[DRBD-user] md5 sums mismatch

Łukasz Oleś lukaszoles at gmail.com
Mon Jul 18 12:08:59 CEST 2011

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


W dniu 18 lipca 2011 10:38 użytkownik Felix Frank <ff at mpexnet.de> napisał:
> Hi,
>
> you should keep stuff on-list :-)

yeah, wrong button :)

> On 07/18/2011 10:29 AM, Łukasz Oleś wrote:
>> W dniu 11 lipca 2011 10:45 użytkownik Felix Frank <ff at mpexnet.de> napisał:
>>> On 07/11/2011 10:32 AM, Łukasz Oleś wrote:
>>>> Online verify found 88 4k block out of sync!
>>>
>>> 350k out of 3TB?
>>>
>>> I'd venture to say this is what can result from unfortunate bit-flips
>>> somewhere along the way.
>>>
>>> Yes, these are bad to have. Yet, as far as I know, few systems can even
>>> protect you from data corruption happening on the way from RAM to HDD,
>>> so I disbelieve you're facing a huge problem.
>>
>> During last week we have made more tests. Firstly, we repeated the
>> test and again there where mismatches. What is more interesting
>> corrupted files were on source volume.
>
> Like I wrote - data can be corrupted between your CPU/Mem and HDD. It's
> really hard to guard against this.
> I've read that ZFS does have defences, but using it atop DRBD will prove
> difficult ;-)
>
>> Then we repeated test again but files were copied localy(not via
>> iscsi) and drbd was disconnected. Files were ok.
>>
>> Any ideas?
>
> Not really. Your iSCSI target cannot likely take the blame, since all it
> does is feeding data to DRBD. If your Secondary received sound data,
> that means that the iSCSI target fed sound data to DRBD.
>
> DRBD cannot be blamed either, because it does little more than hand down
> data to your HDD controller.
>
> I have to admit that it's strange, however. My "misfortune theory"
> doesn't really hold if only one machine is affected. How did you make
> sure you were writing sound data during your final test?

I'm copying files with known md5 sums

-- 
Łukasz Oleś



More information about the drbd-user mailing list