CS 632 Lecture 6

1
CS 632 Lecture 6 Recovery
Principles of Transaction-Oriented Database
Recovery Theo Haerder, Andreas Reuter, 1983
ARIES A Transaction Recovery Method Supporting
Fine-Granularity Locking and Partial Rollbacks
Using Write-Ahead Logging C. Mohan, Don
Haderle, Bruce Lindsay, Hamid Piralesh, and
Peter Schwartz, 1992
2
Principles of Transaction-Oriented Database
Recovery

1. DATABASE RECOVERY WHAT IT IS EXPECTED TO DO
2. What Is a Transaction
3. Which Failures Have To Be Anticipated
4. Summary of Recovery Actions
5. THE MAPPING HIERARCHY OF A DBMS
6. The Mapping Process
7. The Storage Hierarchy
8. Different Views of a Database
9. Mapping Concepts for Updates
10. CRASH RECOVERY
11. Dependencies Between Buffer Management and
    Recovery Component
12. Classification of Log Data
13. Checkpoints
14. ARCHIVE RECOVERY

3

• What is a transaction?
• actions executed indivisibly
• four properties required (the ACID properties)
• Atomicity
• all actions in the transaction happen, or none
  happen
• Consistency
• if a transaction preserves the consistency of the
  DB
• Isolation
• execution of one transaction is isolated from
  that of other transactions
• Durability
• if a transaction commits, its effects persist

4
Which Failures Have to Be Anticipated?

• Transaction Failures
• caused by bad input or other violations of
  consistency
• System Failures
• caused by
• bugs in the DBMS code
• operating system fault
• hardware failure
• occur less frequently
• Media Failures
• caused by
• software errors
• hardware errors
• physical errors

5
Summary of Recovery Actions

• Transaction Undo
• when a transaction aborts itself
• Global Undo
• when recovering from a system failure
• incomplete transactions
• Partial Redo
• when recovering from a system failure
• complete transactions
• Global Redo
• archive recovery
• full redundancy

6
(No Transcript)
7
(No Transcript)
8
Different Views of a Database

• Current database
• the contents are found on disk or in the DB
  buffer
• Materialized database
• the state that the DBMS finds at restart after a
  crash
• Physical database
• all blocks of the on-line copy

• Three types of update operations
• modification of page contents affects the
  current database
• write affects the physical database
• propagation affects the materialized database

9
Mapping Concepts for Updates
Two schemes for propagation

• Two types of propagation strategies
• ATOMIC
• NOT ATOMIC

10
Dependencies Between Buffer Management and
Recovery Component

• UNDO Recovery Actions
• STEAL undo actions necessary
• NO STEAL undo actions not necessary
• REDO Recovery Actions
• FORCE redo actions not necessary
• NO FORCE redo actions necessary
• Partial REDO no logging necessary
• Global REDO logging necessary

11
Classification of Log Data

• Physical State Logging on Page Level
• before image after image
• Physical Transition Logging on Page Level
• differences between old and new states
• Physical State Logging on Access Path Level
• Physical/Logical Transition Logging on the Access
  Path
• Logical Logging on the Record-Oriented Level

12
Checkpoints

• limit the amount of REDO recovery required after
  a crash
• involve three steps
• write a BEGIN_CHECKPOINT record
• write all checkpoint data
• write an END_CHECKPOINT record
• types
• Transaction-Oriented Checkpoints
• Transaction-Consistent Checkpoints
• Action-Consistent Checkpoints
• Fuzzy Checkpoints

13
(No Transcript)
14
Conclusion

• taxonomy for classifying the implementation
  techniques for
• database recovery
• four criteria
• Propagation
• ATOMIC/ NOT ATOMIC
• Buffer Handling
• STEAL / NO STEAL
• EOT Processing
• FORCE / NO FORCE
• Checkpointing

15
(No Transcript)
16
ARIES A Transaction Recovery Method Supporting
Fine-Granularity Locking and Partial Rollbacks
Using Write-Ahead Logging
ARIES Algorithm for Recovery and Isolation
Exploiting Semantics
17
Goals

• Simplicity
• Operation logging
• Flexible storage management
• Partial rollbacks
• Flexible buffer management
• Recovery Independence
• Logical undo
• Parallelism and fast recovery
• Minimal overhead

18
Normal Processing

• Updates
• Total or Partial Rollbacks
• savepoints
• Transaction Termination
• Checkpoints
• fuzzy checkpoints

Restart Processing

• Analysis Pass
• Redo Pass
• Undo Pass

19
(No Transcript)
20
Log Records

• LSN (Log Sequence Number)
• unique id for each log record
• Type
• type of record
• TransId
• id of transaction that wrote the record
• PrevLSN
• LSN of the previous log record written by the
  same
• transaction that wrote the record
• PageID
• id of the page referred by the log
• UndoNextLSN
• present in CLR
• Data
• redo/undo data describing the update

21
Page Structure

• page_LSN the LSN of the log record describing
  the latest
• update to the page

Transaction Table

• TransId the id of the transaction
• State
• LastLSN
• UndoNextLSN

Dirty Pages Table

• PageId
• RecLSN

22
Restart Processing

• Analysis Pass
• Redo Pass
• Undo Pass
• Selective or Deferred Restart

23
Media Recovery

• fuzzy image copy (fuzzy archive dump)
• image copy checkpoint
• media recovery redo point

Nested Top Actions

• top actions
• nested top actions