Skip to contentSkip to conversation

devices get kicked from RAID about once a month

DISCUSSION Stefan /*St0fF*/ Hübner 3 participants 3 messages Jun 2, 2010 – Jun 3, 2010 MD RAID

lore
Share & tools
lore
#0 lore raw

-------- Original-Nachricht --------
Betreff: Re: devices get kicked from RAID about once a month
Datum: Wed, 02 Jun 2010 19:08:58 +0200
Von: Stefan /*St0fF*/ Hübner <[email protected]> Antwort an: [email protected]
An: Dan Christensen <[email protected]>

Am 02.06.2010 18:33, schrieb Dan Christensen:

Quoted 10 lines
Quoted 4 lines · John Robinson

My Samsung Spinpoint F1's can have TLER enabled using a more recent smartctl. It's not appeared as part of a formal release yet but a patch went in to r3065 in SVN:
http://sourceforge.net/apps/trac/smartmontools/log/trunk/smartmontools

Thanks. I got the svn3077 from Debian testing, but it doesn't seem to be supported with my drives:

Quoted 6 lines
# smartctl -l scterc /dev/sda
smartctl 5.40 2010-03-16 r3077 [i686-pc-linux-gnu] (local build)
Copyright (C) 2002-10 by Bruce Allen,
http://smartmontools.sourceforge.net

Warning: device does not support SCT Commands

There you have it: the drives are not supporting SCT-ERC. Which I can say is true: I also own an SP2504C.

Quoted 2 lines

Any other suggestions?

Not really, it's up to Neil to export some sysfs-variable, where you could tune how long a drive may take to respond to some command.

The "trying hard to get the data back" can take up to a few minutes (somewhere I read about 2, somewhere else about 3 minutes). And what really is happening is: after some timeout the mdraid "thinks correctly", that the drive cannot provide the requested sector. To prevent failure, it reconstructs the "missing" data and issues a write request. Unfortunately the drive still tries to reconstruct the data itself and will not respond. After those write requests failing, mdraid drops the disk.

The ERC-setting is volatile. You'd have to issue it on every reboot, and on every hotswap. But at first you'd have to get drives that support this setting.

stefan

Quoted 7 lines

Dan

--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to [email protected]
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 [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#1 lore raw
Quoted 6 lines · On Wed, 02 Jun 2010 20:29:46 +0200 Stefan /*St0fF*/ Hübner <[email protected]>…
Quoted 2 lines

Any other suggestions?

Not really, it's up to Neil to export some sysfs-variable, where you could tune how long a drive may take to respond to some command.

Nope. md doesn't do any timeouts.
You need to look for, or ask for, such variables at the scsi/sata layer.

NeilBrown
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#2 lore raw
Quoted 4 lines · Neil Brown

On Wed, 02 Jun 2010 20:29:46 +0200
Stefan /*St0fF*/ Hübner <[email protected]> wrote:

Quoted 9 lines
Quoted 6 lines
Quoted 2 lines

Any other suggestions?

Not really, it's up to Neil to export some sysfs-variable, where you could tune how long a drive may take to respond to some command.

Nope. md doesn't do any timeouts.

That's the problem. A timeout between getting the timeout status and trying the rewrite is really needed to have any hope of recovery.

If there were a write intent bitmap for the drive, perhaps the drive could enter some "may be recovering" state and writes, including the one to rewrite the sector, could be help off for some few minutes. I say that, knowing that there is at least some similar code working for network attached drives, which seem to survive a brief network issue.

Telling the user a write intent bitmap is needed and making use of it sound at all practical as a use for some existing code?

Quoted 3 lines

You need to look for, or ask for, such variables at the scsi/sata layer.

The need for a delay between timeout and rewrite

Signature · 9 lines
Bill Davidsen <[email protected]>
 "We can't solve today's problems by using the same thinking we
  used in creating them." - Einstein


--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html