[DRBD-user] Multiple volumes per resource - replication consistency

[Anton Polyakov]

Hi Guys
 
Happy new year!
 
I would like to have two volumes (say /dev/data and /dev/binlog) which are used by the same application to be synchronized from my primary node to the other node (bin logs and data dir of MySQL). In DRBD docs it says that resource is a replication group containing multiple volumes and “DRBD ensures write fidelity across all volumes in the resource.”. 
MySQL writes data to bin log and data log volumes and data dir contains snapshot while binlog contains dirty segments. There is a relation between the two - binlogs periodically merged back to data volume and purged, so to recover on the other side data volume must contain recent enough snapshot such that binlogs can be applied on top of it.
Following that I need two volumes to maintain write causality and be replicated strictly together – i.e. avoid situation when one volume lags or doesn’t replicate while the other one does, otherwise when I failover on the other side I will observe inconsistent picture. 

I put the following configuration to DRBD:

resource mysql_resource {
       startup {
               wfc-timeout     1;
        }
       disk {
               no-disk-flushes;
               no-md-flushes;
               fencing resource-and-stonith;
               on-io-error call-local-io-error;
               disk-timeout 0;
        }
       volume 0 {
               device    /dev/drbd1;
               disk      /dev/data;
               meta-disk internal;
        }
        volume 1 {
               device    /dev/drbd2;
               disk      /dev/binlog;
               meta-disk internal;
        }
 
        on ip-10-15-0-135 {
               address   10.15.0.135:1120;
        }
        on peer-host {
               address   10.15.1.61:1120;
        }
}
 
Now two questions:
Can I achieve such consistency guarantees by using DRBD resource with two volumes?
If yes and a resource is a single logical replication unit (in a sense of described guarantees), why are GIs different per volume? Can I have situations when only one volume is disconnected or needs to resync? Can I have situations when one volume is Priomary/Secondary while the other is Secondary/Primary?
More generally, what would be the motivation to have volumes grouped into resources if synchronization still happens on volume level? (Or maybe I am confused since documentation does refer to resources being unit of synchronization (like sync bitmask is per reaource), but get/set -gi accepts volume)
 
Anton Polyakov
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20190101/0ca19869/attachment-0001.htm>