InnoDB MVCC Architecture (by 권건우)

© Copyrights 2001~2017 EXEM CO.,LTD. All Rights Reserved.
InnoDB MVCC Architecture
엑셈 | 권건우 상무

02. MySQL/InnoDB MVCC Architecture
01. MySQL/InnoDB Undo Architecture
Table of Agenda

MySQL Deep Internal
MySQL Deep InternalMVCC Architecture 발전
1970
1978
1981
E.F. Codd
PHILIP A. BERNSTEIN AND NATHAN GOODMAN
- the data independence objective
- the communicability objective
- the set-processing objective
Our approach is quite different. We regard an object that is modifiable as a
sequence of immutable versions. each version is the state of the object after
an update is made to the object.
Multiversion Concurrency Control
1984 The first shipping, commercial database software product featuring MVCC
was Digital's VAX Rdb/ELN. The second was InterBase, both of which are still
active, commercial products.
Jim Starkey
1986 Oracle version 6 에서 Rollback Segment 도입
Bob Miner
1999 Mikheev Vadim B.
PostgreSQL 6.5 에 MVCC 아키텍쳐를 도입함
David P. Reed

MySQL Deep Internal
MySQL Deep InternalMVCC Architecture의 두가지 접근법
• PostgreSQL, Firebird/Interbase, SQL Server 해당
• 데이터베이스에 다중 버전의 레코드를 저장
• 더 이상 필요하지 않을 때 모아둔 레코드를 버림
접근법 1
• Oracle, MySQL/InnoDB 해당
• 최신 버전의 데이터만 데이터베이스 내에 저장
• 언두를 이용하여 이전 버전의 데이터를 재구성
접근법 2

MySQL/InnoDB Undo Architecture

MySQL Deep Internal
MySQL Deep Internal
slot 0 4
slot 1 5
…
slot 1023
(1,5) Undo log
06d1 Rec001, A
T1.ibd space f9
Page
0
Page
1
Page
2
Page
3
Page
4
…
Page
64
MySQL/InnoDB Undo Architecture
ibdata1 space 0
Page
0
Page
1
Page
2
Page
3
Page
4
Page
5
…
(0,5) TRX_SYS Rseg 0 (0,6)
Rseg 1
…
Rseg 33 (1,3)
Rseg 127
undo001 space 1
Page
0
Page
1
Page
2
Page
3
Page
4
Page
5
… Page
8
ibtmp1
Rseg
1
Rseg
2
Rseg
3
Rseg
4
…
Rseg
32
Page (f9,3)
Rec001, A
(TRX_ID 0x778c,
Roll_PTR (5,06d1))
(1,3) Rollback Segment
System Tablespace
User Tablespace
Undo Tablespace
Temp Tablespace

MySQL Deep Internal
MySQL Deep Internal
slot 0 4
slot 1
…
slot 1023
(1,4) Undo log
0110 Rec001, A
T1.ibd space f9
Page
0
Page
1
Page
2
Page
3
Page
4
…
Page
64
Page (f9,3)
Rec001, A
(TRX_ID 0x778b,
Roll_PTR (4,0110))
MySQL/InnoDB Undo Architecture (select 시)
0x7f0208608000
Data Page (f9,3)
Rec001, A
(TRX_ID 0x778b,
Roll_PTR (4,0110))
Insert A 후, Select * from t1;
TRX_SYS
max_trx_id = 0x778c
rw_trx_list->count = 0
mvcc->m_views->count = 1
Rseg 0 (0,6)
Rseg 1
…
Rseg 33 (1,3)
Rseg 127
ibdata1 space 0
Page
0
Page
1
…
Page
5
…
(0,5) TRX_SYS
undo001 space 1
Page
0
Page
1
Page
2
Page
3
Page
4
Page
5
… Page
8
Buffer Pool
User Tablespace
System Tablespace
Undo Tablespace

MySQL Deep Internal
MySQL Deep Internal
slot 0 4
slot 1 5
…
slot 1023
(1,5) Undo log
06d1 Rec001, A
T1.ibd space f9
Page
0
Page
1
Page
2
Page
3
Page
4
…
Page
64
Page (f9,3)
Rec001, B
(TRX_ID 0x778c,
Roll_PTR (5,06d1))
MySQL/InnoDB Undo Architecture (update 시)
0x7f0208608000
Data Page (f9,3)
Rec001, B
(TRX_ID 0x778c,
Roll_PTR (5,06d1))
A를 B로 update (Update t1 Set c2 = B;)
TRX_SYS
max_trx_id = 0x778d
Rseg 0 (0,6)
Rseg 1
…
Rseg 33 (1,3)
Rseg 127
ibdata1 space 0
Page
0
Page
1
…
Page
5
…
(0,5) TRX_SYS
undo001 space 1
Page
0
Page
1
Page
2
Page
3
Page
4
Page
5
… Page
8
0x7f0208608000
Data Page (f9,3)
Rec001, A
(TRX_ID 0x778b,
Roll_PTR (4,0110))
TRX_ID : 0x778c
addr=0x1b74a750
page_no=5
offset=0x6d1
Buffer Pool
User Tablespace
System Tablespace
Undo Tablespace

