SoftwareRaid

Summary

mdadm currently has a gaggle of open bugs, and every cycle the RAID ISO tests produce new and interesting bugs. It seems like we're doing something a bit wrong with Software RAID. There are some valid solutions here: https://wiki.ubuntu.com/ReliableRaid which should be discussed and either refuted or implemented.

Release Note

A number of long standing bugs in software RAID have been fixed, and a clear specification for how mdadm should work in Ubuntu has been written.

Rationale

mdadm, and the kernel's MD driver, are responsible for users' data storage. It must be as reliable as possible and the ways in which it works or does not work should be well known and documented.

Currently there are conflicting bug reports for mdadm in Ubuntu, and they all seem like potential bugs. This blueprint will attempt to remove any of the ambiguity of how mdadm is expected to work in Ubuntu. We also intend to assert the intentions of the upstream (mdadm and kernel md) authors' whenever possible.

User stories

As a system administrator, I want to have confidence that data stored on software RAID arrays will not be lost or corrupted on boot, and if there is a problem, I expect to be informed of it and for my configuration decisions to be respected.

Assumptions

Design

The design is a large part of the work that needs to be done, as it involves reviewing existing reports and known fail modes and outlining how the system should handle them.

Note that this will incorporate elements from previous efforts to improve the RAID system, such as ReliableRaid, where possible.

Implementation

Pending Design section.

Test/Demo Plan

Testing will continue with the existing ISO testing for RAID 1, though other automated tests will be added if possible using the new automated server install testing suite.

Unresolved issues


CategorySpec

Oneiric/SoftwareRaid (last edited 2011-05-19 22:45:15 by cpe-76-94-215-209)