CHAPTER 5: DATABASE RECOVERY

TECHNIQUES

DR. MAJD ALHAWAMDEH

 Introduction

 Recovery algorithms
 Recovery concepts
 Write-ahead logging

 In-place versus shadow updates

 Rollback

 Deferred update

 Immediate update

 Certain recovery techniques best used with specific

concurrency control methods

Slide 22- 2
 22.1 Recovery Concepts

 Recovery process restores database to most recent

consistent state before time of failure
 Information kept in system log
 Typical recovery strategies
 Restore backed-up copy of database
 Best in cases of extensive damage
 Identify any changes that may cause inconsistency
 Best in cases of noncatastrophic failure
 Some operations may require redo

Slide 22- 3
 Recovery Concepts (cont’d.)

 Deferred update techniques

 Do not physically update the database until after transaction
commits
 Undo is not needed; redo may be needed

 Immediate update techniques

 Database may be updated by some operations of a transaction
before it reaches commit point
 Operations also recorded in log

 Recovery still possible

Slide 22- 4
 Recovery Concepts (cont’d.)

 Undo and redo operations required to be idempotent

 Executing operations multiple times equivalent to executing
just once
 Entire recovery process should be idempotent

 Caching (buffering) of disk blocks

 DBMS cache: a collection of in-memory buffers

 Cache directory keeps track of which database items are in the

buffers

Slide 22- 5
 Recovery Concepts (cont’d.)

 Cache buffers replaced (flushed) to make space for

new items
 Dirty bit associated with each buffer in the cache
 Indicates whether the buffer has been modified
 Contents written back to disk before flush if dirty bit
equals one
 Pin-unpin bit
 Page is pinned if it cannot be written back to disk yet

Slide 22- 6
 Recovery Concepts (cont’d.)

 Main strategies
 In-place updating
 Writes the buffer to the same original disk location
 Overwrites old values of any changed data items

 Shadowing
 Writes an updated buffer at a different disk location, to maintain
multiple versions of data items
 Not typically used in practice

 Before-image: old value of data item

 After-image: new value of data item

Slide 22- 7
 Recovery Concepts (cont’d.)

 Write-ahead logging
 Ensure the before-image (BFIM) is recorded

 Appropriate log entry flushed to disk

 Necessary for UNDO operation if needed

 UNDO-type log entries

 REDO-type log entries

Slide 22- 8
 Recovery Concepts (cont’d.)

 Steal/no-steal and force/no-force

 Specify rules that govern when a page from the database cache
can be written to disk
 No-steal approach
 Cache buffer page updated by a transaction cannot be written
to disk before the transaction commits
 Steal approach
 Recovery protocol allows writing an updated buffer before the
transaction commits

Slide 22- 9
 Recovery Concepts (cont’d.)

 Force approach
 All pages updated by a transaction are immediately written to
disk before the transaction commits
 Otherwise, no-force

 Typical database systems employ a steal/no-force

strategy
 Avoids need for very large buffer space
 Reduces disk I/O operations for heavily updated pages

Slide 22- 10
 Recovery Concepts (cont’d.)

 Write-ahead logging protocol for recovery algorithm

requiring both UNDO and REDO
 BFIM of an item cannot be overwritten by its after image until
all UNDO-type log entries have been force-written to disk
 Commit operation of a transaction cannot be completed until
all REDO-type and UNDO-type log records for that transaction
have been force-written to disk

Slide 22- 11
 Checkpoints in the System Log and Fuzzy
Checkpointing

 Taking a checkpoint
 Suspend execution of all transactions temporarily

 Force-write all main memory buffers that have been modified

to disk
 Write a checkpoint record to the log, and force-write the log to
the disk
 Resume executing transactions

 DBMS recovery manager decides on checkpoint

interval

Slide 22- 12
 Checkpoints in the System Log and Fuzzy
Checkpointing (cont’d.)

 Fuzzy checkpointing
 System can resume transaction processing after a
begin_checkpoint record is written to the log
 Previous checkpoint record maintained until end_checkpoint
record is written

Slide 22- 13
 Transaction Rollback

 Transaction failure after update but before commit

 Necessary to roll back the transaction

 Old data values restored using undo-type log entries

 Cascading rollback
 If transaction T is rolled back, any transaction S that has read
value of item written by T must also be rolled back
 Almost all recovery mechanisms designed to avoid this

Slide 22- 14
Figure 22.1 Illustrating
cascading rollback (a
process that never occurs
in strict or cascadeless
schedules) (a) The read
and write operations of
three transactions (b)
System log at point of
crash (c) Operations
before the crash

Slide 22- 15
 Transactions that Do Not Affect the Database

 Example actions: generating and printing messages

and reports
 If transaction fails before completion, may not want
user to get these reports
 Reports should be generated only after transaction reaches
commit point
 Commands that generate reports issued as batch
jobs executed only after transaction reaches commit
point
 Batch jobs canceled if transaction fails