MySQL/InnoDB MVCC Architecture
1) MVCC
2) Readview

MySQL Deep Internal
MySQL Deep Internal
slot 0 4
slot 1 5
…
slot 1023
(1,5) Undo log
MySQL/InnoDB MVCC Architecture
TRX_SYS
max_trx_id = 0x778d
Rseg 0 (0,6)
Rseg 1
…
Rseg 33 (1,3)
Rseg 127
ibdata1 space 0
Page
0
Page
1
…
Page
5
…
(0,5) TRX_SYS
undo001 space 1
Page
0
Page
1
Page
2
Page
3
Page
4
Page
5
… Page
8
System Tablespace
Undo Tablespace
Readview Readview Readview Readview
TRX_STRUCTURE TRX_STRUCTURE TRX_STRUCTURE TRX_STRUCTURE

MySQL Deep Internal
MySQL Deep InternalMySQL/InnoDB MVCC Architecture
Rseg 0 (0,6)
Rseg 1
…
Rseg 33 (1,3)
Rseg 127
ibdata1 space 0
Page
0
Page
1
…
Page
5
…
System Tablespace
slot 0 4
slot 1
…
slot 1023
(1,4) Undo log(1,3) Rollback Segment
Undo Tablespace
Undo Page (1,4)
Rec001, A
(TRX_ID 0x778b)
1. 초기상태
Insert into t1
values(‘Rec001’,’A’);
Commit ;
Session 0
T1.ibd space f9
Page
0
…
Page
3
Page
4
…
Page
64
Page (f9,3)
Rec001, A
(TRX_ID 0x778b,
Roll_PTR (4,110))
Buffer Pool
User Tablespace
TRX_SYS
max_trx_id = 0x778c

MySQL Deep Internal
Session 0
Rseg 0 (0,6)
Rseg 1
…
Rseg 33 (1,3)
Rseg 127
ibdata1 space 0
Page
0
Page
1
…
Page
5
…
(0,5) TRX_SYS
System Tablespace
T1.ibd space f9
Page
0
…
Page
3
Page
4
…
Page
64
Page (f9,3)
Rec001, A
(TRX_ID 0x778b,
Roll_PTR (4,110))
Buffer Pool
User Tablespace
slot 0 4
slot 1
…
slot 1023
Undo Tablespace
TRX_SYS
max_trx_id = 0x778c
Select * from t1;
Rec001, A
Readview :
0x35dd108
Session 1
Readview : 0x35dd108
m_low_limit_no = 0x778c
m_low_limit_id = 0x778c
m_up_limit_id = 0x778c
m_ids->m_size = 0
0x7f0208608000
Data Page (f9,3)
Rec001, A
(TRX_ID 0x778b,
ROLL_PTR (4,110)
2. 세션1 Select

MySQL Deep Internal
Rseg 0 (0,6)
Rseg 1
…
Rseg 33 (1,3)
Rseg 127
ibdata1 space 0
Page
0
Page
1
…
Page
5
…
(0,5) TRX_SYS
System Tablespace
T1.ibd space f9
Page
0
…
Page
3
Page
4
…
Page
64
Page (f9,3)
Rec001, B
(TRX_ID 0x778c)
Buffer Pool
User Tablespace
slot 0 4
slot 1 5
…
slot 1023
Undo Tablespace
TRX_SYS
max_trx_id = 0x778d
m_ids->m_size = 0
TRX ID 0x778c
addr=0x1b74a750
0x7f02083b4000
Undo Page (1,5)
Rec001, A
(TRX_ID 0x778b)
0x7f0208608000
Data Page (f9,3)
Rec001, B
(TRX_ID 0x778c,
ROLL_PTR (5,6d1))
Undo Page (1,5)
(0x6d1)
Rec001, A
(TRX_ID 0x778b)
(1,5) Undo log
Update t1
Set c2 = B;
Session 0
Select * from t1;
Rec001, A
Readview :
0x35dd108
Session 1
3. 세션0 Update

