Raid1 failure rules?

Fri, 21 Dec 2001 16:19:34 -0500

I asked this question on the linux-raid email list and absolutely no one responded to it.  I am not that surpised since of all the RAID info out there, the nuts and bolts rules are seemingly not discussed by anyone (at least not for free).

So, I come to PLUG where many questions are answered or at least responded to.  Anyone have any ideas on where to get info on the subject, other than reading the source (most accurate but most time consuming)?

Alan

--<original post to linux-raid below>--

Date:     2001-12-19 17:16:04

Hello,

As I am investigating software RAID1 operation in Linux, I am finding more questions after I find my initial answers.  Specifically, I am wondering about mirror failure rules.  For example:

If the read from one mirrored drive fails, the data is taken from the other drive.  Fine.  But, is the drive that failed the read that one time now declared entirely bad or does the RAID code attempt to recover that bad sector by re-writing the data from the good drive?  Or, is there some threshold number for read errors from a drive that must be reached before it is declared as bad?  What about write errors?

I guess I want to know exactly what errors and what rules regarding errors must be met for a mirrored drive to be marked bad.  Is there a reference or resource somewhere, industry standard or otherwise, that defines this?  Even just a pointer to the source code that makes these decisions would be fine.

Alan