Slide 22- 16
22.2 NO-UNDO/REDO Recovery Based on
Deferred Update
 Deferred update concept
 Postpone updates to the database on disk until the transaction
completes successfully and reaches its commit point
 Redo-type log entries are needed

 Undo-type log entries not necessary

 Can only be used for short transactions and transactions that

change few items
 Buffer space an issue with longer transactions

Slide 22- 17
 NO-UNDO/REDO Recovery Based on
Deferred Update (cont’d.)
 Deferred update protocol
 Transaction cannot change the database on disk until it
reaches its commit point
 All buffers changed by the transaction must be pinned until the
transaction commits (no-steal policy)
 Transaction does not reach its commit point until all its
REDO-type log entries are recorded in log and log buffer is
force-written to disk

Slide 22- 18
NO-UNDO/REDO Recovery Based on Deferred
Update (cont’d.)

Figure 22.2 An example of a recovery timeline to illustrate the effect of checkpointing

Slide 22-19
 22.3 Recovery Techniques Based
on Immediate Update

 Database can be updated immediately

 No need to wait for transaction to reach commit point

 Not a requirement that every update be immediate

 UNDO-type log entries must be stored

 Recovery algorithms
 UNDO/NO-REDO (steal/force strategy)

 UNDO/REDO (steal/no-force strategy)

Slide 22- 20
Figure 22.3 An example
of recovery using
deferred update with
concurrent transactions
(a) The READ and
WRITE operations of
four transactions (b)
System log at the point
of crash

Slide 22- 21
 22.4 Shadow Paging

 No log required in a single-user environment

 Log may be needed in a multiuser environment for the
concurrency control method
 Shadow paging considers disk to be made of n fixed-
size disk pages
 Directory with n entries is constructed
 When transaction begins executing, directory copied into
shadow directory to save while current directory is being used
 Shadow directory is never modified

Slide 22- 22
 Shadow Paging (cont’d.)

 New copy of the modified page created and stored

elsewhere
 Current directory modified to point to new disk block
 Shadow directory still points to old disk block
 Failure recovery
 Discard current directory

 Free modified database pages

 NO-UNDO/NO-REDO technique

Slide 22- 23
Shadow Paging (cont’d.)

Figure 22.4 An example of shadow paging

Slide 22- 24
 22.5 The ARIES Recovery Algorithm

 Used in many IBM relational database products

 Uses a steal/no-force approach for writing
 Concepts
 Write-ahead logging

 Repeating history during redo

 Retrace all database system actions prior to crash to reconstruct
database state when crash occurred
 Logging changes during undo
 Prevents ARIES from repeating completed undo operations if
failure occurs during recovery

Slide 22- 25
 The ARIES Recovery Algorithm (cont’d.)

 Analysis step
 Identifies dirty (updated) pages in the buffer and set of
transactions active at the time of crash
 Determines appropriate start point in the log for the REDO
operation
 REDO
 Reapplies updates from the log to the database

 Only necessary REDO operations are applied

Slide 22- 26
 The ARIES Recovery Algorithm (cont’d.)

 UNDO
 Log is scanned backward

 Operations of transactions that were active at the time of the

crash are undone in reverse order
 Every log record has associated log sequence number
(LSN)
 Indicates address of log record on disk
 Corresponds to a specific change of some transaction

Slide 22- 27
 ARIES Recovery Example

Figure 22.5 An
example of recovery
in ARIES (a) The log
at point of crash (b)
The Transaction and
Dirty Page Tables at
time of checkpoint
(c) The Transaction
and Dirty Page
Tables after the
analysis phase

Slide 16-28
 22.6 Recovery in Multidatabase Systems

 Two-level recovery mechanism

 Global recovery manager (coordinator) needed to
maintain recovery information
 Coordinator follows two-phase commit protocol
 Phase 1: Prepare for commit message
 Ready to commit or cannot commit signal returned
 Phase 2: Issue commit signal
 Either all participating databases commit the effect
of the transaction or none of them do

Slide 22- 29
Recovery in Multidatabase Systems (cont’d.)

 Always possible to recover to a state where either the

transaction is committed or it is rolled back
 Failure during phase 1 requires rollback
 Failure during phase 2 means successful transaction
can recover and commit

Slide 22- 30
 22.7 Database Backup and Recovery
from Catastrophic Failures

 Database backup
 Entire database and log periodically copied onto inexpensive
storage medium
 Latest backup copy can be reloaded from disk in case of
catastrophic failure
 Backups often moved to physically separate locations
 Subterranean storage vaults

Slide 22- 31
 Database Backup and Recovery
from Catastrophic Failures (cont’d.)

 Backup system log at more frequent intervals and

copy to magnetic tape
 System log smaller than database
 Can be backed up more frequently
 Benefit: users do not lose all transactions since last database
backup

Slide 22- 32
 22.8 Summary

 Main goal of recovery

 Ensure atomicity property of a transaction

 Caching
 In-place updating versus shadowing
 Before and after images of data items
 UNDO and REDO operations
 Deferred versus immediate update
 Shadow paging
 Catastrophic failure recovery

Slide 22- 33