Re: SES Enclosure Management.

[Benjamin ESTRABAUD <be@mpstor.com>]

On 14/02/12 20:53, Robert Woodworth wrote:
> On 02/14/2012 01:42 PM, Joe Landman wrote:
>> On 02/14/2012 03:31 PM, NeilBrown wrote:
>>> On Tue, 14 Feb 2012 10:30:37 -0700 Robert Woodworth
>>> <robertjwoodworth@gmail.com>  wrote:
>>>
>>>> Has anyone ever thought of integrating SES managed enclosures into the
>>>> kernel RAID system?  I briefly looked through the archives and have
>>>> not found anything on the topic.
>>>>
>>>> Some HW based RAID controllers do this flawlessly now, there is no
>>>> reason why the kernel RAID cannot also.  (LSI MegaRAID)
>>>>
>>>> 1) When a drive is part if a managed enclosure, the RAID system should
>>>> address it by location instead of by enumerated device node.  The SES
>>>> device in the enclosure can map the physical slot to a physical drive.
>>>>   The RAID admin (mdamd) should be able to add/fail/identify devices
>>>> based on slot.
>>>
>>> Does this just mean that the admin should using names in 
>>> /dev/disk/by-path/
>>> rather than /dev/sdXX to address devices?  What can md or mdadm do 
>>> to help?
>>>
>>
>> Not sure on the SES (or SGPIO side), but one of the things we've been 
>> doing has been to create a file with disk placement "coordinates", so 
>> as to map serial number and device to physical location.
>>
>
> With real SES managed enclosures, you issue a SCSI command to read SES 
> Page1 and Page2 to get the details about the drives in any given 
> slot.  This currently works fine in Linux with the sg_utils3 package.  
> From the command line, 'sg_ses -p 2 /dev/sgXX`  where the device is 
> the SES device.
>
> Take a look at your systems, if you see a device at 
> /sys/class/enclosure/XXXX/ then you have a managed enclosure attached.
>

Hi,

True, but your definition of SES "slot 0" might be different from 
someone else's. The end user will still need to map "slot 0" to a 
physical slot on his system, instead of mapping a drive's serial number 
to a physical slot. There is no standard for slot numbering and most 
vendors use different schemes.
>
>>>>
>>>> 2) If the RAID system fails a drive, it should notify the SES
>>>> management and turn on the fail bit and the fail LED.
>>>
>>> "mdadm --monitor" will run a script on drive failure.  This could 
>>> easily
>>> notify the SES management.
>>
>> Yes, we are using this now for notifications and logging.
>>
>>>
>>> So maybe all we need here is a script to plug in to mdadm... Would 
>>> you like
>>> to write one?
>>>
>>
>> Just need a "standard" SES (or SGPIO) mechanism to hook into, and we 
>> should be able to support this.  Right now we have to work through 
>> HBA scripts.
> A true managed enclosure has nothing to do with the HBA.  A managed 
> enclosure provides a device on the SCSI bus and you exclusively 
> communicate with that device regardless of the HBA.  Most HW RAIDs 
> (LSI MegaRAID) will hide the SES device exactly like they hide the 
> physical disks.
>
>>
>>>>
>>>> 3) The RAID system should be able to turn on the 'identify' bit and
>>>> LEDs for an array or a single drive.
>>>
This is fine too, as you pointed out SES gives you information on disks, 
but in order to map a block device sda to a device in SES, won't you 
need to rely on its SAS address? or even the SAS address arbitrarily 
given by the expander to a SATA device?

Your expander's SES device might display this address for both SAS and 
SATA drives in SES page 0xa, but most SES devices won't, at least for 
SATA drives, as the SES spec specifically requires SATA drive's address 
to be set to 0x0.

So if drive "sd"a fails, you'll be able to tell which expander it is 
attached using kernel infos, tell its SAS address as you see it, but you 
won't be able to map it to a particular slot number in SES, and will not 
be able to set its failed LED reliably.
>>> Again, it sounds like you just need a script to ask mdadm which 
>>> devices are
>>> included in a given array, and then do whatever magic is needed to 
>>> turn on
>>> the light.
>>> It is fairly easy to extra the device list from the output of
>>>    mdadm --detail --brief --verbose /dev/md/whatever
>>>
>>> but it might be good to make it easier to extra from
>>>    mdadm --detail --export /dev/md/whatever
>>>
>>> Would you like to write such a script?
>>>
>>>>
>>>> I'm currently doing firmware on a managed enclosure.  Although my
>>>> vendor only supports LSI MegaRAID, there is no reason why my
>>>> enclosures cannot work in the same manner on a kernel RAID system.
>>>>
>>>>
>>>>
>>>> Request for comments...
>>>
>>> It sounds to me like you just need a few scripts to provide some
>>> enclosure-specific functionality.  I would be happy to include them 
>>> in the
>>> mdadm distribution.
>>>
>>> Or maybe there is something that I didn't understand??
>>>
>>> Thanks,
>>> NeilBrown
>>>
>>>
Regards,
Ben.
>>>
>>>> -- 
>>>> To unsubscribe from this list: send the line "unsubscribe 
>>>> linux-raid" in
>>>> the body of a message to majordomo@vger.kernel.org
>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>
> -- 
> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>