Crash Recovery

1
Crash Recovery

• Chapter 22.3

2
Buffers

• Main memory locations which are used to transfer
  data to and from secondary storage
• Dirty bit and dirty page
• Dirty bit and pinCount are initially set to 0
• Replacement strategies
• First-in-first-out (FIFO)
• Least recently used (LRU)

3
Buffer management

• A steal policy buffer manager writes a buffer to
  disk before a transaction commits
• Alternative policy no-steal
• A force policy all pages updated by a
  transaction are immediately written to disk when
  the transaction commits.
• Alternative policy no-force

4
Recovery Facilities

• Backup mechanism
• Logging
• Checkpoints
• Recovery algorithms

5
The Log

• History of actions executed by the DBMS
• Stable storage
• Log tail
• Actions are recorded in chronological order

6
Log Sequence Number

• Monotonically increasing order
• Every page (object) contains pageLSN
• PageLSN - the most recent LSN from the log

7
Actions recorded to the Log

• Start
• Updating an object
• Update record is appended to the log tail
• pageLSN is set to the current LSN
• Commit
• Commit record is appended to the log
• Log tail is written to stable storage

8
Actions recorded to the Log (cont)

• Abort
• Abort log record is appended to the log
• Undo is initiated
• End
• Undoing updates
• After the action described in update log record
  is undone, a compensation log record (CLR) is
  appended to the log.

9
Update Log Record

• Common fields for all log records
• LSN, prevLSN, transID, type
• Fields for update records only
• PageID
• Length of changes in bytes
• Offset (start byte)
• Before-image
• After-image

10
Compensation Log Record

• Is written just before the action recorded in
  Update log record is undone.
• The same fields as for Update record
• Field undonextLSN
• LSN from prevLSN of correspondent Update log
  record

11
The Write-Ahead Log Protocol

• All Update log records must be written to stable
  storage first
• Log records
• Transaction records
• Systems records

12
Checkpoints

• A point of synchronization between database and
  transaction log
• All buffers are force-written to secondary
  storage
• Checkpoints in AREIS
• Begin_checkpoint record
• End_checkpoint record
• Master record holds LSN of the begin_checkpoint
  record

13
Recovery Using Deferred Update

• Updates are not written to the database until the
  transaction T commits
• T starts
• Start log record
• T commits
• Commit log record
• Log tail is force-written to disk
• Updates are implemented

14
Recovery Using Deferred Update (cont)

• T aborts
• Abort log record
• Ignore the log records related to T
• Recovery after the crash
• Most recent checkpoint
• Transactions with Start and Commit log records
  are redone
• Transactions with Start and Abort log records are
  ignored
• Transactions that have neither Commit nor Abort
  log records receive Abort log records

15
Recovery Using Immediate Update

• Updates are written to the database as they occur
• T starts
• Start log record
• T writes
• Update log record
• Update is written to the buffer
• Update is applied to the database when the buffer
  content is transferred to secondary storage
• T commits
• Commit log record

16
Recovery Using Immediate Update (cont)

• T aborts
• Updates are undone in reverse order
• Recovery after the crash
• Most recent checkpoint
• Transactions with Start and Commit log records
  are redone
• Transactions with Start log record that dont
  have Commit log record are undone

17
Recovery (Example 10)
T1
T2
T3
T4
T5
T6
tc
tf
18
ARIES

• Algorithm for Recovery and Isolation Exploiting
  Semantics
• Works with steal, no-force policies
• Three main phases
• Analysis
• Redo
• Undo

19
Basic Principles

• Write-ahead logging
• Repeating history during Redo phase
• Logging of all changes during Undo phase

20
Analysis Phase

• Scan log from the checkpoint to the end
• Main goal
• To determine the point in the log at which to
  start the Redo step
• To determine dirty pages at the moment of the
  crash
• To identify active transactions at the time of
  the crash

21
Analysis Step Result

• Transaction Table contains accurate list of
  transactions active at the moment of the crash
• (actions must be undone)
• Dirty Page Table contains list of modified pages
  that must be recorded to secondary storage
• (actions must be redone)

22
Redo Phase

• Repeating history paradigm
• Redo(Redo(A)) Redo(A)
• Redo actions are recorded to the log
• The result
• The database is brought to the state as before
  the crash

23
Undo Phase

• Scans backward from the end of the log
• Undo(Undo(A)) Undo(A)
• For every Undo action the CLR is added to the log
  
• Removes all changes caused by incomplete
  transactions