ceph pool of size 4 with min_size of 2 safe?

[BriRey]

Hi All,

I'm hoping for a concise, relatively authoritative answer:

Is it safe to have a ceph pool of size 4 and a min_size setting of 2? Or is it inherently risky from a "split-brain" concern (or any other) perspective?

We are running five nodes with five OSDs per node. All five nodes are monitors and managers.

Some initial googling of this question seems to indicate that "split brain" is not a concern. . .

Ceph does not have a concept of "osd quorum" or "electing a primary
PG". The mons are in a PAXOS quorum, and the mon leader decides which
OSD is primary for each PG. No need to worry about a split OSD brain.

. . . nevertheless the overall judgement on whether a size 4, min_size 2 is advisable (or inadvisable for any reason) would be greatly appreciated.

Thanks!

--B

 

[Alwin]

Is it safe to have a ceph pool of size 4 and a min_size setting of 2? Or is it inherently risky from a "split-brain" concern (or any other) perspective?

Size is the number of replicas Ceph tries to achieve for a healthy cluster. Min_size is the number of replicas, when Ceph stops writing data. Read operations are still possible. Depending on your crushmap (eg. failure domains) it may or may not lead to read-only or even data loss.

 

[BriRey]

Size is the number of replicas Ceph tries to achieve for a healthy cluster. Min_size is the number of replicas, when Ceph stops writing data. Read operations are still possible. Depending on your crushmap (eg. failure domains) it may or may not lead to read-only or even data loss.

Thank you for your reply. I am familiar with those facts. Should this reply be read to say as well that there is no further concern or worry ("split-brain" issues, etc.)? Put another way, there is no inherent potential issues with a 4/2 (size/min_size)?

 

[spirit]

the quorum is managed by monitor deamon (so you need 3 monitors minimum to avoid split brain).

for replication/sizz, you can use 2,3,4 as you want. (it's more a security of how much disk you can loose at the same time)