MySQL Deep Internal
Rseg 0 (0,6)
Rseg 1
…
Rseg 33 (1,3)
Rseg 127
ibdata1 space 0
Page
0
Page
1
…
Page
5
…
(0,5) TRX_SYS
System Tablespace
T1.ibd space f9
Page
0
…
Page
3
Page
4
…
Page
64
Page (f9,3)
Rec001, B
(TRX_ID 0x778c)
Buffer Pool
User Tablespace
slot 0 4
slot 1 5
…
slot 1023
Undo Tablespace
TRX_SYS
max_trx_id = 0x778e
m_views->start = 0x35dd068
m_low_limit_no = 0x778e
m_low_limit_id = 0x778e
m_up_limit_id = 0x778e
m_ids->m_size = 0
0x7f02083b4000
Undo Page (1,5)
Rec001, A
(TRX_ID 0x778b)
0x7f0208608000
Data Page (f9,3)
Rec001, B
(TRX_ID 0x778c,
ROLL_PTR (5,6d1))
Undo Page (1,5)
(0x6d1)
Rec001, A
(TRX_ID 0x778b)
(1,5) Undo log
Session 0
Select * from t1;
Rec001, A
Readview :
0x35dd108
Session 1
4. 세션2 Select
Select * from t1;
Rec001, B
Readview :
0x35dd068
Session 2
m_ids->m_size = 0

MySQL Deep Internal
Rseg 0 (0,6)
Rseg 1
…
Rseg 33 (1,3)
Rseg 127
ibdata1 space 0
Page
0
Page
1
…
Page
5
…
(0,5) TRX_SYS
System Tablespace
T1.ibd space f9
Page
0
…
Page
3
Page
4
…
Page
64
Page (f9,3)
Rec001, C
(TRX_ID 0x778e)
Buffer Pool
User Tablespace
slot 0 4
slot 1 5
…
slot 1023
Undo Tablespace
TRX_SYS
max_trx_id = 0x778e
m_views->start = 0x35dd068
m_ids->m_size = 0
(1,5) Undo log
Update t1
set c2 = C;
Session 0
Select * from t1;
Rec001, A
Readview :
0x35dd108
Session 1
5. 세션0 Update
Select * from t1;
Rec001, B
Readview :
0x35dd068
Session 2
m_ids->m_size = 0
TRX ID 0x778e
addr=0x1b74a750
0x7f02083b4000
Undo Page (1,5)
(0x6d1)
Rec001, A
(TRX_ID 0x778b)
0x7f02083b4000
Undo Page (1,5)
(0x7be)
Rec001, B
(TRX_ID 0x778c
ROLL_PTR (5,6d1))
0x7f0208608000
Data Page (f9,3)
Rec001, B
(TRX_ID 0x778c,
ROLL_PTR (5,6d1))
Undo Page (1,5)
0x6d1
Rec001, A
(TRX_ID 0x778b)
Undo Page (1,5)
(0x7be)
Rec001, B
(TRX_ID 0x778c
ROLL_PTR (5,6d1))

MySQL Deep Internal
Rseg 0 (0,6)
Rseg 1
…
Rseg 33 (1,3)
Rseg 127
ibdata1 space 0
Page
0
Page
1
…
Page
5
…
(0,5) TRX_SYS
System Tablespace
T1.ibd space f9
Page
0
…
Page
3
Page
4
…
Page
64
Page (f9,3)
Rec001, C
(TRX_ID 0x778e)
Buffer Pool
User Tablespace
slot 0 4
slot 1 5
…
slot 1023
Undo Tablespace
TRX_SYS
max_trx_id = 0x7790
m_views->start = 0x35ddfc8
Readview : 0x35dcfc8
m_low_limit_no = 0x7790
m_low_limit_id = 0x7790
m_up_limit_id = 0x7790
m_ids->m_size = 0
(1,5) Undo log
Session 0
Select * from t1;
Rec001, A
Readview :
0x35dd108
Session 1
6. 세션3 Select
Select * from t1;
Rec001, B
Readview :
0x35dd068
Session 2
m_ids->m_size = 0
0x7f02083b4000
Undo Page (1,5)
(0x6d1)
Rec001, A
(TRX_ID 0x778b)
0x7f02083b4000
Undo Page (1,5)
(0x7be)
Rec001, B
(TRX_ID 0x778c
ROLL_PTR (5,6d1))
0x7f0208608000
Data Page (f9,3)
Rec001, C
(TRX_ID 0x778e,
ROLL_PTR (5,7be))
Undo Page (1,5)
0x6d1
Rec001, A
(TRX_ID 0x778b)
Undo Page (1,5)
(0x7be)
Rec001, B
(TRX_ID 0x778c
ROLL_PTR (5,6d1))
Select * from t1;
Rec001, C
Readview :
0x35ddfc8
Session 2
m_ids->m_size = 0

MySQL Deep Internal
MySQL Deep InternalReadview
정의 및 목적
• 트랜잭션이 발생하면, COMMIT 시점이 아닌 트랜잭션 시작 시점의 TRX_ID 가 저장됨
• 쿼리 시작시점의 액티브 트랜잭션들의 집합 (COMMIT/ROLLBACK 하지 않은 상태)
• MVCC 구현에 반드시 필요
Component Description
low_limit_no
- 해당 값을 Purge thread가 참조하여 불필요한 Undo log들을 처리하는 purge 작업에 사용된다.
- low_limit_no = Readview 생성시 trx_sys의 max_trx_id값
low_limit_id
- low_limit_id보다 크거나 같은 TRX_ID (trx_id >= low_limit_id) 는 해당 readview에서 조회 안 됨
- low_limit_id = Readview 생성시 trx_sys의 max_trx_id값
up_limit_id
- up_limit_id(trx_id < up_limit_id) 보다 작은 트랜잭션
TRX_ID 는 해당 readview에서 모두 조회됨
- up_limit_id = readview 생성시, active 트랜잭션 중 가장 작은 TRX_ID
trx_ids[] - 모든 액티브 트랜잭션 id로 구성된 그룹
view_list - Readview list (doubly linked list)

MySQL Deep Internal
New TRX Structure
mutex
max_trx_id = 0x98fd
rw_trx_list
count = 3
start = 0x4fe4fae0
mvcc->m_views
count = 3
start = 0x25b9f08
…
trx_sys Trx linked list
trx id = 0x98fc
update_undo :
page_no = 6
offset = 0x86c
trx_list
prev = 0x00
next = 0x4fe4fe70
…
trx id = 0x98fb
update_undo :
page_no = 7
offset = 0x370
trx_list
prev = 0x4fe4fae0
next = 0x4fe4f750
…
trx id = 0x98f8
update_undo :
page_no = 5
offset = 0x2c4
trx_list
prev = 0x4fe4fe70
next = 0x00
…
trx0 0x4fe4fae0 trx1 0x4fe4fe70 trx2 0x4fe4f750
New Readview
Readview linked list
m_low_limit_no = 0x98fd
m_low_limit_id = 0x98fd
m_up_limit_id = 0x98f8
m_view_list
prev = 0x0
next = 0x25b9fa8
m_ids->m_size = 3
m_ids[0] = 0x98f8
m_ids[1] = 0x98fb
m_ids[2] = 0x98fc
readview0 0x25b9f08
m_low_limit_no = 0x98fc
m_low_limit_id = 0x98fc
m_view_list
prev = 0x25b9f08
next = 0x25ba048
m_ids->m_size = 2
m_ids[0] = 0x98f8
m_ids[1] = 0x98fb
readview1 0x25b9fa8
m_low_limit_no = 0x98fb
m_low_limit_id = 0x98fb
m_view_list
prev = 0x25b9fa8
next = 0x0
m_ids->m_size = 1
m_ids[0] = 0x98f8
readview2 0x25ba048

MySQL Deep Internal
1. t1 Table 생성
create table t1 (id int primary key, comment
char(50), filler char(2)) engine = innodb;
2. Secondary index 생성
create index test_idx on test(comment);
T1.ibdspace173
Page
0
Page
1
Page
2
Page
3
Page
4
Page
5
…
UNDO001space1
Page
0
Page
1
Page
2
Page
3
Page
4
Page
5
…
Secondary Index
source : InnoDBTransaction/Lock/Multi-version analysis, 网易杭研-何登成
Clustering Index
Page 4

MySQL Deep Internal
3. TRX_ID가 0x96f6 일때, t1에 1건 Insert
insert into t.test values (65, 'aaa',' ') ;
T1.ibdspace173
Page
0
Page
1
Page
2
Page
3
Page
4
Page
5
…
Deleted: 0
65
DB_TRX_ID (0x96f6)
DB_ROLL_PTR (4,110)
aaa
Deleted: 0
aaa
PK: 65
UNDO001space1
Page
0
Page
1
Page
2
Page
3
Page
4
Page
5
…
Insert Undo
aaa
PK: 65
Secondary Index
Clustering Index
Page 4

