[DRBD-user] drbd related scsi-errrors?

Lars Ellenberg Lars.Ellenberg at linbit.com
Mon May 17 19:12:52 CEST 2004

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


/ 2004-05-17 17:12:06 +0200
\ Bernd Schubert:
Content-Description: signed data
> Hello,
> 
> I'm pretty new to drbd and we are currently setting up a new server, so I have 
> not so much experience with drbd.
> 
> Well, last week we observed some strange scsi-errors during a synchronization 
> with the failover server. Here are some more details:
> 
> main server: 
> 	- connected with a transtec 5008 scsi/ide raid system
> 	- three partitions (240GB out of 2 TB) are connected via drbd to a failover 
> server
> 
> failover server: 
> 	- 250 GB ide disk
> 
> The synchronization is initiated by setting down the network between both 
> systems, stopping drbd on the server, setting eth2 up again and finally 
> restarting drbd.
> 
> Here are some funny results when the scsi errors happen:
> 
>         - With my original drbd configuration this only happens when all 3 
> partions are synchronized simultaneously. During a synchronizion of only one 
> or two partitions it doesn't happen.
> 
>         - When I set the maximum synchronization speed of two partitions to 
> only 5MB/s and leave the third maximum speed at 100MB/s, it also doesn't 
> happen.
> 
>         - After scratching my head for a while and overthinking the problem, I 
> had the idea that the scsi-errors might occur due to the nice value of -18 
> for the synchronization processes (suggested drbd-configuration value). So I 
> went ahead and set this value to -5, which makes the scsi-errors also 
> disappear.

> Any thoughts about it? I'm also in contact with Justin T. Gibbs (the adaptec 
> scsi maintainer).

scsi errors are way below drbd. the usage pattern and heavy load on the
scsi bus caused by three drbd devices concurrently syncing obviously
triggers some corner case somewhere in the scsi stack.

maybe you should leave the other sync parameters alone, but use
different "sync-group"s, so you ar not thrashing the scsi device
that much.  sync then will happen one device at a time.

> Thanks,
> 	Bernd
> 
> 
> PS: drbd is 0.6.12



More information about the drbd-user mailing list