Re: read error recovery threshold

[NeilBrown <neilb@suse.de>]

[-- Attachment #1: Type: text/plain, Size: 1163 bytes --]

On Mon, 15 Sep 2014 10:56:11 -0600 Eric Mei <meijia@gmail.com> wrote:

> Hi,
> 
> After a read error detected, RAID6 will initiate a recovery procedure
> try to correct it, until the number of read error exceeds a threshold,
> which is "conf->max_nr_stripes" (see raid5_end_read_request()), I'm
> wondering the reasoning behind this. To me the threshold seems a drive
> property, but max_nr_stripes is a array-wide cache setting and can be
> changed at runtime. In our specific case, we observed a drive emitting
> lots of read errors without being marked as faulty because the larger
> max_nr_stripes
> setting.
> 
> Look at other part of MD code, there is "mddev::max_corr_read_errors"
> which is set to 20, but only RAID10 makes use of it. Also the comment
> above MD_DEFAULT_MAX_CORRECTED_READ_ERRORS says "...We divide the read
> error count by 2 for every hour elapsed between read errors", but I
> don't see any code matching this description.
> 
> Any thoughts? Thanks

Yes, it is inconsistent.
It wasn't designed to be inconsistent, it just happened.
Patch with good justification will be looked on kindly.

Thanks,
NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]