RE: RAID showing all devices as spares after partial unplug

[Mike Hartman <mike@hartmanipulation.com>]

On Sat, Sep 17, 2011 at 11:07 PM, Mike Hartman
<mike@hartmanipulation.com> wrote:
> Yikes. That's a pretty terrifying prospect.
>
> On Sat, Sep 17, 2011 at 10:57 PM, Jim Schatzman
> <james.schatzman@futurelabusa.com> wrote:
>> Mike-
>>
>> See my response below.
>>
>> Good luck!
>>
>> Jim
>>
>>
>> At 07:34 PM 9/17/2011, Mike Hartman wrote:
>>>On Sat, Sep 17, 2011 at 9:16 PM, Jim Schatzman
>>><james.schatzman@futurelabusa.com> wrote:
>>>> Mike-
>>>>
>>>> I have seen very similar problems. I regret that electronics engineers cannot design more secure connectors. eSata connector are terrible - they come loose at the slightest tug. For this reason, I am gradually abandoning eSata enclosures and going to internal drives only. Fortunately, there are some inexpensive RAID chassis available now.
>>>>
>>>> I tried the same thing as you. I removed the array(s) from mdadm.conf and I wrote a script for "/etc/cron.reboot" which assembles the array, "no-degraded". Doing this seems to minimize the damage caused by drives prior to a reboot. However, if the drives are disconnected while Linux is up, then either the array will stay up but some drives will become stale or the array will be stopped. The behavior I usually see is that all the drives that went offline now become "spare".
>>>>
>>>
>>>That sounds similar, although I only had 4/11 go offline and now
>>>they're ALL spare.
>>>
>>>> It would be nice if md would just reassemble the array once all the drives come back online. Unfortunately, it doesn't. I would run mdadm -E against all the drives/partitions, verifying that the metadata all indicates that they are/were part of the expected array.
>>>
>>>I ran mdadm -E and they all correctly appear as part of the array:
>>>
>>>for d in /dev/sd[cdfhjklmn]1 /dev/md1p1 /dev/md3p1; do echo $d; mdadm
>>>-E $d | grep Role; done
>>>
>>>/dev/sdc1
>>>   Device Role : Active device 5
>>>/dev/sdd1
>>>   Device Role : Active device 4
>>>/dev/sdf1
>>>   Device Role : Active device 2
>>>/dev/sdh1
>>>   Device Role : Active device 0
>>>/dev/sdj1
>>>   Device Role : Active device 10
>>>/dev/sdk1
>>>   Device Role : Active device 7
>>>/dev/sdl1
>>>   Device Role : Active device 8
>>>/dev/sdm1
>>>   Device Role : Active device 9
>>>/dev/sdn1
>>>   Device Role : Active device 1
>>>/dev/md1p1
>>>   Device Role : Active device 3
>>>/dev/md3p1
>>>   Device Role : Active device 6
>>>
>>>But they have varying event counts (although all pretty close together):
>>>
>>>for d in /dev/sd[cdfhjklmn]1 /dev/md1p1 /dev/md3p1; do echo $d; mdadm
>>>-E $d | grep Event; done
>>>
>>>/dev/sdc1
>>>         Events : 1756743
>>>/dev/sdd1
>>>         Events : 1756743
>>>/dev/sdf1
>>>         Events : 1756737
>>>/dev/sdh1
>>>         Events : 1756737
>>>/dev/sdj1
>>>         Events : 1756743
>>>/dev/sdk1
>>>         Events : 1756743
>>>/dev/sdl1
>>>         Events : 1756743
>>>/dev/sdm1
>>>         Events : 1756743
>>>/dev/sdn1
>>>         Events : 1756743
>>>/dev/md1p1
>>>         Events : 1756737
>>>/dev/md3p1
>>>         Events : 1756740
>>>
>>>And they don't seem to agree on the overall status of the array. The
>>>ones that never went down seem to think the array is missing 4 nodes,
>>>while the ones that went down seem to think all the nodes are good:
>>>
>>>for d in /dev/sd[cdfhjklmn]1 /dev/md1p1 /dev/md3p1; do echo $d; mdadm
>>>-E $d | grep State; done
>>>
>>>/dev/sdc1
>>>          State : clean
>>>   Array State : .A..AA.AAAA ('A' == active, '.' == missing)
>>>/dev/sdd1
>>>          State : clean
>>>   Array State : .A..AA.AAAA ('A' == active, '.' == missing)
>>>/dev/sdf1
>>>          State : clean
>>>   Array State : AAAAAAAAAAA ('A' == active, '.' == missing)
>>>/dev/sdh1
>>>          State : clean
>>>   Array State : AAAAAAAAAAA ('A' == active, '.' == missing)
>>>/dev/sdj1
>>>          State : clean
>>>   Array State : .A..AA.AAAA ('A' == active, '.' == missing)
>>>/dev/sdk1
>>>          State : clean
>>>   Array State : .A..AA.AAAA ('A' == active, '.' == missing)
>>>/dev/sdl1
>>>          State : clean
>>>   Array State : .A..AA.AAAA ('A' == active, '.' == missing)
>>>/dev/sdm1
>>>          State : clean
>>>   Array State : .A..AA.AAAA ('A' == active, '.' == missing)
>>>/dev/sdn1
>>>          State : clean
>>>   Array State : .A..AA.AAAA ('A' == active, '.' == missing)
>>>/dev/md1p1
>>>          State : clean
>>>   Array State : AAAAAAAAAAA ('A' == active, '.' == missing)
>>>/dev/md3p1
>>>          State : clean
>>>   Array State : .A..AAAAAAA ('A' == active, '.' == missing)
>>>
>>>So it seems like overall the array is intact, I just need to convince
>>>it of that fact.
>>>
>>>> At that point, you should be able ro re-create the RAID. Be sure you list the drives in the correct order. Once the array is going again, mount the resulting partitions RO and verify that the data is o.k. before going RW.
>>>
>>>Could you be more specific about how exactly I should re-create the
>>>RAID? Should I just do --assemble --force?
>>
>>
>>
>>  -->  No. As far as I know, you have to use "-C"/"--create".  You need to use exactly the same array parameters that were used to create the array the first time. Same metadata version. Same stripe size. Raid mode the same. Physical devices in the same order.
>>
>> Why do you have to use "--create", and thus open the door for catastropic error?? I have asked the same question myself. Maybe, if more people ping Neil Brown on this, he may be willing to find another way.
>>
>>
>>


