Re: WWN

You’re absolutely correct.  10 years ago, “good” wasn’t even an option, but it was a reason why I wanted the transport split out so that SCSI/SPI disk change detection could be handled differently from SATA and from SAS.  Since then, it’s just been inertia, like Warner surmised.

Scott


> On Apr 18, 2022, at 3:28 PM, David Gwynne <david@gwynne.id.au> wrote:
> 
> don't let perfection get in the way of doing something good.
> 
>> On 19 Apr 2022, at 04:11, Scott Long <scottl@samsco.org> wrote:
>> 
>> It’s been inertia from the hardware standpoint.  SATA was notorious for having unusable WWNs, serial numbers, etc.  It’s also been complicated by the varying quality of SATL and SATA-pass-thru implementations  in controllers (i.e. LSI).  The Spectra guys probably have a lot of good experience to share here.
>> 
>> Keep VM’s in mind though.  Emulated environments don’t always try very hard to generate unique IDs for storage.  It might be possible to heuristically figure out what kind of environment the system is in and whether the IDs are dependable.
>> 
>> Scott
>> 
>> 
>>> On Apr 18, 2022, at 11:47 AM, Warner Losh <imp@bsdimp.com> wrote:
>>> 
>>> Is there a reason we don't rely primarily on WWN changing to detect a disk change at a particular location? I know it's not universally available, but anything made in the last 15 or 20 years should have one if my research is correct... Or is this just a case of inertia?
>>> 
>>> I'm looking at making ahci a little more resilient to transient outages, and thought it might be best to key primarily off this and secondarily off other changed information when that's not available. If I had a WWN, then I'd know the disk that was gone for 500ms is the same one and I could resume its operations and still detect that someone unplugged drive A and plugged in drive B.
>>> 
>>> Warner
>>> 
>> 
>> 
>