[Box Backup] Restoring from a hardware failure

[Ben Summers]

On 22 Jul 2005, at 10:01, Dennis Speekenbrink wrote:

> Per Thomsen wrote:
>
>
>> All,
>> I'm using a hardware raid card on my bbstored server, and I had a  
>> drive failure the other day. Switched out the drive, and  
>> everything seemed to work fine.
>>
>> However, one of the clients was unable to back up. I found the  
>> following in one of the backupdirs under that account (output of  
>> 'ls -li'):
>>
>>
> Going for a hardware RAID setup myself for my new server, this  
> worries me.
> Could you please offer a few more details on the original setup and  
> what seemed to cause the crash.
>
> It would seem to me that any hardware RAID setup (except  
> 0:striping) is _invented_ to prevent these issues from arising, no?
>
> Thanks for any insights.

A RAID device can only store the data sent to it by the host. If you  
switch the machine off in the middle of writing something important,  
you'll still get corruption.

This is why you also need a tolerant filing system. Journalling at  
the minimum. And even then, you have to hope there aren't bugs in the  
filing system code which might cause bad stuff to be written.


>
> For your problem, it would seem to me that removing (or moving, try  
> never to destroy data unless your sure you won't need it anymore)  
> would cause the client to re-send all original data to the server.   
> But like Ben posted, let bbstoreaccounts do it's thing before re- 
> allowing the client's connection.

ESPECIALLY if you've told the client to back up a directory where  
there used to be stuff, and there isn't yet because it needs to be  
restored. You wouldn't want the client to mark everything deleted on  
the server, would you?

[ I wanted to add a feature to stop this accident, but there was such  
an uproar on the list I didn't. ]


> The only way I see this going wrong, is when the client has  
> dataloss now, it will not be available on the server yet.

Yes.

Ben