SLIDE 1
Modern techniques for transaction-
- riented database recovery
Modern techniques for transaction- oriented database recovery - - PowerPoint PPT Presentation
Modern techniques for transaction- oriented database recovery - - PowerPoint PPT Presentation
Modern techniques for transaction- oriented database recovery Caetano Sauer My pleas 1.Demand on-demand recovery 2.Make the log great again Demand on-demand recovery ARIES Recovery info B A B A Transactions Pages requiring requiring
SLIDE 2
SLIDE 3
My pleas
1.Demand on-demand recovery 2.Make the log great again
SLIDE 4
Demand on-demand recovery
SLIDE 5
- 2. REDO scan
A B A B Buffer pool Database Pages requiring redo Transactions requiring undo
- 3. UNDO scan
- 1. Log analysis
ARIES
system mostly unavailable during three phases Recovery info LOG
SLIDE 6
ARIES restart
6
Recovery info
- 2. REDO scan
A B A B Buffer pool Database
- 3. UNDO scan
- 1. Log analysis
system mostly unavailable during three phases LOG
Pages requiring redo Transactions requiring undo
SLIDE 7
Recovery info
Instant recovery
B Buffer pool fix(B) Aborting txn.
(pre-failure)
Running txn.
(post-failure)
lock(f) lock conflict!
same recovery actions, different schedule redo and undo on demand, without waiting for log scans
Pages requiring redo Transactions requiring undo + locks
...
LOG per-page log chains
SLIDE 8
Failure classes
Failure class Loss Typical cause Response Transaction Single-transaction progress Deadlock, constraint violation Rollback System Server process (in-memory state) Software fault, power loss Restart Media Persistent database contents Hardware fault Restore Single page Local integrity Partial writes, wear-out Repair
SLIDE 9
Make the log great again
SLIDE 10
Log DB ?
Dual storage Single storage
Main memory Main memory “... a DBMS is really two DBMSs, one managing the database as we know it and a second one managing the log.”
Michael Stonebraker
write-
- ptimized
read-
- ptimized
Throw away the database!
SLIDE 11
The log as a partitioned B-tree
P0 P1 P2 LSN domain
y z w
B A B A B A
x like an LSM, but with page identifiers and their log records
SLIDE 12
The log as a partitioned B-tree (implementation)
P0 P1 P2 unsorted partition sorted partitions B A page → LSN mapping LSN domain
y z w
Partitioned log index log appends (txn. commit)
(sort low-water mark)
B A B A B A
u x
SLIDE 13
Merging
Before:
B A B A B A z, w x, z
After:
B A B A z, w x, y y, z merge
SLIDE 14
Log records Data structure fetch append Log index reorg. volatile persistent
FineLine:
WAL Log records Database Data structure write read volatile persistent append
Write-ahead logging:
FineLine
SLIDE 15
FineLine recovery
Log records Data structure Indexed log volatile persistent
SLIDE 16
fetch Indexed log volatile persistent
FineLine recovery
SLIDE 17
fetch Indexed log volatile persistent
FineLine recovery
SLIDE 18
fetch Indexed log volatile persistent
FineLine recovery
SLIDE 19
➢ nodes recovered automatically during fetch ➢ volatile structures = in-memory database ➢ no undo, no dirty pages, no checkpoints, no offline log scans
fetch append Indexed log volatile persistent
FineLine
In-memory database
SLIDE 20
In-memory databases Log-structured storage DBMS Buffer mgmt. Physiological logging
SLIDE 21
My pleas
1.Demand on-demand recovery 2.Make the log great again
SLIDE 22
Standing on the shoulders of Giants
SLIDE 23