Cluster File System Node Failure Recovery via Demand-Driven State Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cluster file systems face inefficiencies in recovering from node failures due to costly synchronous checkpointing processes that require extensive inter-machine communication, which can be more resource-intensive than the file operations themselves.
Innovation Solution
Implementing a recovery mechanism that utilizes file agents to compile 'snapshots' of file states before node failures, promoting file server secondaries to primaries, and employing a reference counting subsystem to track object references, allowing for demand-driven file state recovery without the need for extensive data replication.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If synchronous checkpointing is used to replicate file server primary information to file server secondaries, then reliability is improved through failure recovery capability, but device complexity and resource consumption increase due to extensive inter-machine communication
Solution Approach 1:
The patent performs file state snapshots and caching in advance during normal operation, so that when a failure occurs, the recovery process can immediately use the pre-prepared cached state without needing to perform expensive synchronous replication at the moment of failure
Solution Approach 2:
The patent creates simplified copies of file state information in the form of snapshots cached on file server secondaries. These copies are maintained asynchronously and contain sufficient information for recovery, avoiding the need for complex real-time replication mechanisms
2Reliability
If checkpoint information is replicated to all file server secondaries for every file operation, then reliability is improved, but productivity decreases due to the cost being more than the file operation itself
Solution Approach 1:
The patent applies partial replication by caching checkpoint information only on selected file server secondaries rather than all secondaries. The reference counting mechanism determines which secondaries need the checkpoint data, allowing the system to perform replication only where necessary, thus reducing overhead while maintaining reliability
Solution Approach 2:
The patent distributes checkpoint information selectively to specific file server secondaries based on local needs determined by reference counting. Each secondary holds checkpoint data only if it has active references to it, creating a localized optimization where replication effort matches actual demand
3Reliability
If full file state recovery is performed immediately upon node failure, then reliability is improved through quick service restoration, but use of energy and resources increase due to unnecessary recovery of files not accessed after failure
Solution Approach 1:
The patent implements a dynamic recovery approach where the system first performs a reference count sweep to identify which files actually need recovery based on active references. This dynamic assessment allows the system to adapt the recovery scope to actual needs, avoiding static full-recovery approaches that waste resources on unnecessary file restorations
Data Source
AI summary
A cluster file system including a file server primary configured to satisfy a plurality of file operation requests, a file server secondary configured to be promoted to the file server primary in the event of a failure of a node on which the file server primary is located, a plurality of file agents configured to process the plurality of file operation requests on the node, and a reference counting subsystem configured to track an object reference of the plurality of file agents.