MySQL:Innodb恢复的学习笔记

本文涉及的产品
云数据库 RDS MySQL Serverless,0.5-2RCU 50GB
简介:

本文只是记录无可读性,供自己参考


MySQL · 引擎特性 · InnoDB 崩溃恢复过程

enum {
    SRV_FORCE_IGNORE_CORRUPT = 1,    /*!< let the server run even if it
                    detects a corrupt page */
    SRV_FORCE_NO_BACKGROUND    = 2,    /*!< prevent the main thread from
                    running: if a crash would occur
                    in purge, this prevents it */
    SRV_FORCE_NO_TRX_UNDO = 3,    /*!< do not run trx rollback after
                    recovery */
    SRV_FORCE_NO_IBUF_MERGE = 4,    /*!< prevent also ibuf operations:
                    if they would cause a crash, better
                    not do them */
    SRV_FORCE_NO_UNDO_LOG_SCAN = 5,    /*!< do not look at undo logs when
                    starting the database: InnoDB will
                    treat even incomplete transactions
                    as committed */
    SRV_FORCE_NO_LOG_REDO = 6    /*!< do not do the log roll-forward
                    in connection with recovery */
};

innodb中的 3个lsn

innodb的lsn和oracle的scn一样,是一个重要的概念。比如

  • 在flush list中正是是使用low lsn作为链表的条件
    参考buf_page_t中的lsn_t oldest_modification;
  • 在checkpoint中记录的也是lsn
    参考宏
#define LOG_CHECKPOINT_NO        0
#define LOG_CHECKPOINT_LSN        8
#define LOG_CHECKPOINT_OFFSET        16
#define LOG_CHECKPOINT_LOG_BUF_SIZE    24
  • 在物理文件中每个块最后的刷新lsn
    参考宏FIL_PAGE_LSN
  • 在写日志落盘的时候也是以lsn为标准的
    参考函数log_write_up_to

实际上lsn就是表示的日志量的字节数,是一个累加的值,在5.7中表现为:

/* Type used for all log sequence number storage and arithmetics */
typedef    ib_uint64_t        lsn_t;

及一个8字节非负的整数。最大值及2的64次方。有了这种物理上概念,lsn很容易换算为当前日志的偏移量。

下面描述一下和检查点相关的几个lsn

  • ibdata第一个块FIL中的lsn(flush lsn):ibdata的26后面8字节是在innodb 干净关闭的时候进行更新的,如果不正常关闭不会进行写入(FIL_PAGE_FILE_FLUSH_LSN)
  • redolog中MLOG_CHECKPOINT的lsn: redolog MLOG_CHECKPOINT lsn的写入是在每次checkpoint的时候同步写入的.干净关闭会更新。
  • redolog header中的lsn:是在每次checkpoint的时候异步写入的在MLOG_CHECKPOINT写入之后.干净关闭会更新。
    我们表示为lsn1/lsn2/lsn3

正常关闭3个lsn是相等的,如果非正常关闭innodb,lsn1不会更新,因此lsn3必然不和lsn1相等,则判定需要进行carsh recovery。

if (checkpoint_lsn != flush_lsn) {

    ......
        if (!recv_needed_recovery) {

            ib::info() << "The log sequence number " << flush_lsn
                << " in the system tablespace does not match"
                " the log sequence number " << checkpoint_lsn
                << " in the ib_logfiles!"; //出现这个警告说明需要恢复了,因为没有干净的关闭数据库,那么flush_lsn一定比checkpoint_lsn小

            if (srv_read_only_mode) {
                ib::error() << "Can't initiate database"
                    " recovery, running in read-only-mode.";
                log_mutex_exit();
                return(DB_READ_ONLY);
            }
            recv_init_crash_recovery(); //初始化
        }
    }

