name, stages.timer_wait FROM events_statements_history_long statement join events_stages_history_long stages on statement.event_id=stages.nesting_event_id WHERE statement.sql_text = 'update test_icp set y=y+1 where x=1'; +----------+----------+-------------------------------------+--------------------------------+----------------+ | EVENT_ID | event_id | sql_text | event_name | timer_wait | +----------+----------+-------------------------------------+--------------------------------+----------------+ | 5816 | 5817 | update test_icp set y=y+1 where x=1 | stage/sql/init | 195543000 | | 5816 | 5819 | update test_icp set y=y+1 where x=1 | stage/sql/checking permissions | 22730000 | | 5816 | 5821 | update test_icp set y=y+1 where x=1 | stage/sql/Opening tables | 66079000 | | 5816 | 5827 | update test_icp set y=y+1 where x=1 | stage/sql/init | 89116000 | | 5816 | 5828 | update test_icp set y=y+1 where x=1 | stage/sql/System lock | 218744000 | | 5816 | 5832 | update test_icp set y=y+1 where x=1 | stage/sql/updating | 6001362045000 | | 5816 | 5968 | update test_icp set y=y+1 where x=1 | stage/sql/end | 10435000 | | 5816 | 5969 | update test_icp set y=y+1 where x=1 | stage/sql/query end | 85979000 | | 5816 | 5983 | update test_icp set y=y+1 where x=1 | stage/sql/closing tables | 56562000 | | 5816 | 5990 | update test_icp set y=y+1 where x=1 | stage/sql/freeing items | 83563000 | | 5816 | 5992 | update test_icp set y=y+1 where x=1 | stage/sql/cleaning up | 4589000 | +----------+----------+-------------------------------------+--------------------------------+----------------+ 查看wait事件: SELECT event_id, event_name, source, timer_wait, object_name, index_name, operation, nesting_event_id FROM events_waits_history_long WHERE nesting_event_id = 5832; *************************** 1. row *************************** event_id: 5832 event_name: wait/io/table/sql/handler source: handler.cc:2782 timer_wait: 6005946156624 object_name: test_icp index_name: PRIMARY operation: fetch 从结果来看,waits表中记录了一个fetch等待事件,但并没有更细的innodb行锁等待事件统计。 (3).模拟MDL锁等待的例子 会话A执行一个大查询select count(*) from test_slow,会话B执行表结构变更alter table test_slow modify c2 varchar(152);通过如下语句可以得到alter语句的执行过程,重点关注“stage/sql/Waiting for table metadata lock”阶段。 SELECT statement.EVENT_ID, stages.event_id, statement.sql_text, stages.event_name as stage_name, stages.timer_wait as stage_time FROM events_statements_history_long statement left join events_stages_history_long stages on statement.event_id=stages.nesting_event_id WHERE statement.sql_text = 'alter table test_slow modify c2 varchar(152)'; +-----------+-----------+----------------------------------------------+----------------------------------------------------+---------------+ | EVENT_ID | event_id | sql_text | stage_name | stage_time | +-----------+-----------+----------------------------------------------+----------------------------------------------------+---------------+ | 326526744 | 326526745 | alter table test_slow modify c2 varchar(152) | stage/sql/init | 216662000 | | 326526744 | 326526747 | alter table test_slow modify c2 varchar(152) | stage/sql/checking permissions | 18183000 | | 326526744 | 326526748 | alter table test_slow modify c2 varchar(152) | stage/sql/checking permissions | 10294000 | | 326526744 | 326526750 | alter table test_slow modify c2 varchar(152) | stage/sql/init | 4783000 | | 326526744 | 326526751 | alter table test_slow modify c2 varchar(152) | stage/sql/Opening tables | 140172000 | | 326526744 | 326526760 | alter table test_slow modify c2 varchar(152) | stage/sql/setup | 157643000 | | 326526744 | 326526769 | alter table test_slow modify c2 varchar(152) | stage/sql/creating table | 8723217000 | | 326526744 | 326526803 | alter table test_slow modify c2 varchar(152) | stage/sql/After create | 257332000 | | 326526744 | 326526832 | alter table test_slow