我的学习记录,可能有误请谅解,也提供了一些源码接口供有兴趣的朋友调试。
源码版本:percona 5.7.14
本文并不准备说明如何开启记录慢查询,只是将一些重要的部分进行解析。如何记录慢查询可以自行参考官方文档:
- 5.4.5 The Slow Query Log
本文使用了Percona 版本开启来了参数log_slow_verbosity,得到了更详细的慢查询信息。通常情况下信息没有这么多,但是一定是包含关系,本文也会使用Percona的参数解释说明一下这个参数的含义。
一、慢查询中的时间
实际上慢查询中的时间就是时钟时间,是通过操作系统的命令获得的时间,如下是Linux中获取时间的方式
while (gettimeofday(&t, NULL) != 0)
{}
newtime= (ulonglong)t.tv_sec * 1000000 + t.tv_usec;
return newtime;
实际上就是通过OS的API gettimeofday函数获得的时间。
二、慢查询记录的依据
- long_query_time:如果执行时间超过本参数设置记录慢查询。
- log_queries_not_using_indexes:如果语句未使用索引记录慢查询。
- log_slow_admin_statements:是否记录管理语句。(如ALTER TABLE,ANALYZE TABLE, CHECK TABLE, CREATE INDEX, DROP INDEX, OPTIMIZE TABLE, and REPAIR TABLE.)
本文主要讨论long_query_time参数的含义。
三、long_query_time参数的具体含义
如果我们将语句的执行时间定义为如下:
实际消耗时间 = 实际执行时间+锁等待消耗时间
那么long_query_time实际上界定的是实际执行时间,所以有些情况下虽然语句实际消耗的时间很长但是是因为锁等待时间较长而引起的,那么实际上这种语句也不会记录到慢查询。
我们看一下log_slow_applicable函数的代码片段:
res= cur_utime - thd->utime_after_lock;
if (res > thd->variables.long_query_time)
thd->server_status|= SERVER_QUERY_WAS_SLOW;
else
thd->server_status&= ~SERVER_QUERY_WAS_SLOW;
这里实际上清楚的说明了上面的观点,是不是慢查询就是通过这个函数进行的判断的,非常重要。我可以清晰的看到如下公式:
- res (实际执行时间 ) = cur_utime(实际消耗时间) - thd->utime_after_lock( 锁等待消耗时间)
实际上在慢查询中记录的正是 - Query_time:实际消耗时间
- Lock_time:锁等待消耗时间
但是是否是慢查询其评判标准却是实际执行时间及Query_time - Lock_time
其中锁等待消耗时间( Lock_time)我现在已经知道的包括:
- MySQL层 MDL LOCK等待消耗的时间。(Waiting for table metadata lock)
- MySQL层 MyISAM表锁消耗的时间。 (Waiting for table level lock)
- InnoDB层 行锁消耗的时间。
四、MySQL是如何记录锁时间
我们可以看到在公式中utime_after_lock( 锁等待消耗时间Lock_time)的记录也就成了整个公式的关键,那么我们试着进行debug。
1、MySQL层utime_after_lock的记录方式
不管是 MDL LOCK等待消耗的时间还是 MyISAM表锁消耗的时间都是在MySQL层记录的,实际上它只是记录在函数mysql_lock_tables的末尾会调用的THD::set_time_after_lock进行的记录时间而已如下:
void set_time_after_lock()
{
utime_after_lock= my_micro_time();
MYSQL_SET_STATEMENT_LOCK_TIME(m_statement_psi, (utime_after_lock - start_utime));
}
那么这里可以解析为代码运行到mysql_lock_tables函数的末尾之前的所有的时间都记录到utime_after_lock时间中,实际上并不精确。但是MDL LOCK的获取和MyISAM表锁的获取都包含在里面。所以即便是select语句也可能会看到Lock_time并不为0。下面是部分栈帧:
#0 THD::set_time_after_lock (this=0x7fff28012820) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_class.h:3414
#1 0x0000000001760d6d in mysql_lock_tables (thd=0x7fff28012820, tables=0x7fff28c16b58, count=1, flags=0) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/lock.cc:366
#2 0x000000000151dc1a in lock_tables (thd=0x7fff28012820, tables=0x7fff28c165b0, count=1, flags=0) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_base.cc:6700
#3 0x00000000017c4234 in Sql_cmd_delete::mysql_delete (this=0x7fff28c16b50, thd=0x7fff28012820, limit=18446744073709551615)
at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_delete.cc:136
2、InnoDB层的行锁的utime_after_lock记录方式
InnoDB引擎层调用通过thd_set_lock_wait_time调用thd_storage_lock_wait函数完成的,部分栈帧如下:
#0 thd_storage_lock_wait (thd=0x7fff2c000bc0, value=9503561) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_class.cc:798
#1 0x00000000019a4b2a in thd_set_lock_wait_time (thd=0x7fff2c000bc0, value=9503561)
at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/handler/ha_innodb.cc:1784