MLOG_CHECKPOINT

  • log_checkpoint 函数由master线程调用,以及关闭数据库的时候调用,不断的向redo log中写入MLOG_CHECKPOINT和MLOG_FILE_NAME

log_group_checkpoint 写入checkpoint信息到log header

#0  mtr_t::commit_checkpoint (this=0x7fff761fb830, checkpoint_lsn=697558445) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/mtr/mtr0mtr.cc:592
#1  0x0000000001ceb4b9 in fil_names_clear (lsn=697558445, do_write=true) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/fil/fil0fil.cc:7067
#2  0x0000000001a521cc in log_checkpoint (sync=true, write_always=false) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1927
#3  0x0000000001b856f2 in srv_master_do_idle_tasks () at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/srv/srv0srv.cc:2596
#4  0x0000000001b85b6b in srv_master_thread (arg=0x0) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/srv/srv0srv.cc:2744
#5  0x0000003f74807aa1 in start_thread () from /lib64/libpthread.so.0
#6  0x0000003f740e8bcd in clone () from /lib64/libc.so.6

正常innodb运行的情况下 checkpoint是由master线程触发。我们知道脏数据通过page clean线程和lru manager线程是在不断写盘的,那么在进行异常重启的的时候我们必须要知道一个恢复的起点,但是这个起点是不能记录在内存中必要固化到磁盘,恢复的时候读取这个点以后的redo进行恢复,而checkpoint就是完成这个事情下面是checkpoint的执行流程。

正常情况下master会每秒进行检查点其作用有:(参考log_checkpoint函数)
1、检查是否有自上次检查点以来的脏数据写盘了
2、如果有则在redo里面会为每个修改过的文件写入MLOG_FILE_NAME,完成后写入一个总的MLOG_CHECKPOINT(参考fil_names_clear函数)
MLOG_FILE_NAME主要记录至上次检查点以来更改过的数据文件
MLOG_CHECKPOINT主要记录检查点的lsn
3、如果有则在redo header中写入相应的检查点信息包含(异步写)(参考log_group_checkpoint函数)

Log sequence number 697794162
Log flushed up to 697794162
Pages flushed up to 697794162
Last checkpoint at 697794153

697794162-697794153 = 9 刚好是MLOG_CHECKPOINT的长度
oldest_lsn <= log_sys->last_checkpoint_lsn + SIZE_OF_MLOG_CHECKPOINT
SIZE_OF_MLOG_CHECKPOINT=9

redo的写入有6个途径:
1、master 线程每秒调用 栈帧(可能是idle可能是active 和检测是否需要插入缓存合并有关)
2、master 线程每秒checkpoint调用 (可能是idle可能是active 和检测是否需要插入缓存合并有关)
3、page clean 线程调用 栈帧 / Force the log to the disk before writing the modified block /
4、主线程commit 调用 栈帧
5、innodb shutdown
6、redo buffer不足

1、master线程调用 栈帧(可能是idle可能是active 和检测是否需要插入缓存合并有关)

#0  log_group_write_buf (group=0x33f29f8, buf=0x7fffa5b38000 "\200\024", len=512, pad_len=0, start_lsn=697764864, new_data_offset=166)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1145
#1  0x0000000001a50f95 in log_write_up_to (lsn=697765068, flush_to_disk=true) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1493
#2  0x0000000001a51163 in log_buffer_sync_in_background (flush=true) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1553
#3  0x0000000001b84bd1 in srv_sync_log_buffer_in_background () at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/srv/srv0srv.cc:2312
#4  0x0000000001b85666 in srv_master_do_idle_tasks () at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/srv/srv0srv.cc:2586
#5  0x0000000001b85b6b in srv_master_thread (arg=0x0) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/srv/srv0srv.cc:2744

2、master 线程checkpoint调用 (可能是idle可能是active 和检测是否需要插入缓存合并有关)

#0  log_group_write_buf (group=0x33f29f8, buf=0x7fffa5a38000 "\200\024\002", len=1024, pad_len=0, start_lsn=697789952, new_data_offset=139)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1145
#1  0x0000000001a50f95 in log_write_up_to (lsn=697790725, flush_to_disk=true) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1493
#2  0x0000000001a52247 in log_checkpoint (sync=true, write_always=false) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1934
#3  0x0000000001b856f2 in srv_master_do_idle_tasks () at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/srv/srv0srv.cc:2596
#4  0x0000000001b85b6b in srv_master_thread (arg=0x0) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/srv/srv0srv.cc:2744

3、page clean 线程调用 栈帧 / Force the log to the disk before writing the modified block /

#0  log_group_write_buf (group=0x33f29f8, buf=0x7fffa5a38000 "\200\024\002", len=13312, pad_len=1024, start_lsn=697778176, new_data_offset=468)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1145
#1  0x0000000001a50f95 in log_write_up_to (lsn=697790015, flush_to_disk=true) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1493
#2  0x0000000001c704c7 in buf_flush_write_block_low (bpage=0x7fffc0cae940, flush_type=BUF_FLUSH_LIST, sync=false)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/buf/buf0flu.cc:1035
#3  0x0000000001c70cea in buf_flush_page (buf_pool=0x33247d8, bpage=0x7fffc0cae940, flush_type=BUF_FLUSH_LIST, sync=false)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/buf/buf0flu.cc:1237
#4  0x0000000001c717f4 in buf_flush_try_neighbors (page_id=..., flush_type=BUF_FLUSH_LIST, n_flushed=0, n_to_flush=25)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/buf/buf0flu.cc:1466
#5  0x0000000001c71b57 in buf_flush_page_and_try_neighbors (bpage=0x7fffc0cae940, flush_type=BUF_FLUSH_LIST, n_to_flush=25, count=0x7fffa02867c0)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/buf/buf0flu.cc:1558
#6  0x0000000001c72862 in buf_do_flush_list_batch (buf_pool=0x33247d8, min_n=25, lsn_limit=18446744073709551615)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/buf/buf0flu.cc:1846
#7  0x0000000001c72cb6 in buf_flush_batch (buf_pool=0x33247d8, flush_type=BUF_FLUSH_LIST, min_n=25, lsn_limit=18446744073709551615)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/buf/buf0flu.cc:1926
#8  0x0000000001c73104 in buf_flush_do_batch (buf_pool=0x33247d8, type=BUF_FLUSH_LIST, min_n=25, lsn_limit=18446744073709551615, n_processed=0x7fffa0286938)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/buf/buf0flu.cc:2071
#9  0x0000000001c734ee in buf_flush_lists (min_n=25, lsn_limit=18446744073709551615, n_processed=0x7fffa02869c8)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/buf/buf0flu.cc:2202
#10 0x0000000001c76a97 in buf_flush_page_cleaner_coordinator (arg=0x0) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/buf/buf0flu.cc:3362

4、主线程commit 调用 栈帧

#0  log_group_write_buf (group=0x33f29f8, buf=0x7fffa5a38000 "\200\024\002", len=2560, pad_len=0, start_lsn=697762816, new_data_offset=230)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1145
#1  0x0000000001a50f95 in log_write_up_to (lsn=697765030, flush_to_disk=true) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1493
#2  0x0000000001a51087 in log_buffer_flush_to_disk (sync=true) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:1524
#3  0x00000000019a9157 in innobase_flush_logs (hton=0x2e9fdd0, binlog_group_flush=true)
    at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/handler/ha_innodb.cc:4407
#4  0x0000000000f65893 in flush_handlerton (thd=0x0, plugin=0x7ffff03588e8, arg=0x7ffff0358944) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/handler.cc:2606
#5  0x00000000015d7716 in plugin_foreach_with_mask (thd=0x0, func=0xf65835 <flush_handlerton(THD*, plugin_ref, void*)>, type=1, state_mask=4294967287, 
    arg=0x7ffff0358944) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_plugin.cc:2318
#6  0x0000000000f658ef in ha_flush_logs (db_type=0x0, binlog_group_flush=true) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/handler.cc:2617
#7  0x000000000185733d in MYSQL_BIN_LOG::process_flush_stage_queue (this=0x2e02c80, total_bytes_var=0x7ffff0358a88, rotate_var=0x7ffff0358a87, 
    out_queue_var=0x7ffff0358a78) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/binlog.cc:8541
#8  0x000000000185899f in MYSQL_BIN_LOG::ordered_commit (this=0x2e02c80, thd=0x7fff2c000b70, all=false, skip_commit=false)
    at /root/mysql5.7.14/percona-server-5.7.14-7/sql/binlog.cc:9189
#9  0x000000000185700c in MYSQL_BIN_LOG::commit (this=0x2e02c80, thd=0x7fff2c000b70, all=false) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/binlog.cc:8440
#10 0x0000000000f63df8 in ha_commit_trans (thd=0x7fff2c000b70, all=false, ignore_global_read_lock=false)

