Database Recovery Concepts

 

Database Recovery Concepts

1. Introduction

Database recovery refers to the process of restoring the database to a consistent state after a failure such as system crashes, transaction errors, or disk failures.

The main objectives of recovery are:

  • Atomicity – Incomplete transactions must be undone.

  • Durability – Committed transactions must remain permanent.

To support recovery, the DBMS maintains a system log that records all updates performed by transactions.

2. Types of Failures

Database failures are classified into two types:

1. Catastrophic Failure

These failures cause physical damage to the database.

Examples:

  • Disk crash

  • Hardware failure

Recovery method:

  • Restore database from backup

  • Redo committed transactions using the log.

2. Non-Catastrophic Failure

These failures do not damage the database physically.

Examples:

  • System crash

  • Transaction abort

  • Software errors

Recovery is done using log records.

3. Recovery Algorithms

There are two main recovery techniques.

TechniqueDescription
Deferred UpdateDatabase updated only after transaction commits
Immediate UpdateDatabase updated before commit

4. Deferred Update (NO-UNDO / REDO)

In deferred update, the database is not updated until the transaction commits.

  • Updates are stored in buffers or logs.

  • If transaction fails → no undo required.

  • If system crashes after commit → redo operations are required.

Thus it is called:

NO-UNDO / REDO

5. Immediate Update

In immediate update, the database may be updated before the transaction commits.

If a transaction fails:

  • Changes already written to disk must be undone.

Recovery may require:

  • Undo of uncommitted transactions

  • Redo of committed transactions

Two variations exist:

AlgorithmDescription
UNDO/REDO        Both undo and redo required
UNDO/NO-REDO        Only undo required

6. Write-Ahead Logging (WAL)

Write-Ahead Logging ensures that log records are written to disk before database updates.

Rules of WAL:

  1. Before a data item is updated, its before image must be written to the log.

  2. A transaction cannot commit until all its log records are written to disk.

This guarantees correct recovery.

7. Buffer Management

DBMS uses main memory buffers (cache) to store disk pages temporarily.

Important components:

Dirty Bit

  • Indicates whether a page has been modified.

ValueMeaning
0        Not modified
1    Modified

Pin/Unpin Bit

  • Determines whether a page can be written back to disk.

8. Steal / No-Steal and Force / No-Force Policies

Steal / No-Steal

PolicyMeaning
No-Steal            Updated pages cannot be written before commit
Steal            Updated pages may be written before commit

Force / No-Force

PolicyMeaning
Force        All updates written to disk before commit
No-Force        Updates may remain in memory

Most DBMS use:

Steal + No-Force

Which requires UNDO and REDO operations.

9. Checkpoints

A checkpoint is a special log record that indicates that all modified buffers have been written to disk.

Format:

[checkpoint, list of active transactions]

Purpose:

  • Reduces recovery time

  • During crash recovery, the system only examines log records after the last checkpoint.

10. Fuzzy Checkpointing

Normal checkpointing pauses transaction processing.

To avoid delays, fuzzy checkpointing is used.

Steps:

  1. Write begin_checkpoint record

  2. Continue transaction execution

  3. Flush buffers gradually

  4. Write end_checkpoint

11. Transaction Rollback

If a transaction fails before commit, the system performs rollback.

Rollback process:

  • Use log records

  • Restore before images (BFIM) of data items

  • Undo all write operations.

12. Cascading Rollback

Cascading rollback occurs when one transaction depends on another.

Example:

  • T1 writes a value

  • T2 reads that value

  • If T1 aborts → T2 must also abort.

To avoid cascading rollback, DBMS uses strict or cascadeless schedules.

Conclusion

Database recovery techniques ensure that the database remains consistent and reliable after failures.
Key components of recovery include:

  • Log files

  • Write-Ahead Logging

  • Checkpoints

  • Undo and Redo operations

Modern DBMS typically use steal/no-force strategy with WAL, which supports efficient recovery.

Comments

Post a Comment

Popular posts from this blog

Database Management Systems DBMS PCCST402 Semester 4 KTU CS 2024 Scheme

 About Me Database Management Systems DBMS PCCST402 Scheme and Syllabus Model Question Paper - KTU BTech DBMS PCCST402 2024 Scheme Module-1 Introduction to DBMS Introduction to DBMS Example  Database Database System Vs Traditional File Processing Advantages of using DBMS Database System Concepts and Architecture Data Models, Schemas and Instances   Three schema architecture and Data Independence Database Languages and Interfaces Centralized and Client Server Architectures in DBMS Classification of Database Management Systems Conceptual Data Modeling and Database Design Conceptual Modeling and the Role of the ER Model in Database Design Entities and Attributes  Company Database Application - Example Entity Types, Entity Sets, Keys, and Value Sets in the ER Model Initial Conceptual Design of the COMPANY Database Relationship Types in the ER Model Relationship Degree, Role Names, and Recursive Relationships Constraints on Binary Relationship Types Attributes of Relation...
Read more

Data Models, Schemas and Instances

  Data Models, Schemas and Instances  1. Data Abstraction In databases, data abstraction means hiding the complex details of how data is stored and focusing only on what is important. Different users can view data at different levels of detail depending on their needs. 2. Data Models A data model is a collection of concepts used to describe the structure of a database. It defines: Data types Relationships among data Constraints on data Operations for accessing and updating data Data models help us understand and design databases without worrying about low-level storage details. Categories of Data Models a) High-level / Conceptual Data Models Close to how users think about data. Use concepts like entities , attributes , and relationships . Example: ER (Entity–Relationship) Model . b) Representational (Implementation) Data Models In between conceptual and physical models. Easy for users to understand but still close to how data is st...
Read more

Introduction to Database Management System -DBMS

Image
In today’s digital world, almost every application—from mobile banking and e-commerce to university portals and hospital systems—depends on the efficient storage, organization, and retrieval of data. As the amount of information generated by individuals and organizations continues to grow, managing this data in a structured and meaningful way becomes essential. This is where database systems play a central role. A database can be thought of as a well-organized collection of related data, designed so that it can be easily accessed, managed, and updated. But the database itself is only one component. To manage it effectively, we rely on a specialized software package known as a Database Management System (DBMS) . A DBMS provides a systematic and automated way to store large volumes of data, enforce rules for consistency and correctness, protect data from unauthorized access, and allow multiple users to work with the data simultaneously without conflicts. A database typically has three...
Read more