Light-Weight Checkpoints for Transaction Log Restoration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing redo logs face challenges in checkpointing, including reducing the amount of state saved and determining safe points for checkpointing, which limits the usability of logical redo logs for data mining and replication, as they require extensive time and resources to restore systems.
Innovation Solution
Implementing light-weight checkpoints that save only the active transactions' state at specific safe locations in the transaction stream, allowing for efficient extension of logs and selection of checkpoints based on mean time to make the extending log, thereby reducing the cost and time of checkpointing and improving usability for data mining and replication.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional checkpoints are made to enable system restoration, then recoverability is improved, but the time and resources required to make and restore from checkpoints increase significantly
Solution Approach 1:
The patent segments the transaction log into multiple extents, where each extent contains a portion of the log records. Checkpoints are made at extent boundaries rather than requiring full log processing, enabling incremental restoration. The log is divided into manageable chunks that can be processed independently, reducing the time penalty of checkpoint operations while maintaining complete recoverability.
Solution Approach 2:
The patent performs preliminary organization of transaction logs into extent-based structures during normal operation, pre-grouping records by extent identifiers. This preliminary action ensures that when a checkpoint is needed, the system can quickly locate and process only the relevant extent portions rather than scanning entire log files, significantly reducing checkpoint creation and restoration time.
2Reliability
If complete transaction state is saved at checkpoints, then restoration accuracy is improved, but the storage space required for checkpoints increases
Solution Approach 1:
The patent extracts only the essential extent identification information from complete transaction states for checkpoint storage. Instead of saving full transaction state data, the system stores minimal extent markers that identify which log extents have been processed. This extraction approach maintains restoration accuracy by preserving the critical boundary information while dramatically reducing checkpoint storage requirements.
Solution Approach 2:
The patent uses lightweight extent marker objects instead of heavy transaction state snapshots. These minimal checkpoint markers serve their purpose of enabling restoration and can be quickly regenerated from the extent-based log structure, functioning as disposable, low-cost checkpoint representations that maintain accuracy without consuming significant storage space.
3Loss of information
If the entire redo log is processed for data mining, then complete transaction information is obtained, but the time and computational resources required increase
Solution Approach 1:
The patent segments the redo log into extents with unique identifiers, enabling data mining operations to target specific extent ranges rather than processing the entire log. Users can specify extent ID ranges to mine, allowing selective extraction of relevant transaction information while excluding unrelated portions, thus maintaining information completeness for the queried range while dramatically improving mining efficiency.
Solution Approach 2:
The patent performs preliminary organization of redo log records into extent-based structures with indexed extent identifiers during normal logging operations. This preliminary action enables data mining tools to quickly locate and process only the relevant extent portions based on extent ID ranges, avoiding the need to scan and process the entire redo log while ensuring complete transaction information is obtained for the specified range.
Data Source
AI summary
Techniques for making light-weight checkpoints in logs of streams of transactions and for extending the logs from the checkpoints. The state saved in the light weight checkpoint need only include the state of all transactions that are active at a prior point in the log and are still active at the checkpoint. A log is extended from the checkpoint by processing new transactions in the transaction stream beginning at the prior point to produce the extending log. When the checkpoint is reached, the state saved at the checkpoint is used to continue to produce the extending log. Techniques are further disclosed for selecting points in the stream of transactions at which the checkpoints may be made and for determining the distance between checkpoints. As disclosed, the log in which the checkpoints are used is a logical log made from a physical log produced by a relational database system. The logical log may be used for replication and for data mining.