入口函数
innobase_start_or_create_for_mysql

recv_recovery_from_checkpoint_start

1、此函数需要输入flush的lsn及脏数据写入到的位置

这个数据从FIL_PAGE_FILE_FLUSH_LSN中读取,这个值
只在ibdata的第一个page有效其他均为0,他的读取来自于
函数Datafile::validate_first_page,
其中
*flush_lsn = mach_read_from_8(
m_first_page + FIL_PAGE_FILE_FLUSH_LSN);


其写入由函数  正常shutdown才会写入,非正常关闭不会写入,正常运行检查点也不会写入
fil_write_flushed_lsn 写入
mach_write_to_8(buf + FIL_PAGE_FILE_FLUSH_LSN, lsn);

栈帧
    #0  fil_write_flushed_lsn (lsn=696973727) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/fil/fil0fil.cc:1946
    #1  0x0000000001a538a7 in logs_empty_and_mark_files_at_shutdown () at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/log/log0log.cc:2464
    #2  0x0000000001b915a9 in innobase_shutdown_for_mysql () at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/srv/srv0start.cc:2803
    #3  0x00000000019a8ffc in innobase_end (hton=0x2e9edd0, type=HA_PANIC_CLOSE) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/innobase/handler/ha_innodb.cc:4360
    #4  0x0000000000f62621 in ha_finalize_handlerton (plugin=0x3015cd0) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/handler.cc:813
    #5  0x00000000015d3d25 in plugin_deinitialize (plugin=0x3015cd0, ref_check=true) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_plugin.cc:995
    #6  0x00000000015d410e in reap_plugins () at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_plugin.cc:1077
    #7  0x00000000015d6073 in plugin_shutdown () at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_plugin.cc:1845
    #8  0x0000000000ebf7eb in clean_up (print_message=true) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/mysqld.cc:1336
    #9  0x0000000000ec701b in mysqld_main (argc=98, argv=0x2e9cf08) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/mysqld.cc:5386
    #10 0x0000000000ebd604 in main (argc=10, argv=0x7fffffffe458) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/main.cc:25

2、此函数首先建立红黑树用于恢复并且做 force recovery判断

   (srv_force_recovery >= SRV_FORCE_NO_LOG_REDO) 
   然后调用 
   recv_find_max_checkpoint 此函数就是找到最大的checkpoint lsn、redo文件,以及field 就是相对于一个日志而讲所在的offset set (512*1 or 512*3)_
   读取如下:
                   group->lsn = mach_read_from_8(
                  buf + LOG_CHECKPOINT_LSN);
                group->lsn_offset = mach_read_from_8(
                  buf + LOG_CHECKPOINT_OFFSET);
                checkpoint_no = mach_read_from_8(
                  buf + LOG_CHECKPOINT_NO);

3、recv_group_scan_log_recs

   循环读取64K(RECV_SCAN_SIZE)日志到redo buffer(log_group_read_log_seg)
   recv_scan_log_recs
     本函数将64K 的redo 通过每个512 bytes block大小循环加入到(recv_sys_add_to_parsing_buf) 扫描到parse buffer 
     代码片段拷贝
          ut_memcpy(recv_sys->buf + recv_sys->len,
              log_block + start_offset, end_offset - start_offset);
            
           当每次扫描了64K*80的日志量的时候会输出,级5M
           ib::info() << "Doing recovery: scanned up to"
                " log sequence number " << scanned_lsn;
     并且进行分析对这加入到 parse buffer的日志进行分析(recv_parse_log_recs) /其中parse buffer 为2M(宏RECV_PARSING_BUF_SIZE大小)
        recv_parse_log_recs函数对每次加入 parse buffer的64k进行分析主要分析是MLOG_SINGLE_REC_FLAG还是MLOG_MULTI_REC_END 同时确认MLOG_MULTI_REC_END
        是否完整,完成后加入到hash table,函数是调用recv_parse_log_rec进行 (type spaceid page_no data)的剥离。 