#2 0x0000000001a4b50f in lock_wait_suspend_thread (thr=0x7fff2c088200) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/lock/lock0wait.cc:363
#3 0x0000000001b0ec9b in row_mysql_handle_errors (new_err=0x7ffff0317d54, trx=0x7ffff2f2e5d0, thr=0x7fff2c088200, savept=0x0)
at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/row/row0mysql.cc:772
#4 0x0000000001b4fe61 in row_search_mvcc (buf=0x7fff2c087640 "\377", mode=PAGE_CUR_G, prebuilt=0x7fff2c087ac0, match_mode=0, direction=0)
at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/row/row0sel.cc:5940
函数本身还是很简单自己看看就知道了就是相加而已如下:
void thd_storage_lock_wait(THD *thd, long long value)
{
thd->utime_after_lock+= value;
}
五、Percona中的log_slow_verbosity参数
这是Percona的解释:
Specifies how much information to include in your slow log. The value is a comma-delimited string, and can contain any combination of the following values:
- microtime: Log queries with microsecond precision (mandatory).
- query_plan: Log information about the query’s execution plan (optional).
- innodb: Log InnoDB statistics (optional).
- minimal: Equivalent to enabling just microtime.
- standard: Equivalent to enabling microtime,innodb.
- full: Equivalent to all other values OR’ed together.
总之在Percona中可以修改这个参数获得更加详细的信息大概的格式如下:
# Time: 2018-05-30T09:30:12.039775Z
# User@Host: root[root] @ localhost [] Id: 10
# Schema: test Last_errno: 1317 Killed: 0
# Query_time: 19.254508 Lock_time: 0.001043 Rows_sent: 0 Rows_examined: 0 Rows_affected: 0
# Bytes_sent: 44 Tmp_tables: 0 Tmp_disk_tables: 0 Tmp_table_sizes: 0
# InnoDB_trx_id: 0
# QC_Hit: No Full_scan: No Full_join: No Tmp_table: No Tmp_table_on_disk: No
# Filesort: No Filesort_on_disk: No Merge_passes: 0
# InnoDB_IO_r_ops: 0 InnoDB_IO_r_bytes: 0 InnoDB_IO_r_wait: 0.000000
# InnoDB_rec_lock_wait: 0.000000 InnoDB_queue_wait: 0.000000
# InnoDB_pages_distinct: 0
SET timestamp=1527672612;
select count(*) from z1 limit 1;
六、输出的详细解释
本节将会进行详细的解释,全部的慢查询的输出都来自于函数File_query_log::write_slow ,有兴趣的同学可以自己看看,我这里也会给出输出的位置和含义,其中含义部分可能给出的是源码中的注释。
1、第一部分时间
# Time: 2018-05-30T09:30:12.039775Z
对应的代码:
my_snprintf(buff, sizeof buff,"# Time: %s\n", my_timestamp);
其中my_timestamp取值来自于
thd->current_utime();
实际上就是:
while (gettimeofday(&t, NULL) != 0)
{}
newtime= (ulonglong)t.tv_sec * 1000000 + t.tv_usec;
return newtime;
可以看到实际就是调用gettimeofday系统调用得到的系统当前时间。
注意:
对于5.6来讲还有一句判断
if (current_time != last_time)
如果两次打印的时间秒钟一致则不会输出时间,只有通过后面介绍的
SET timestamp=1527753496;
来判断时间,5.7.14没有看到这样的代码。
2、第二部分用户信息
# User@Host: root[root] @ localhost [] Id: 10
对应的代码:
buff_len= my_snprintf(buff, 32, "%5u", thd->thread_id());
if (my_b_printf(&log_file, "# User@Host: %s Id: %s\n", user_host, buff)
== (uint) -1)
goto err;
}
user_host是一串字符串,参考代码:
size_t user_host_len= (strxnmov(user_host_buff, MAX_USER_HOST_SIZE,
sctx->priv_user().str
? sctx->priv_user().str : "",
"[", sctx_user.length ? sctx_user.str :
(thd->slave_thread ? "SQL_SLAVE" : ""),
"] @ ",
sctx_host.length ? sctx_host.str : "", " [",
sctx_ip.length ? sctx_ip.str : "", "]",
NullS) - user_host_buff);
解释如下:
- root: m_priv_user - The user privilege we are using. May be "" for anonymous user。
