RaidfileImprovements
The current raidfile code cannot write to degraded stores, and there is no means of rebuilding a lost disc. This means that, while it's good at protecting against disc failures, recovery is inconvenient.
A small-ish project, suitable for one developer to undertake in a short timeframe, is to extend the raidfile support.
Scan the raidfile sets to find broken files, then use the existing code in lib/raidfile to repair any damage by opening the file, with raidfile, twice for read and write, and copy the contents from one to the other.
Off-load the transform from one file to the three raid files to a separate daemon. Use this to queue up written files so that writing to a degraded store is possible. You will have to modify the raidfile code to avoid choosing the offline disc for storage of the first file.
Make sure the user, group and permissions of the files are maintained.
The raidfile code offers an opportunity to build redundant clusters of three machines on a local network with shared discs over NFS. Use the above make this an efficient and supported configuration.
- Introduction
- Overview
- System Requirements
- Comparison With Other Systems
- Endorsements
- Installation
- Common Tasks
- Related Projects