那么recv_group_scan_log_recs为第一层循环,循环扫描64K日志到redo buffer
然后调用recv_scan_log_recs以block单位(512b)大小循环加入到parse buffer,这是通过函数recv_sys_add_to_parsing_buf完成
等到64k加入到parse buffer后调用recv_parse_log_recs函数对每次加入 parse buffer的64k进行分析,分析是循环以record为单位的,主要分析MLOG_SINGLE_REC_FLAG还是MLOG_MULTI_REC_END,同时确保MLOG_MULTI_REC_END记录是完整的。完成后recv_parse_log_recs函数 还要将其加入到hash table函数为recv_add_to_hash_table,本函数是通过recv_parse_log_rec进行 (type spaceid page_no data)的剥离。

recv_apply_hashed_log_recs 最后完成分析完成后的日志应用,从hash table中

作者微信:
微信.jpg

相关实践学习
基于CentOS快速搭建LAMP环境
本教程介绍如何搭建LAMP环境,其中LAMP分别代表Linux、Apache、MySQL和PHP。
全面了解阿里云能为你做什么
阿里云在全球各地部署高效节能的绿色数据中心,利用清洁计算为万物互联的新世界提供源源不断的能源动力,目前开服的区域包括中国(华北、华东、华南、香港)、新加坡、美国(美东、美西)、欧洲、中东、澳大利亚、日本。目前阿里云的产品涵盖弹性计算、数据库、存储与CDN、分析与搜索、云通信、网络、管理与监控、应用服务、互联网中间件、移动服务、视频服务等。通过本课程,来了解阿里云能够为你的业务带来哪些帮助 &nbsp; &nbsp; 相关的阿里云产品:云服务器ECS 云服务器 ECS(Elastic Compute Service)是一种弹性可伸缩的计算服务,助您降低 IT 成本,提升运维效率,使您更专注于核心业务创新。产品详情: https://www.aliyun.com/product/ecs
相关文章
|
1月前
|
存储 关系型数据库 MySQL
MySQL InnoDB数据存储结构
MySQL InnoDB数据存储结构
|
1月前
|
存储 缓存 关系型数据库
MySQL的varchar水真的太深了——InnoDB记录存储结构
varchar(M) 能存多少个字符,为什么提示最大16383?innodb怎么知道varchar真正有多长?记录为NULL,innodb如何处理?某个列数据占用的字节数非常多怎么办?影响每行实际可用空间的因素有哪些?本篇围绕innodb默认行格式dynamic来说说原理。
834 6
MySQL的varchar水真的太深了——InnoDB记录存储结构
|
2月前
|
存储 缓存 关系型数据库
MySQL - 存储引擎MyISAM和Innodb
MySQL - 存储引擎MyISAM和Innodb
|
15天前
|
存储 关系型数据库 MySQL
MySQL引擎对决:深入解析MyISAM和InnoDB的区别
MySQL引擎对决:深入解析MyISAM和InnoDB的区别
31 0
|
2月前
|
关系型数据库 MySQL
MySQL学习笔记
MySQL学习笔记
|
2月前
|
安全 关系型数据库 MySQL
某教程学习笔记(一):09、MYSQL数据库漏洞
某教程学习笔记(一):09、MYSQL数据库漏洞
19 0
|
2月前
|
存储 关系型数据库 MySQL
《高性能Mysql》学习笔记(二)
《高性能Mysql》学习笔记(二)
136 0
|
2月前
|
存储 SQL 关系型数据库
《高性能Mysql》学习笔记(一)
《高性能Mysql》学习笔记(一)
94 0
|
2月前
|
存储 SQL 关系型数据库
Mysql专栏 - mysql、innodb存储引擎、binlog的工作流程
Mysql专栏 - mysql、innodb存储引擎、binlog的工作流程
80 0
|
3月前
|
存储 算法 关系型数据库
MySQL相关(八)- innodb行级锁深入剖析
MySQL相关(八)- innodb行级锁深入剖析
45 0