- [root]: m_user - user of the client, set to NULL until the user has been read from the connection。
- localhost: m_host - host of the client。
- []:client IP m_ip - client IP。
- Id: 10 thd->thread_id()实际上就是show processlist出来的id。
3、第三部分schema等信息
# Schema: test Last_errno: 1317 Killed: 0
对应的代码:
"# Schema: %s Last_errno: %u Killed: %u\n"
(thd->db().str ? thd->db().str : ""),
thd->last_errno, (uint) thd->killed,
Schema:
m_db Name of the current (default) database.If there is the current (default) database, "db" contains its name. If there is no current (default) database, "db" is NULL and "db_length" is 0. In other words, "db", "db_length" must either be NULL, or contain a valid database name.Last_errno:
Variable last_errno contains the last error/warning acquired during query execution.-
Killed: 这里代表的是终止的错误码。源码中如下:
enum killed_state
{
NOT_KILLED=0,
KILL_BAD_DATA=1,
KILL_CONNECTION=ER_SERVER_SHUTDOWN,
KILL_QUERY=ER_QUERY_INTERRUPTED,
KILL_TIMEOUT=ER_QUERY_TIMEOUT,
KILLED_NO_VALUE /* means neither of the states */
};
在错误码中代表如下:
{ "ER_SERVER_SHUTDOWN", 1053, "Server shutdown in progress" },
{ "ER_QUERY_INTERRUPTED", 1317, "Query execution was interrupted" },
{ "ER_QUERY_TIMEOUT", 1886, "Query execution was interrupted, max_statement_time exceeded" },
4、第四部分执行信息
这部分可能是大家最关心的部分,很多信息也是默认输出都会输出的。
# Query_time: 19.254508 Lock_time: 0.001043 Rows_sent: 0 Rows_examined: 0 Rows_affected: 0
# Bytes_sent: 44 Tmp_tables: 0 Tmp_disk_tables: 0 Tmp_table_sizes: 0
# InnoDB_trx_id: 0
对应代码:
my_b_printf(&log_file,
"# Schema: %s Last_errno: %u Killed: %u\n"
"# Query_time: %s Lock_time: %s Rows_sent: %llu"
" Rows_examined: %llu Rows_affected: %llu\n"
"# Bytes_sent: %lu",
(thd->db().str ? thd->db().str : ""),
thd->last_errno, (uint) thd->killed,
query_time_buff, lock_time_buff,
(ulonglong) thd->get_sent_row_count(),
(ulonglong) thd->get_examined_row_count(),
(thd->get_row_count_func() > 0)
? (ulonglong) thd->get_row_count_func() : 0,
(ulong) (thd->status_var.bytes_sent - thd->bytes_sent_old)
my_b_printf(&log_file,
" Tmp_tables: %lu Tmp_disk_tables: %lu "
"Tmp_table_sizes: %llu",
thd->tmp_tables_used, thd->tmp_tables_disk_used,
thd->tmp_tables_size)
snprintf(buf, 20, "%llX", thd->innodb_trx_id);及thd->innodb_trx_id
- Query_time:语句执行的时间及实际消耗时间 。
- Lock_time:包含MDL lock和InnoDB row lock和MyISAM表锁消耗时间的总和及锁等待消耗时间。前面已经进行了描述(实际上也并不全是锁等待的时间只是锁等待包含在其中)。
我们来看看Query_time和Lock_time的源码来源,它们来自于Query_logger::slow_log_write函数如下:
query_utime= (current_utime > thd->start_utime) ?
(current_utime - thd->start_utime) : 0;
lock_utime= (thd->utime_after_lock > thd->start_utime) ?
(thd->utime_after_lock - thd->start_utime) : 0;
下面是数据current_utime 的来源,
current_utime= thd->current_utime();
实际上就是:
while (gettimeofday(&t, NULL) != 0)
{}
newtime= (ulonglong)t.tv_sec * 1000000 + t.tv_usec;
return newtime;
获取当前时间而已
对于thd->utime_after_lock的获取我已经在前文进行了描述,不再解释。
Rows_sent:发送给mysql客户端的行数,下面是源码中的解释
Number of rows we actually sent to the clientRows_examined:InnoDB引擎层扫描的行数,下面是源码中的解释。(备注栈帧1)
Number of rows read and/or evaluated for a statement. Used for slow log reporting.