MySQL Deep Internal
4. Primary key 업데이트
update t.test set id = 67 where id = 65 ;
T1.ibdspace173
Page
0
Page
1
Page
2
Page
3
Page
4
Page
5
…
Deleted: 1
65
DB_TRX_ID (0x96fc)
DB_ROLL_PTR (7,110)
aaa
Deleted: 1
aaa
PK: 65
UNDO001space1
Page
0
Page
1
Page
2
Page
3
Page
4
…
Page
7
Update Undo
aaa
PK: 65
• 새로운 레코드 (67, aaa)가 추가됨
(Insert 발생하는 것처럼 Insert Undo에 저장)
• 이전 레코드 (65, aaa)는 삭제 표시됨 (삭제 flag, trx_id, roll_ptr를
설정하고, before-image를 Update Undo에 저장)
• Secondary Index 페이지에서도 레코드 (aaa, 65) 는 삭제 설정하고
레코드 (aaa,67)가 추가됨
Deleted: 0
67
DB_TRX_ID (0x96fc)
DB_ROLL_PTR (4,11d)
aaa
Deleted: 0
aaa
PK: 67
Insert Undo
aaa
PK: 67
Secondary Index
Clustering Index
Page 4 Page 7

MySQL Deep Internal
5. Secondary key 업데이트
update t.test set comment = 'ccc' where id = 67 ;
T1.ibdspace173
Page
0
Page
1
Page
2
Page
3
Page
4
Page
5
…
Deleted: 1
65
DB_TRX_ID (0x96fc)
DB_ROLL_PTR (7,110)
aaa
Deleted: 1
aaa
PK: 65
UNDO001space1
Page
0
Page
1
Page
2
Page
3
Page
4
…
Page
7
Update Undo
aaa
PK: 65
• 레코드 (67,aaa)를 레코드 (67,ccc)로 변경하고, before-image를
Update Undo에 저장
• Secondary Index 페이지에서도 레코드 (aaa,67)는 삭제 설정되고,
레코드 (ccc,67)가 추가됨.
Deleted: 0
67
DB_TRX_ID (0x9700)
DB_ROLL_PTR (7,224)
ccc
Deleted: 1
aaa
PK: 67
Update Undo
aaa
PK: 67
Deleted: 0
ccc
PK: 67
Secondary Index
Clustering Index
Page 7

MySQL Deep Internal
Session 0 (DML)
Session 1 (DML)
Session 2 (DML)
Session 3 (Read)
Session 4 (Read)
Session 5 (Read)
UPDATE
(67, ‘ccc’)
INSERT
(65, ‘aaa’)
UPDATE
(67, ‘aaa’)
0x96f5
0x96f6
0x96fb 0x96fc 0x97000x96ff 0x97060x96fd 0x97090x9705
READVIEW 0 0x25c0ee8
UP:0x96f5 ~ LOW:0x96fd
Trx:0x96f5, 0x96fb, 0x96fc
Select * from test (65,aaa)
READVIEW 1 0x25c0e48
UP:0x96fb ~ LOW:0x9705
Trx:0x96fb, 0x96ff, 0x9700
Select * from test (67,aaa) READVIEW 2 0x25c0da8
UP:0x9706~ LOW:0x9709
Trx:0x9706
Select * from test (67,ccc)
TX 0x96f5
TX 0x96fb
TX 0x96fc
TX 0x9700
TX 0x96ff
TX 0x9706

MySQL Deep Internal
MySQL Deep Internal
Readview
select * from t1;
case up_limit_id low_limit_id trx_ids [ ] query result
Readview0 Primary Key 변경 0x96f5 0x96fd 0x96f5, 0x96fb, 0x96fc (65, aaa)
Readview1 Secondary Key 변경 0x96fb 0x9705 0x96fb, 0x96ff, 0x9700 (67, aaa)
Readview2
해당 테이블에
트랜잭션이 없을 경우
0x9706 0x9709 0x9706 (67, ccc)

NAVER http://cafe.naver.com/playexem
ITPUB (中) http://blog.itpub.net/31135309/
Wordpress https://playexem.wordpress.com/
Slideshare http://www.slideshare.net/playexem
교육 문의 edu@ex-em.com
EXEM Research & Contents Team
Youtube https://www.youtube.com/channel/UC5wKR
_-A0eL_Pn_EMzoauJg
Tudou (中) http://www.tudou.com/home/maxgauge/

감사합니다

InnoDB MVCC Architecture (by 권건우)

Recommended

More Related Content

What's hot (20)

Similar to InnoDB MVCC Architecture (by 권건우) (20)

Recently uploaded (20)

InnoDB MVCC Architecture (by 권건우)