Is there any way to construct the exact create command using the info
given by mdadm -E? This array started as a RAID 5 that was reshaped
into a 6 and then grown many times, so I don't have a single original
create command lying around to reference.

I know the devices and their order (as previously listed) - are all
the other options I need to specify part of the -E output? If so, can
someone clarify how that maps into the command?

Here's an example output:

mdadm -E /dev/sdh1

/dev/sdh1:
         Magic : a92b4efc
       Version : 1.2
   Feature Map : 0x1
    Array UUID : 714c307e:71626854:2c2cc6c8:c67339a0
          Name : odin:0  (local to host odin)
 Creation Time : Sat Sep  4 12:52:59 2010
    Raid Level : raid6
  Raid Devices : 11

 Avail Dev Size : 2929691614 (1396.99 GiB 1500.00 GB)
    Array Size : 26367220224 (12572.87 GiB 13500.02 GB)
 Used Dev Size : 2929691136 (1396.99 GiB 1500.00 GB)
   Data Offset : 2048 sectors
  Super Offset : 8 sectors
         State : clean
   Device UUID : 384875df:23db9d35:f63202d0:01c03ba2

Internal Bitmap : 2 sectors from superblock
   Update Time : Thu Sep 15 05:10:57 2011
      Checksum : f679cecb - correct
        Events : 1756737

        Layout : left-symmetric
    Chunk Size : 256K

  Device Role : Active device 0
  Array State : AAAAAAAAAAA ('A' == active, '.' == missing)

Mike


>>>>
>>>> Jim
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> At 04:16 PM 9/17/2011, Mike Hartman wrote:
>>>>>I should add that the mdadm command in question actually ends in
>>>>>/dev/md0, not /dev/md3 (that's for another array). So the device name
>>>>>for the array I'm seeing in mdstat DOES match the one in the assemble
>>>>>command.
>>>>>
>>>>>On Sat, Sep 17, 2011 at 4:39 PM, Mike Hartman <mike@hartmanipulation.com> wrote:
>>>>>> I have 11 drives in a RAID 6 array. 6 are plugged into one esata
>>>>>> enclosure, the other 4 are in another. These esata cables are prone to
>>>>>> loosening when I'm working on nearby hardware.
>>>>>>
>>>>>> If that happens and I start the host up, big chunks of the array are
>>>>>> missing and things could get ugly. Thus I cooked up a custom startup
>>>>>> script that verifies each device is present before starting the array
>>>>>> with
>>>>>>
>>>>>> mdadm --assemble --no-degraded -u 4fd7659f:12044eff:ba25240d:
>>>>>> de22249d /dev/md3
>>>>>>
>>>>>> So I thought I was covered. In case something got unplugged I would
>>>>>> see the array failing to start at boot and I could shut down, fix the
>>>>>> cables and try again. However, I hit a new scenario today where one of
>>>>>> the plugs was loosened while everything was turned on.
>>>>>>
>>>>>> The good news is that there should have been no activity on the array
>>>>>> when this happened, particularly write activity. It's a big media
>>>>>> partition and sees much less writing then reading. I'm also the only
>>>>>> one that uses it and I know I wasn't transferring anything. The system
>>>>>> also seems to have immediately marked the filesystem read-only,
>>>>>> because I discovered the issue when I went to write to it later and
>>>>>> got a "read-only filesystem" error. So I believe the state of the
>>>>>> drives should be the same - nothing should be out of sync.
>>>>>>
>>>>>> However, I shut the system down, fixed the cables and brought it back
>>>>>> up. All the devices are detected by my script and it tries to start
>>>>>> the array with the command I posted above, but I've ended up with
>>>>>> this:
>>>>>>
>>>>>> md0 : inactive sdn1[1](S) sdj1[9](S) sdm1[10](S) sdl1[11](S)
>>>>>> sdk1[12](S) md3p1[8](S) sdc1[6](S) sdd1[5](S) md1p1[4](S) sdf1[3](S)
>>>>>> sdh1[0](S)
>>>>>>       16113893731 blocks super 1.2
>>>>>>
>>>>>> Instead of all coming back up, or still showing the unplugged drives
>>>>>> missing, everything is a spare? I'm suitably disturbed.
>>>>>>
>>>>>> It seems to me that if the data on the drives still reflects the
>>>>>> last-good data from the array (and since no writing was going on it
>>>>>> should) then this is just a matter of some metadata getting messed up
>>>>>> and it should be fixable. Can someone please walk me through the
>>>>>> commands to do that?
>>>>>>
>>>>>> Mike
>>>>>>
>>>>>--
>>>>>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
>>
>>
>
--
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