An examined row is defined as a row that is read and/or evaluated
according to a statement condition, including increate_sort_index(). Rows may be counted more than once, e.g., a statement including ORDER BY could possibly evaluate the row in filesort() before reading it for e.g. update.Rows_affected:涉及到修改的话(比如DML语句)这是受影响的行数。
for DML statements: to the number of affected rows;
for DDL statements: to 0.
Bytes_sent:发送给客户端的实际数据的字节数,它来自于
(ulong) (thd->status_var.bytes_sent - thd->bytes_sent_old)Tmp_tables:临时表的个数。
Tmp_disk_tables:磁盘临时表的个数。
Tmp_table_sizes:临时表的大小。
以上三个指标来自于:
thd->tmp_tables_used
thd->tmp_tables_disk_used
thd->tmp_tables_size
这三个指标增加的位置对应在free_tmp_table函数中如下:
thd->tmp_tables_used++;
if (entry->file)
{
thd->tmp_tables_size += entry->file->stats.data_file_length;
if (entry->file->ht->db_type != DB_TYPE_HEAP)
thd->tmp_tables_disk_used++;
}
- InnoDB_trx_id:事物ID,也就是trx->id,/*!< transaction id */
5、第五部分优化器相关信息
# QC_Hit: No Full_scan: No Full_join: No Tmp_table: No Tmp_table_on_disk: No
# Filesort: No Filesort_on_disk: No Merge_passes: 0
这一行来自于如下代码:
my_b_printf(&log_file,
"# QC_Hit: %s Full_scan: %s Full_join: %s Tmp_table: %s "
"Tmp_table_on_disk: %s\n" \
"# Filesort: %s Filesort_on_disk: %s Merge_passes: %lu\n",
((thd->query_plan_flags & QPLAN_QC) ? "Yes" : "No"),
((thd->query_plan_flags & QPLAN_FULL_SCAN) ? "Yes" : "No"),
((thd->query_plan_flags & QPLAN_FULL_JOIN) ? "Yes" : "No"),
((thd->query_plan_flags & QPLAN_TMP_TABLE) ? "Yes" : "No"),
((thd->query_plan_flags & QPLAN_TMP_DISK) ? "Yes" : "No"),
((thd->query_plan_flags & QPLAN_FILESORT) ? "Yes" : "No"),
((thd->query_plan_flags & QPLAN_FILESORT_DISK) ? "Yes" : "No"),
这里注意一个处理的技巧,这里query_plan_flags中每一位都代表一个含义,这样存储既能存储足够多的信息同时存储空间也很小,是C/C++中常用的方式。
- QC_Hit: No:是否query cache命中。
- Full_scan: 此处相当于Select_scan 的含义,是否进行了全扫描包括using index。
- Full_join: 此处相当于Select_full_join 的含义,是否被驱动表使用到了索引,如果没有使用到索引则为YES。
考虑如下的执行计划
mysql> desc select *,sleep(1) from testuin a,testuin1 b where a.id1=b.id1;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+
| 1 | SIMPLE | a | NULL | ALL | NULL | NULL | NULL | NULL | 5 | 100.00 | NULL |
| 1 | SIMPLE | b | NULL | ALL | NULL | NULL | NULL | NULL | 5 | 20.00 | Using where; Using join buffer (Block Nested Loop) |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+
2 rows in set, 1 warning (0.00 sec)
如此输出如下:
# QC_Hit: No Full_scan: Yes Full_join: Yes
- Tmp_table:是否使用了临时表,在函数create_tmp_table中设置。
- Tmp_table_on_disk:是否使用了磁盘临时表,如果时候innodb引擎则在create_innodb_tmp_table函数中设置。
- Filesort:是否进行了排序,在函数filesort中设置。
- Filesort_on_disk:是否使用了磁盘排序,同样在函数filesort中设置,但是设置之前会进行是否需要磁盘排序文件的判断。
- Merge_passes: 进行多路归并排序,归并的次数。
Variable query_plan_fsort_passes collects information about file sort passes
acquired during query execution.
6、第六部分InnoDB相关信息
# InnoDB_IO_r_ops: 0 InnoDB_IO_r_bytes: 0 InnoDB_IO_r_wait: 0.000000
# InnoDB_rec_lock_wait: 0.000000 InnoDB_queue_wait: 0.000000
# InnoDB_pages_distinct: 0
这一行来自于如下代码:
char buf[3][20];
snprintf(buf[0], 20, "%.6f", thd->innodb_io_reads_wait_timer / 1000000.0);
snprintf(buf[1], 20, "%.6f", thd->innodb_lock_que_wait_timer / 1000000.0);
snprintf(buf[2], 20, "%.6f", thd->innodb_innodb_que_wait_timer / 1000000.0);
if (my_b_printf(&log_file,
"# InnoDB_IO_r_ops: %lu InnoDB_IO_r_bytes: %llu "
"InnoDB_IO_r_wait: %s\n"
"# InnoDB_rec_lock_wait: %s InnoDB_queue_wait: %s\n"
"# InnoDB_pages_distinct: %lu\n",
thd->innodb_io_reads, thd->innodb_io_read,
buf[0], buf[1], buf[2], thd->innodb_page_access)
== (uint) -1)
- InnoDB_IO_r_ops:物理IO读取次数。
- InnoDB_IO_r_bytes:物理IO读取的总字节数。
- InnoDB_IO_r_wait:物理IO读取等待的时间。innodb 使用 BUF_IO_READ标记为物理io读取繁忙,参考函数buf_wait_for_read。
- InnoDB_rec_lock_wait:等待行锁消耗的时间。在函数que_thr_end_lock_wait中设置。
- InnoDB_queue_wait: 等待进入innodb引擎消耗的时间,在函数srv_conc_enter_innodb_with_atomics中设置。(参考http://blog.itpub.net/7728585/viewspace-2140446/)
- InnoDB_pages_distinct: innodb访问的页数,包含物理和逻辑IO,在函数buf_page_get_gen的末尾通过_increment_page_get_statistics函数设置。
7、第七部分set timestamp
SET timestamp=1527753496;
这一句来自源码,注意源码注释解释就是获取的服务器的当前的时间(current_utime)。
/*
This info used to show up randomly, depending on whether the query
checked the query start time or not. now we always write current
timestamp to the slow log
*/
end= my_stpcpy(end, ",timestamp=");
end= int10_to_str((long) (current_utime / 1000000), end, 10);
if (end != buff)
{
*end++=';';
*end='\n';
if (my_b_write(&log_file, (uchar*) "SET ", 4) ||
my_b_write(&log_file, (uchar*) buff + 1, (uint) (end-buff)))
goto err;
}
七、总结
本文通过查询源码解释了一些关于MySQL慢查询的相关的知识,主要解释了慢查询是基于什么标准进行记录的,同时输出中各个指标的含义,当然这仅仅是我自己得出的结果,如果有不同意见可以一起讨论。
备注栈帧1:
本栈帧主要跟踪Rows_examined的变化及 join->examined_rows++;的变化
(gdb) info b
Num Type Disp Enb Address What
1 breakpoint keep y 0x0000000000ebd5f3 in main(int, char**) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/main.cc:25
breakpoint already hit 1 time
4 breakpoint keep y 0x000000000155b94f in do_select(JOIN*) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_executor.cc:872
breakpoint already hit 5 times
5 breakpoint keep y 0x000000000155ca39 in evaluate_join_record(JOIN*, QEP_TAB*) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_executor.cc:1473
breakpoint already hit 20 times
6 breakpoint keep y 0x00000000019b4313 in ha_innobase::index_first(uchar*)
at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/handler/ha_innodb.cc:9547
breakpoint already hit 4 times
7 breakpoint keep y 0x00000000019b45cd in ha_innobase::rnd_next(uchar*)
at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/handler/ha_innodb.cc:9651
8 breakpoint keep y 0x00000000019b2ba6 in ha_innobase::index_read(uchar*, uchar const*, uint, ha_rkey_function)
at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/handler/ha_innodb.cc:9004
breakpoint already hit 3 times
9 breakpoint keep y 0x00000000019b4233 in ha_innobase::index_next(uchar*)
at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/handler/ha_innodb.cc:9501
breakpoint already hit 5 times
#0 ha_innobase::index_next (this=0x7fff2cbc6b40, buf=0x7fff2cbc7080 "\375\n") at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/handler/ha_innodb.cc:9501
#1 0x0000000000f680d8 in handler::ha_index_next (this=0x7fff2cbc6b40, buf=0x7fff2cbc7080 "\375\n") at /root/mysql5.7.14/percona-server-5.7.14-7/sql/handler.cc:3269
#2 0x000000000155fa02 in join_read_next (info=0x7fff2c007750) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_executor.cc:2660
#3 0x000000000155c397 in sub_select (join=0x7fff2c007020, qep_tab=0x7fff2c007700, end_of_records=false)
at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_executor.cc:1274
#4 0x000000000155bd06 in do_select (join=0x7fff2c007020) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_executor.cc:944
作者微信: