Good luck, till it crashes it's fine. There are most likely intermittent network problems that don't allow Ceph to work properly. Anyhow, if you don't have a physical private network, I can't recommend it for Ceph. IMO, a better alternative is ZFS with storage replication.
Ceph down if one node is down
- Thread starter bizzarrone
- Start date
-
- Tags
- ceph cluster proxmox 5.3
What luck? What's wrong with 3 node setup?
Network problem's on 3 different setups?
I have a dedicated 10gbps switch in one off the Hetzner setups and the same problem occurs.
Nothing wrong with a three node setup.
Your statement didn't sound like all of them are at peril.
Dedicated for your cluster alone? Or shared with other customers.
In any case, IO is halted when min_size is not met. With one node failing the min_size is still good and IO will continue. The crushmap provided would not hinder the data placement. So I believe that you have network issue and be it just the configuration.
Thank you Alwin for answering
The 10gbs is only for my setup.
I reduce the min_size to 1 and now disconnecting one node is wokring as expected.
A 3/1 configuration is working on all 3 clusters with one node down.
My question is - why a 3/2 configuration is blocking IO?
The 10gbs is only for my setup.
I reduce the min_size to 1 and now disconnecting one node is wokring as expected.
A 3/1 configuration is working on all 3 clusters with one node down.
My question is - why a 3/2 configuration is blocking IO?
I'd still argue its the network setup. With min_size 1 each node can also just write locally.