2

On fresh drbd setup (no filesystem on drbd0, just synced) I have invalidated "node2" state and full sync started:

root@node1:~# drbdadm invalidate-remote drbd0
root@node1:~# drbd-overview
 0:drbd0/0  SyncSource Secondary/Secondary UpToDate/Inconsistent
        [>....................] sync'ed:  0.2% (153188/153464)Mfinish: 1:04:59 speed: 40,180 (40,180) K/sec

and having drbd0 sync from node1 to node2 in progress I have set it primary on node1 for a moment, created ext4, and set it back to standby and then I went to node2 (inconsitent) and made it as primary without any warnings (sic!)

root@node2:~# drbd-overview
 0:drbd0/0  SyncTarget Secondary/Secondary Inconsistent/UpToDate
        [====>...............] sync'ed: 26.7% (112500/153464)Mfinish: 0:47:58 speed: 40,000 (40,296) want: 74,040 K/sec
root@node2:~# drbdadm primary drbd0
root@node2:~# drbd-overview
 0:drbd0/0  SyncTarget Primary/Secondary Inconsistent/UpToDate
        [====>...............] sync'ed: 27.1% (111912/153464)Mfinish: 0:50:37 speed: 37,724 (40,252) want: 75,600 K/sec
root@node2:~#

next, I mounted the ext4 from inconsistent device filesystem without any problems: root@node2:~# mount /dev/drbd0 /mnt

then in that /mnt I created a few thousands of rsa key pairs (all the time on inconsistend drbd0 on node2 and sync from node1 in progres)

then I unmounted /mnt on node2 and mounted it on node1 and I see there all the files (sync node1->node2 still in progress) which I have created on the inconsistent state drbd0 on node2.

next I created another thousands of files on node1 and remounted /mnt to node2 and I see all the files which were created on node1 and node2 (during full sync on drbd0 on node2 in porgress = inconsistent state of it)

I have checked all files checksums in each step and all were ok.

How it works?? Is that by design? Why there was no split-brain state when I activated drbd0 on inconsistent device? What is the real state of drbd0 device on inconsistent node that is can be used as a primary for read/write? Was the resync done in both ways: from node1-node2 full sync (as node2 was invalidated in first step) and from inconsistent node node2 -> node1 the changes/files I created when I mounted it there?

After sync completed, final status is ok but I am still confused... 

root@node2~# drbd-overview
 0:drbd0/0  Connected Primary/Secondary UpToDate/UpToDate /mnt ext4 148G 1.7G 145G 2%

thank you in advance for any explaination,

best regards,

DonJ
  • 351
  • 2
  • 18

1 Answers1

2

Completely by design. When you made the node with the Inconsistent disk primary it then is essentially reading and writing across the network to the peer with the UpToDate disk.

If you were to break the network connection while the Inconsistent disk was primary, things would have gone south real quick. ;-)

The same also applies if the disk fails on the primary node. DRBD will switch to diskless and then push all reads and writes to the peer node (assuming the peer has a disk which is UpToDate. This can help avoid a service outage due to local disk failure.

Dok
  • 396
  • 1
  • 6