Cluster Metadata Recovery Using Epoch Undo Records
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Distributed file systems with a single metadata server represent a single point of failure and computational bottleneck, and redundant metadata servers do not adequately address the issue of file system consistency during failures, leading to potential file system inconsistencies due to lost metadata across multiple metadata servers.
Innovation Solution
Implementing a method that uses an 'epoch' of undo records to create a virtual instantaneous snapshot of a consistent file system image, allowing file servers to roll back to this state in case of failures, and maintaining and advancing the epoch value to purge unneeded undo records, thereby ensuring file system recovery without invoking coordinated wait conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single metadata server is used in a distributed file system, then the system structure is simple and easy to manage, but it creates a single point of failure and computational bottleneck
Solution Approach 1:
The patent divides the single metadata server into multiple metadata servers (MDS0, MDS1, etc.), each managing a portion of the namespace. This segmentation eliminates the single point of failure while distributing the computational load, thereby improving reliability without proportionally increasing complexity.
Solution Approach 2:
The patent introduces a new dimension of organization by dividing the namespace into hierarchical segments (e.g., /user/*, /home/*) that can be distributed across multiple metadata servers. This dimensional approach allows the system to scale horizontally while maintaining manageable complexity through structured delegation.
2Reliability
If redundant metadata servers are implemented to eliminate single point of failure, then system reliability improves, but file system consistency during failures becomes problematic
Solution Approach 1:
The patent pre-establishes epoch boundaries and undo records before failures occur. Each metadata server maintains a log of transactions organized by epochs, with undo information ready to revert to consistent states. This preliminary preparation ensures that when failures happen, consistency can be restored without complex coordination.
Solution Approach 2:
The patent converts the potential harm of distributed metadata servers (inconsistency risk) into a benefit by using epoch-based undo records. The distributed architecture's flexibility is transformed into an advantage where each server can independently rollback to known good states, actually improving consistency recovery compared to centralized systems.
3Productivity
If multiple metadata servers are used to distribute computational load, then the computational bottleneck is relieved, but file system consistency across servers becomes difficult to maintain
Solution Approach 1:
The patent implements a feedback mechanism where metadata servers periodically exchange epoch information and transaction logs. This feedback loop allows servers to stay synchronized about the current consistent state, enabling load distribution while maintaining consistency through coordinated awareness of epoch boundaries.
Solution Approach 2:
The patent introduces epoch boundaries as an intermediary concept that mediates between multiple metadata servers. These epoch markers serve as a common reference point that all servers can use to determine consistency, acting as a virtual coordinator without requiring a actual centralizing authority.
4Reliability
If undo records are maintained for all transactions to enable recovery, then file system recovery capability is improved, but storage space is consumed by historical records
Solution Approach 1:
The patent implements periodic epoch advancement where undo records are retained only for the current and previous epochs. When epochs advance, old undo records are purged. This periodic retention policy maintains sufficient recovery capability while systematically limiting storage consumption to a bounded amount.
Solution Approach 2:
The patent discards old undo records after they have served their purpose of enabling recovery to previous consistent states. By advancing epochs and purging outdated records, the system recovers storage space while maintaining the ability to recover to any recent consistent state, balancing reliability with resource efficiency.
Data Source
AI summary
A method and apparatus are described for recovering a fully consistent file system stored in a cluster file system with multiple metadata servers using an epoch of undo records. The epoch consists of (i) a virtual instantaneous snapshot marking a consistent and valid file system image and (ii) a set of undo records that enable the file servers to roll-back to this fully consistent image associated with the file system. The file system is recovered by rolling back file transactions associated with undo records subsequent to the undo records associated with the epoch snapshot. In addition, the undo records are maintained by advancing the epoch value and purging unneeded undo records.