Motivation
• Undo logging: cannot bring backup DB
copies up to date
• Redo logging: need to keep all modified
blocks in memory until commit
• Solution: undo/redo logging
Update ⇒ <Ti, X, v, w>
v = old value, w = new value
ICS 214A: Database Management Systems
Fall 2002
Lecture 18: Redo/Undo Logging
Professor Chen Li
1
ICS214A
Notes 18
Example
Undo/redo logging
T1 :
• Suppose xact T changes DB element X
from v (old value) to w (new value)
– Write log record <T, X, v, w>
Read (A,t); t ← t×2
Write (A,t);
Read (B,t); t ← t×2
Write (B,t);
Output (A);
Output (B);
A=B
• Rule:
A:8 16
B:8 16
– Before modifying any db element X (v w) on
disk changed by xact T, write the update log
record <T,X,v,w> on disk
– <T COMMIT> can precede or follow any dbelement change on disk
ICS214A
2
Notes 18
A:8 16
B:8 16
<T1, start>
<T1, A, 8, 16>
<T1, B, 8, 16>
<T1, commit>
disk
memory
log
Notice: <T1,COMMIT> log record can be before
or after the two output operations
3
ICS214A
Notes 18
4
1
1
Example 1
Undo/redo Recovery
T1 :
• Redo all the committed xacts in the
order earliest-first (forward)
• Undo all the incomplete xacts in the
order latest-first (backward)
Read (A,t); t ← t×2
Write (A,t);
Read (B,t); t ← t×2
Write (B,t);
Output (A);
Output (B);
A:8
B:8
16
16
A=B
failure!
A:8 16
B:8 16
<T1, start>
<T1, A, 8, 16>
<T1, B, 8, 16>
<T1, commit>
disk
memory
log
Recovery: T1 is committed. Redo: write 16 to
both A and B on disk.
ICS214A
Notes 18
Example 2
T1 :
Read (A,t); t ← t×2
Write (A,t);
Read (B,t); t ← t×2
Write (B,t);
Output (A);
Output (B);
A:8
B:8
16
16
failure!
A:8 16
B:8
disk
Notes 18
ICS214A
Notes 18
Example 3
A=B
memory
Recovery: same as before.
ICS214A
5
T1 :
<T1, start>
<T1, A, 8, 16>
<T1, B, 8, 16>
<T1, commit>
Read (A,t); t ← t×2
Write (A,t);
Read (B,t); t ← t×2
Write (B,t);
Output (A);
Output (B);
A:8
B:8
16
16
6
A=B
failure!
A:8 16
B:8
<T1, start>
<T1, A, 8, 16>
<T1, B, 8, 16>
disk
memory
log
Recovery: T1 incomplete. Undo: old values (“8”) of
A and B are written to disk. Write <T1, abort> to
the log.
log
7
ICS214A
Notes 18
8
2
2
Example 4
T1 :
Read (A,t); t ← t×2
Write (A,t);
Read (B,t); t ← t×2
Write (B,t);
Output (A);
Output (B);
A:8 16
B:8 16
T1 :
failure!
<T1, start>
<T1, A, 8, 16>
<T1, B, 8, 16>
A:8
B:8
memory
Recovery: same as before.
ICS214A
Example 5
A=B
disk
9
• Write (and flush) <START CKPT(T1,…,Tk)>
log record, where T1,…,Tk are active xacts
• Flush to disk all modified data elements.
ICS214A
Redo/Undo Log:
<T1, START>
<T1, A, 4, 5>
<T2, START>
<T1, COMMIT>
<T2, B, 9, 10>
– Recall that in redo logging, we only flush those
data elements by committed xacts.
<START CKPT(T2)>
<T2, C, 14, 15>
<T3, START>
<T3, D, 19, 20>
<END CKPT>
<T2, COMMIT>
<T3, COMMIT>
• Write and flush <END CKPT> log record
Notes 18
failure!
<T1, start>
A:8
B:8
Notes 18
10
Example
Nonquiescent checkpoint: flexible
ICS214A
A:8 16
B:8
A=B
disk
memory
log
Recovery: T1 incomplete. Write <T1, abort> to
the log.
log
Notes 18
Read (A,t); t ← t×2
Write (A,t);
Read (B,t); t ← t×2
Write (B,t);
Output (A);
Output (B);
11
ICS214A
Start checkpoint. T2 active
continue, accept new xacts,
make sure B=10 by T2 is on disk,
and A = 5 by T1 is on disk.
end checkpoint
continue
Notes 18
12
3
3
Recovery: example 1
Redo/Undo Log:
<T1, START>
<T1, A, 4,5>
<T2, START>
<T1, COMMIT>
<T2, B, 9, 10>
<START CKPT(T2)>
<T2, C, 14, 15>
<T3, START>
<T3, D, 19, 20>
• T2 and T3 already committed.
• Since we see <END CKPT> first
(backward), T1’s changes must be on disk.
So T1 can be ignored.
• Redo T2 and T3, forward.
– Do NOT need to look at the log records
before <START CKPT>
– Reason: their changes are already on disk
<T2, C, 14, 15>
<T3, START>
<T3, D, 19, 20>
<END CKPT>
ICS214A
T2 committed.
T3 incomplete
Redo T2 (forward): C = 15 (on disk)
Undo T3 (backward): D = 19 (on disk)
<END CKPT>
<T2, COMMIT>
Notes 18
13
Recovery: example 3
<START CKPT(T2)>
•
•
•
•
<T2, C, 14, 15>
<T3, START>
<T3, D, 19, 20>
<T2, COMMIT>
<T3, COMMIT>
Redo/Undo Log:
<T1, START>
<T1, A, 4,5>
<T2, START>
<T1, COMMIT>
<T2, B, 9, 10>
Redo/Undo Log:
<T1, START>
<T1, A, 4,5>
<T2, START>
<T1, COMMIT>
<T2, B, 9, 10>
<START CKPT(T2)>
<END CKPT>
ICS214A
Recovery: example 2
Notes 18
14
Recovery: example 3
• T2 and T3 incomplete
• Undo T3 (backward): D = 19 (on disk)
• Undo T2 (backward): C = 14 and B = 9
(on disk)
• Therefore, once we write the <END
CKPT> log record, we still need to keep
the earliest <Ti, START> log record for
those Ti’s that were active (when we
started the CKPT) and incomplete during
the ckpt.
Redo/Undo Log:
<T1, START>
<T1, A, 4, 5>
<T2, START>
<T1, COMMIT>
<T2, B, 9, 10>
<START CKPT(T2)>
<T2, C, 14, 15>
<T3, START>
<T3, D, 19, 20>
– In this case: <T2, START>
Notes 18
ICS214A
15
ICS214A
• See <START CKPT(T2)> first (backward)
• T2 active and incomplete.
– Undo (backward): C= 14 and B = 9 (on
disk)
• T3 incomplete
– Undo (backward): D = 19 (on disk)
• T1 complete
– Redo (forward): A = 5 (on disk)
Notes 18
16
4
4
Summary
• Deal with system failures
• Undo logging
• Redo logging
• Undo/Redo
• Checkpoints
ICS214A
Notes 18
17
5
5