Single Node Ceph Recovery

[cclloyd]

If I were running Ceph (testing, no crucial data on this cluster) with only a single node (pools set to replicate OSD instead of host). If the host OS were to fail, but all the Bluestore OSDs were fine, can I import those into a new cluster after reinstalling the host OS? Or is the data lost forever?

 

[Alwin]

They are lost, as the MON DB is located on the OS disk.

 

[cclloyd]

They are lost, as the MON DB is located on the OS disk.

Can that be backed up to a non ceph disk and restored to a reinstalled OS?

 

[Alwin]

The state of the cluster is written to that DB and that is always changing. If there is no crucial data on it, then why bother?

 

[FourtyTwo]

Sorry for necroposting but what will happen if we improve ccloyd's question a bit:

First node is a ceph storage with 4 HDD's + some OS disk.
Second node is monitor only node with OS disk and no storage.

Question: can we somehow recover first node's OSD's if first node's OS disk will fail?

 

[Alwin]

Question: can we somehow recover first node's OSD's if first node's OS disk will fail?

A Ceph cluster hast 3x MONs (quorum). If one fails, the others will still have that data. You can easily test this in a virtual setup.

 

[FourtyTwo]

So if first node fail with OS disk i just need standart steps to recover it?:
1. Remove from cluster
2. Change OS disk, reinstall system, configure
3. Add it back to cluster and recreate MON
4. MON will be synced from 2-nd node (MON only) and i could run old OSD's

Am i right?

 

[Alwin]

Yes, that's it, in an rough outline.

 

[FourtyTwo]

Thank you very much for good and fast answer!