Hello !
Can someone tell me, how md multipathing works, when a drive fails
completly ?
Does this only work with raid-autodetection ?
When no autodetection is done and a drive is missing, would a raidstart
kill the raid, since the drives are now available with other devices (sda
instead of former sdb...) ?
Oktay Akbal
On 2002-09-09T15:08:31,
Oktay Akbal <[email protected]> said:
> Can someone tell me, how md multipathing works, when a drive fails
> completly ?
Well, if the drive (not the path to it) fails _completely_, it won't be
detected by the md autostart (as it can't find the md superblock).
If it fails completely during runtime, all paths but the last one to it will
be disabled, as a drive failure can't be distinguished from a path failure in
the wonderful 2.4 error handling ;-)
But then, all requests send down the last path will fail, because the target
is broken, not the path.
In short, multipathing doesn't help a bit here; how could it?
> Does this only work with raid-autodetection ?
> When no autodetection is done and a drive is missing, would a raidstart
> kill the raid, since the drives are now available with other devices (sda
> instead of former sdb...) ?
I don't understand your question, sorry.
Sincerely,
Lars Marowsky-Br?e <[email protected]>
--
Immortality is an adequate definition of high availability for me.
--- Gregory F. Pfister
> > Does this only work with raid-autodetection ?
> > When no autodetection is done and a drive is missing, would a raidstart
> > kill the raid, since the drives are now available with other devices (sda
> > instead of former sdb...) ?
>
> I don't understand your question, sorry.
Example:
We have sda - sdb (8 drives) and setup up a raidtab to tell linux that
sda and sde are the same sdc - sdd etc.
Now for some random error the server restarts and the former sda (path to
that drive) is no longer available. So now we have sda,sdb...sdg.
We do not use autodetect, but raidstart to activate the raid.
now since the former sda is missing the raidtab does not reflect the
actual setup. The raidtab would read, that sda and sdb are the same
drive, which is not true in that case.
(The device-ordering would not be right for a real setup, but take it as
an example and assume sda-sde sdb-sdf...)
Would the superblock prevent the wrong use of devices ?
(With raid-configuration setup on top of multipathing ?)
Oktay Akbal
S-Tec Datenverarbeitung GmbH
Feuerbachstr. 8
68163 Mannheim
Tel: 0621-4185070
Fax: 0621-4185071
Oktay,
You should really be using devfs in a situation such as yours, where you
need device name consistency during reboots.
On Mon, 9 Sep 2002, Oktay Akbal wrote:
> > > Does this only work with raid-autodetection ?
> > > When no autodetection is done and a drive is missing, would a raidstart
> > > kill the raid, since the drives are now available with other devices (sda
> > > instead of former sdb...) ?
> >
> > I don't understand your question, sorry.
>
> Example:
>
> We have sda - sdb (8 drives) and setup up a raidtab to tell linux that
> sda and sde are the same sdc - sdd etc.
> Now for some random error the server restarts and the former sda (path to
> that drive) is no longer available. So now we have sda,sdb...sdg.
> We do not use autodetect, but raidstart to activate the raid.
>
> now since the former sda is missing the raidtab does not reflect the
> actual setup. The raidtab would read, that sda and sdb are the same
> drive, which is not true in that case.
>
> (The device-ordering would not be right for a real setup, but take it as
> an example and assume sda-sde sdb-sdf...)
>
> Would the superblock prevent the wrong use of devices ?
> (With raid-configuration setup on top of multipathing ?)
>
>
>
>
> Oktay Akbal
> S-Tec Datenverarbeitung GmbH
> Feuerbachstr. 8
> 68163 Mannheim
> Tel: 0621-4185070
> Fax: 0621-4185071
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
[-] Steve Mickeler [ [email protected] ]
[|] Todays root password is brought to you by /dev/random
[+] 1024D/9AA80CDF = 4103 9E35 2713 D432 924F 3C2E A7B9 A0FE 9AA8 0CDF
On 2002-09-09T15:46:15,
Oktay Akbal <[email protected]> said:
> Example:
>
> We have sda - sdb (8 drives) and setup up a raidtab to tell linux that
> sda and sde are the same sdc - sdd etc.
> Now for some random error the server restarts and the former sda (path to
> that drive) is no longer available. So now we have sda,sdb...sdg.
> We do not use autodetect, but raidstart to activate the raid.
>
> now since the former sda is missing the raidtab does not reflect the
> actual setup. The raidtab would read, that sda and sdb are the same
> drive, which is not true in that case.
>
> (The device-ordering would not be right for a real setup, but take it as
> an example and assume sda-sde sdb-sdf...)
>
> Would the superblock prevent the wrong use of devices ?
I hope so. But yes, your setup would break spectularly because the devices
moved and the raids wouldn't go online. This shouldn't happen with the
autostart feature, I think.
Sincerely,
Lars Marowsky-Br?e <[email protected]>
--
Immortality is an adequate definition of high availability for me.
--- Gregory F. Pfister
On Mon, 2002-09-09 at 14:50, Steve Mickeler wrote:
>
> Oktay,
>
> You should really be using devfs in a situation such as yours, where you
> need device name consistency during reboots.
>
There are a collection of nice tools that will run at boot up (eg from
initrd or from the rootfs mount) and generate a set of sun like /dev/
symlinks to the real scsi device by lun.
You don't need to play with experimental stuff like devfs
On Monday September 9, [email protected] wrote:
> > > Does this only work with raid-autodetection ?
> > > When no autodetection is done and a drive is missing, would a raidstart
> > > kill the raid, since the drives are now available with other devices (sda
> > > instead of former sdb...) ?
> >
> > I don't understand your question, sorry.
>
> Example:
>
> We have sda - sdb (8 drives) and setup up a raidtab to tell linux that
> sda and sde are the same sdc - sdd etc.
> Now for some random error the server restarts and the former sda (path to
> that drive) is no longer available. So now we have sda,sdb...sdg.
> We do not use autodetect, but raidstart to activate the raid.
raidstart is broken by design and cannot cope with devices that change
device number (whether name in /dev is preserved or not. There are
device numbers in the superblock which raidstart trusts).
This is one of the reasons that I wrote mdadm
http://www.cse.unsw.edu.au/~neilb/source/mdadm/
This affects all raid levels, not just multipath.
NeilBrown