[DRBD-user] DRBD Replication

Felix Frank ff at mpexnet.de
Fri Jul 27 09:24:38 CEST 2012

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


On 07/26/2012 11:24 PM, Manoj Bhola wrote:
> Hi,
> I have a query and I wonder if someone could shed some light:
> I have a 2 node DRBD set up Primary/Secondary.
> Temporarily the secondary node is unavailable for a period of time. A
> single block of data on the Primary changes multiple times in that
> period. When the failed node comes back online and replication starts,
> for that block of data are all changes sent to the secondary node or
> just the last change?

DRBD cannot sync multiple writes during a resync operation. Nor does it
forward a "change", it just copies the whole block.

The syncronization works in one of two ways:
- QuickSync: The UpToDate node has kept a record of which blocks are
"dirty" locally and must be copied to the peer
- full sync: Contents of each block are copied to the peer

Apart from the quicksync bitmap, a node does not keep track of disk
writes. In order to replay all writes, an arbitrary amount of data would
need keeping somewhere in the metadata (or a limited ring buffer). This
is not done, though.


More information about the drbd-user mailing list