[DRBD-user] Some info

Adam Goryachev mailinglists at websitemanagers.com.au
Wed Oct 11 21:22:28 CEST 2017

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

On 12/10/17 06:14, Gandalf Corvotempesta wrote:
> So, let's assume a raid -> drbd -> lvm
> starting with a single RAID1, what If I would like to add a second
> raid1 converting the existing one to a RAID10 ? drbdadm resize would
> be enoguh ?
Correct, assuming you can convert the raid1 to raid10. You might need to 
start with a 2 device RAID10, best to check that procedure now and 
ensure mdadm will properly support this.
> keeping lvm as the upper layer would be best, I think, because will
> allow me to create logical volumes, snapshot and so on.
You can also do that with raid + lvm + drbd... you just need to create a 
new drbd as you add a new LV, and also resize the drbd after you resize 
the LV.
> what happens if local raid totally fails ? the upper layer will stay
> up thanks to DRBD fetching data from the other node ?
If both drives fail on one node, then raid will pass the disk errors up 
to DRBD, which will mark the local storage as down, and yes, it will 
read all needed data from remote node (writes are always sent to the 
remote node). You would probably want to migrate the remote node to 
primary as quickly as possible, and then work on fixing the storage.
> is "raid -> drbd -> lvm" a standard configuration or something bad? I
> don't want to put in production something "custom" and not supported.
Yes, it is not some bizarre configuration that has never been seen 
before. You also haven't mentioned the size of your proposed raid, nor 
what size you are planning on growing it to?

> How to prevent splitbrains ? Would be enough to bond the cluster
> network ? Any qdevice or fencing to configure ?
Yes, you will always want multiple network paths between the two nodes, 
and also fencing. bonding can be used to improve performance, but you 
should *also* have an additional network or serial or other connection 
between the two nodes which is used for fencing.

> 2017-10-11 21:07 GMT+02:00 Adam Goryachev <mailinglists at websitemanagers.com.au>:
>> On 12/10/17 05:10, Gandalf Corvotempesta wrote:
>>> Previously i've asked about DRBDv9+ZFS.
>>> Let's assume a more "standard" setup with DRBDv8 + mdadm.
>>> What I would like to archieve is a simple redundant SAN. (anything
>>> preconfigured for this ?)
>>> Which is best, raid1+drbd+lvm or drbd+raid1+lvm?
>>> Any advantage by creating multiple drbd resources ? I think that a
>>> single DRBD resource is better for administrative point of view.
>>> A simple failover would be enough, I don't need master-master
>>> configuration.
>> In my case, the best option was raid + lvm + drbd
>> It allows me to use lvm tools to resize each exported resource as required
>> easily:
>> lvmextend...
>> drbdadm resize ...
>> However, the main reason was to improve drbd "performance" so that it will
>> use different counters for each resource instead of a single set of counters
>> for one massive resource.
>> BTW, how would you configure drbd + raid + lvm ?
>> If you do DRBD with a raw drive on each machine, then use raid1 on top
>> within each local machine, then your raw drbd drive dies, the second raid
>> member will not contain or participate with DRBD anymore, so the whole node
>> is failed. This only adds DR ability to recover the user data. I would
>> suggest this should not be a considered configuration at all (unless I'm
>> awake to early and am overlooking something).
>> Actually, assuming machine1 with disk1 + disk2, and machine2 with disk3 +
>> disk4, I guess you could setup drbd1 between disk1 + disk3, and a drbd2 with
>> disk2 + disk4, and then create raid on machine 1 with drbd1+drbd2 and raid
>> on machine2 with drbd1+drbd2 and then use the raid device for lvm. You would
>> need double the write bandwidth between the two machines. When machine1 is
>> primary, and a write for the LV, it will be sent to raid which will send the
>> write to drbd1 and also drbd2. Locally, they are written to disk1 + disk2,
>> but also those 2 x writes will need to send over the network to machine2, so
>> it can be written to disk3 (drbd1) and disk4 (drbd2). Still not a sensible
>> option IMHO.
>> The two valid options would be raid + drbd + lvm or raid + lvm + drbd (or
>> just lvm + drbd if you use lvm to handle the raid as well).
>> Regards,
>> Adam
>> _______________________________________________
>> drbd-user mailing list
>> drbd-user at lists.linbit.com
>> http://lists.linbit.com/mailman/listinfo/drbd-user

More information about the drbd-user mailing list