DRBD splited

[docent]

Hi everybody!

I have the next trouble. Sometime (I don't know when) the DRBD on my cluster splited. And now DRBD is looking like StandAlone on both nodes:

cl8:~# cat /proc/drbd
version: 8.3.2 (api:88/proto:86-90)
GIT-hash: dd7985327f146f33b86d4bff5ca8c94234ce840e build by root@oahu, 2009-09-28 16:27:31
 0: cs: ro:Primary/Unknown ds:UpToDate/DUnknown   r----
    ns:7796056 nr:19049180 dw:28454304 dr:1300228 al:4717 bm:671 lo:0 pe:0 ua:0 ap:0 ep:1 wo:b oos:1324028

cl9:~# cat /proc/drbd
version: 8.3.2 (api:88/proto:86-90)
GIT-hash: dd7985327f146f33b86d4bff5ca8c94234ce840e build by root@oahu, 2009-09-28 16:27:31
 0: cs:StandAlone ro:Primary/Unknown ds:UpToDate/DUnknown   r----
    ns:19065524 nr:10188604 dw:29785584 dr:2664612 al:31899 bm:1606 lo:0 pe:0 ua:0 ap:0 ep:1 wo:b oos:965720

PVE dosn't see this trouble and all VMs are continuing to work fine on both nodes.

 

[dietmar]

Please can you post DRBD related things to the DRBD mailing list - DRBD experts are there?

 

[docent]

I posted this message here because you should consider this situation in the future.

 

[dietmar]

Sure, but I am also subscribed to the DRBD mailing list ;-)

 

[dietmar]

I posted this message here because you should consider this situation in the future.

Both devices are read only, so nothing bad can happen (you will get an error message) - how do you think I should 'consider' that situation?

 

[docent]

Both devices are read only, so nothing bad can happen (you will get an error message) - how do you think I should 'consider' that situation?

This is so, but now I have to stop all virtual machines on one node, transfer them to another (about 1TB), and then restore the DRBD.
I do not know how long the cluster is in a vulnerable state. Cluster manager must show that there is a problem in storage.

May be DRBD volumes for VMs on top of LVM would be better?

 

[dietmar]

I do not know how long the cluster is in a vulnerable state. Cluster manager must show that there is a problem in storage.

The question is why DRBD goes offline? - that should not happen at all.

May be DRBD volumes for VMs on top of LVM would be better?

Why?

 

[dietmar]

I do not know how long the cluster is in a vulnerable state. Cluster manager must show that there is a problem in storage.

Instead you should configure DRBD to send notifications:

http://www.drbd.org/users-guide/s-configure-split-brain-behavior.html#s-split-brain-notification