[DRBD-user] Integrity checksum support

[Theodor Ratkin]

Hi, we are wondering if DRBD has (or is planning to) support data 
integrity checksums, in the likeness of CEPH.

Like most, we are concerned about silent data corruption, and while we 
could leverage checksum methods provided by ZFS or LVM (and others), it 
feels like DRBD is a more appropriate place for it.

For block-level checksums, in cause if a failure, the read operation 
will return I/O error and the node will be switched to diskless by DRBD, 
and it will require manual recovery, possibly forcing full resync.

If checksums were done by DRBD instead, it could just compare them, and 
in case of a failure, mark the device as OUTDATED and resync it from the 
network mirrors.

So, could anyone please shed some light on this matter? I'm aware of 
Checksum-based synchronization, but it doesn't seem like it was made for 
integrity, it's just another sync algorithm that is used for better 
performance it seems.

What if there's a silent corruption on the primary node? Will DRBD grab 
good data from the secondary nodes in this case? In my understanding, 
DRBD always syncs data from primary to secondary, never the other way 
around.