[DRBD-user] Question about 3 nodes, three way replication with WAN link

[Panayiotis Fafakos]

Dear all,

I would like to ask, is if we can have three DRBD nodes like below:

          Main Site                                               DR Site
          Server A (primary)     Server B (primary) Server C (offsite 
secondary)
=============================== ================================ 
============================
Layer 1  Physical Disks Physical Disks Physical Disks
           |                      | |
Layer 2  LVM LVM LVM
           | |                                |
Layer 3  DRBD <--*Protocol C*---> DRBD                             DRBD 
(with no DRBD partner)
           | |                                |
Layer 4 *Stacked DRBD* *Stacked DRBD* <-*WAN/Protocol A*-> *Stacked DRBD*
           | |                                |
Layer 5 LVM                    Inactive LVM Inactive LVM
| |                                |
Layer 6  Filesystem             Unmounted Filesystem Unmounted Filesystem
| |                                |
Layer 7  Applications+VMs NotUsed/Standby NotUsed/Standby


This use-case is like the diagram shown and discussed in the link below
https://linbit.com/drbd-user-guide/users-guide-drbd-8-4/#s-three-way-repl
but the data are replicated to the third server using the secondary 
server (ServerB) as a primary read-only server.
This way, even if the writing application will block, it will only 
affect ServerB while ServerA will continue running at full speed,
this is what we actually hope for at least.

Notes:
a) Layer 3 *DRBD *is *Primary *on *all servers*.
b) Layer 4 *Stacked DRBD* is *Primary *on both *ServerA *and *ServerB* 
(allow-two-primaries). *Stacked DRBD* on *ServerC***is *Secondary* and 
replicates from *ServerB*.
c) ServerB will be used for off-site replication to ServerC, so that the 
slow speed of the WAN link used for replication does not affect ServerA.

Since ServerB and ServerC will be used as backup servers,
actually as hot stand-by in read-only mode(LVM will be in 
"not-available/unavailable" mode),
_is there any possibility that we may have any data corruption?_

/I believe that there should not be any data corruption since ServerB 
will only act as read-only server and will be used only for the off-site 
replication.//
/
We understand that in the above scenario Stacked DRBD metadata and disk 
data are only written from ServerA
since ServerA is the only server that has Layer5, Layer6 and Layer7 
mounted, correct?

If ServerB should ever become Primary for the LVM partition in Layer5,
we would first unmount LVM partition in Layer5, ,
the WAN replication would be switched from between ServerB and ServerC 
to between ServerA and ServerC,
and afterwards ServerB would mount Layer5 LVM, Layer6 Filesystem and 
Layer7 Applications+VMs.
Then ServerB would be the only server to write data, and ServerA would 
send data to ServerC...

Do you think we are bound to have any corruption in the above scenario,
if we make sure that Layer5 LVM and the above Layers will never be 
mounted on more that one servers at a time?

Thank you,

Panos.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20220204/b4d53317/attachment.htm>