modify c2 varchar(152) | stage/sql/Waiting for table metadata lock | 1000181831000 | | 326526744 | 326526835 | alter table test_slow modify c2 varchar(152) | stage/sql/After create | 33483000 | | 326526744 | 326526838 | alter table test_slow modify c2 varchar(152) | stage/sql/Waiting for table metadata lock | 1000091810000 | | 326526744 | 326526841 | alter table test_slow modify c2 varchar(152) | stage/sql/After create | 17187000 | | 326526744 | 326526844 | alter table test_slow modify c2 varchar(152) | stage/sql/Waiting for table metadata lock | 1000126464000 | | 326526744 | 326526847 | alter table test_slow modify c2 varchar(152) | stage/sql/After create | 27472000 | | 326526744 | 326526850 | alter table test_slow modify c2 varchar(152) | stage/sql/Waiting for table metadata lock | 561996133000 | | 326526744 | 326526853 | alter table test_slow modify c2 varchar(152) | stage/sql/After create | 124876000 | | 326526744 | 326526877 | alter table test_slow modify c2 varchar(152) | stage/sql/System lock | 30659000 | | 326526744 | 326526881 | alter table test_slow modify c2 varchar(152) | stage/sql/preparing for alter table | 40246000 | | 326526744 | 326526889 | alter table test_slow modify c2 varchar(152) | stage/sql/altering table | 36628000 | | 326526744 | 326528280 | alter table test_slow modify c2 varchar(152) | stage/sql/end | 43824000 | | 326526744 | 326528281 | alter table test_slow modify c2 varchar(152) | stage/sql/query end | 112557000 | | 326526744 | 326528299 | alter table test_slow modify c2 varchar(152) | stage/sql/closing tables | 27707000 | | 326526744 | 326528305 | alter table test_slow modify c2 varchar(152) | stage/sql/freeing items | 201614000 | | 326526744 | 326528308 | alter table test_slow modify c2 varchar(152) | stage/sql/cleaning up | 3584000 | +-----------+-----------+----------------------------------------------+----------------------------------------------------+---------------+ 从结果可以看到,出现了多次stage/sql/Waiting for table metadata lock阶段,并且间隔1s,说明每隔1s钟会重试判断。找一个该阶段的event_id,通过nesting_event_id关联,确定到底在等待哪个wait事件。 SELECT event_id, event_name, source, timer_wait, object_name, index_name, operation, nesting_event_id FROM events_waits_history_long WHERE nesting_event_id = 326526850; +-----------+---------------------------------------------------+------------------+--------------+-------------+------------+------------+------------------+ | event_id | event_name | source | timer_wait | object_name | index_name | operation | nesting_event_id | +-----------+---------------------------------------------------+------------------+--------------+-------------+------------+------------+------------------+ | 326526851 | wait/synch/cond/sql/MDL_context::COND_wait_status | mdl.cc:1327 | 562417991328 | NULL | NULL | timed_wait | 326526850 | | 326526852 | wait/synch/mutex/mysys/my_thread_var::mutex | sql_class.h:3481 | 733248 | NULL | NULL | lock | 326526850 | +-----------+---------------------------------------------------+------------------+--------------+-------------+------------+------------+------------------+ 通过结果可以知道,产生阻塞的是条件变量MDL_context::COND_wait_status,并且显示了代码的位置。
总结:
本文通过对Performance Schema数据库的介绍,主要用于收集数据库服务器性能参数:①提供进程等待的详细信息,包括锁、互斥变量、文件信息;②保存历史的事件汇总信息,为提供MySQL服务器性能做出详细的判断;③对于新增和删除监控事件点都非常容易,并可以改变mysql服务器的监控周期,例如(CYCLE、MICROSECOND)。通过该库得到数据库运行的统计信息,更好分析定位问题和完善监控信息。类似的监控还有:
打开标准的innodb监控: CREATE TABLE innodb_monitor (a INT) ENGINE=INNODB; 打开innodb的锁监控: CREATE TABLE innodb_lock_monitor (a INT) ENGINE=INNODB; 打开innodb表空间监控: CREATE TABLE innodb_tablespace_monitor (a INT) ENGINE=INNODB; 打开innodb表监控: CREATE TABLE innodb_table_monitor (a INT) ENGINE=INNODB;
参考文章:
https://dev.mysql.com/doc/refman/5.6/en/performance-schema.html
http://www.cnblogs.com/cchust/p/5022148.html
http://www.cnblogs.com/cchust/p/5057498.html
http://www.cnblogs.com/cchust/p/5061131.html