Cache Invalidation via Distributed Soft Lock and Versioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In large-scale cloud-based computer systems, concurrent access to cached data objects can lead to a 'stampede' scenario where multiple application threads attempt to reload the same data from the database, wasting resources and causing long wait times due to missing or invalid cached data objects.
Innovation Solution
Implementing a distributed soft lock mechanism within the cache services to arbitrate among parallel application threads, allowing only one thread to reload the data object, and using an extended TTL strategy with versioning data to determine if the cached object is still valid, thereby minimizing database access and optimizing resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If multiple application threads concurrently access cached data objects, then data access speed is improved, but resource wastage increases due to stampede scenarios
Solution Approach 1:
A distributed soft lock mechanism is introduced as an intermediary to arbitrate among parallel application threads. The lock acts as a mediator that coordinates access to the cache reload operation, allowing only one thread to execute the reload while others wait, thereby preventing the stampede scenario and reducing resource wastage while maintaining concurrent access capability
2Reliability
If cached data objects are invalidated frequently, then data consistency is improved, but wait times increase due to repeated reload operations
Solution Approach 1:
Versioning data is maintained preliminarily alongside cached data objects. When data is updated in the database, the version number is incremented before invalidating the cache. This preliminary action allows the system to quickly determine cache validity without full reload operations, reducing wait times while maintaining data consistency through version comparison
3Reliability
If data objects are reloaded from database, then data freshness is improved, but network and database resource usage increases
Solution Approach 1:
The system implements feedback through versioning data that tracks database changes. Before reloading data from the database, the system checks the version number against the cached version. This feedback mechanism allows the system to determine whether a reload is necessary, ensuring data freshness only when needed and reducing unnecessary network and database resource usage
Data Source
AI summary
A request for a write action to be perform a data entity persisted in primary data sources is received. It is determined whether data changes to the data entity are tracked based on an object version key (OVK) definition of a target data object type. If the data changes to the data entity are tracked based on the OVK definition, an OVK key is generated, based on the OVK definition, from data field values specified in the request for the write action. An OVK version number for the data entity is updated, the OVK version number being identified by the OVK key. A cached data object of the target data object type is caused to be invalidated by the updated OVK version number, the cached data object of the target data object type being derived based on underlying data in the data entity persisted in the primary data sources.