Stream Computing Fault Handling via Checkpoint Replay
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current stream computing systems face challenges in ensuring no-duplicate and no-loss data transmission, particularly in complex computing scenarios like finance and advertisement, where the existing 'exactly once' semantics are not adequately supported, leading to data inconsistency and increased development costs due to the need for additional complex data replay and verification logic.
Innovation Solution
A fault handling method in a stream computing system that records arrival sequences of original data, performs periodic persistence operations, restores and replays data upon failure, and continues encoding computed results to ensure accurate data transmission without duplication or loss, using a checkpoint mechanism and data sequential replay.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If users add additional complex data replay and verification deduplication logic to the traffic layer to guarantee no-duplicate and no-loss of computation results, then data integrity is improved, but device complexity and development cost increase significantly
Solution Approach 1:
The patent introduces an intermediary checkpoint mechanism between data production and consumption. Checkpoints periodically store computation state and data sequences, acting as a mediator that enables fault recovery without requiring complex user-side verification logic. This intermediary layer handles the complexity of data integrity guarantees internally.
Solution Approach 2:
The system performs preliminary actions by proactively creating checkpoints before failures occur. Data sequences and computation states are persisted in advance, enabling rapid recovery without requiring complex post-failure verification logic from users. The checkpoint mechanism prepares recovery data beforehand, reducing both complexity and recovery time.
2Ease of operation
If the system uses simple storage functions like update power to save computation results, then ease of operation is improved, but manufacturing precision and data accuracy deteriorate due to inability to support complex computing scenarios
Solution Approach 1:
The patent segments the data processing function into multiple components: simple storage for checkpoint data, sequential recording modules for data arrival sequences, and replay modules for computation recovery. This segmentation allows the system to maintain simple storage operations while achieving high data accuracy through the coordinated work of multiple specialized components.
Solution Approach 2:
The system creates copies of data sequences and computation states at checkpoint intervals. These copies enable accurate replay of computations without requiring complex storage operations during normal processing. The copy mechanism preserves data accuracy while maintaining operational simplicity through standardized copy-paste semantics.
3Device complexity
If the system implements fault recovery without checkpoint mechanisms, then device complexity is reduced, but reliability deteriorates due to data loss and duplication issues
Solution Approach 1:
The patent implements periodic checkpoint creation at fixed intervals during stream processing. This periodic action provides reliable fault tolerance through regular state snapshots without requiring complex continuous monitoring or adaptive checkpointing logic. The periodic mechanism achieves reliability through simplicity, using fixed-time intervals rather than complex decision-making.
Solution Approach 2:
The system discards computation state periodically by storing it in checkpoints, then recovers from these checkpoints after failures. This discard-recover pattern provides reliable fault tolerance through simple periodic storage and recovery operations, avoiding the need for complex continuous state management or sophisticated recovery algorithms.
4Adaptability or versatility
If the system uses existing stream computing semantics like at most once or at least once, then adaptability to different scenarios is improved, but manufacturing precision deteriorates because exactly once semantics with no-duplicate and no-loss cannot be achieved
Solution Approach 1:
The checkpoint mechanism acts as an intermediary that enables exactly once semantics with no-duplicate and no-loss guarantees. By mediating between data production and consumption through periodic state snapshots, the system achieves precise data accuracy while maintaining flexibility in handling different stream processing scenarios.
Solution Approach 2:
The patent replaces complex mechanical verification and deduplication logic with a simpler checkpoint-based recovery mechanism. This substitution achieves exactly once semantics through state restoration rather than through complex comparison and filtering operations, improving data accuracy while maintaining scenario adaptability.
Data Source
AI summary
The objective of the present invention is to provide a method, apparatus, computing node and computer program product for fault handling in a stream computing system. Here, at a computing node, recording arrival sequences of respective original data from a upstream computing node; performing persistence operation on the respective original data according to a predetermined period; in the case of failure and restart, restoring to-be-computed data in internal storage from the original data subjected to the persistent operation and/or the upstream computing node, and replaying and computing the restored to-be-computed data according to the respective previous arrival sequences; continuing encoding each completely computed result data according to offset of the result data in the last persistent operation period before the failure and transmitting the encoded result data to a next node.

