Re: Paranoia about DegradedArray
On Wed, 29 Oct 2008 17:58:56 -0400, Hal Vaughan wrote:
> On Wednesday 29 October 2008, Hendrik Boom wrote:
>>
>> I'm a bit surprised that none of the messages identifies the other
>> drive, /dev/hdc3. Is this normal? Is that information available
>> somewhere besides the sysadmin's memory?
>
> Luckily it's been at least a couple months since I worked with a
> degraded array, but I *thought* it listed the failed devices as well. It
> looks like the device has not only failed but been removed -- is there a
> chance you removed it after the failure, before running this command?
No. I did not explicitly fail it or remove it. There must have been
some automatic mechanism that did.
>>
>> So presumably the thing to do is
>> mdadm --fail /dev/md0 /dev/hdc3
>> mdadm --remove /dev/md0 /dev/hdc3
>> and then
>> mdadm --add/dev/md0 /dev/hdc3
>
> I think there's a "--readd" that you have to use or something like that,
> but I'd try --add first and see if that works. You might find that hdc3
> has already failed and, form the output above, looks like it's already
> been removed.
In the docs, re-add is specified as something to use if a drive has been
removed *recently*, and then it writes all the blocks that were to have
been written while it was out -- a way of doing an update instead of s
full copy. It doesn't seem relevant in this case.
>
>> Is the --fail really needed in my case? the --detail option seems to
>> have given /dev/hdc3 the status of "removed" (although it failed to
>> mention is was /dev/hdc3).
>
> I've had trouble with removing drives if I didn't manually fail them.
> Someone who knows the inner workings of mdadm might be able to provide
> more information on that.
I wonder if /dev/hdc3 still needs to be manually failed. I wonder if it
is even possible to fail a removed drive...
>
> Yes, paranoia is a good thing in system administration. It's kept me
> from severe problems previously!
And paranoia will make sure I have two complete backups before I actually
do any of this fixup.
- hendrik
>
>
> Hal
Reply to: