0

My startup is thinking about adopting CEPH for block storage (and other features it provides). Most of our infrastructure runs on public cloud (DO,GCP,AWS,etc...).

On public cloud data storage you usually have a bunch of guarantees with regards to data corruption.(RAID etc).

Since these guarantees are provided and managed by our cloud provider is it valid to do a ceph installation without replication?

Note: HA isn't a major concern. data loss however is.

To clarify:

This question could also be rephrased as:

Assume a perfect storage device (i.e it doesn't fail nor suffer any hardware level corruption) because in practice that's what you buy when you purchase a storage volume from a public cloud provider)

Would CEPH still need replication?

put another way: if the underlying storage doesn't fail and I don't mind the data being offline for a while.

Do I still need replication?

Thanks

mohamed emad
  • 1
  • 2
  • You typically deploy Ceph to realise a distributed storage cluster without a single point of failure for reliability, availability, scalability and performance. You need replication for all of those aspects. _-_ Your *"clarifications"* say that you're willing to assume that you need *neither* availability *nor* any added reliability - then why set up Ceph at all? – HBruijn Aug 11 '22 at 15:30
  • convenient APIs and CEPH RBDs . It integrates well with LXD and other openstack solutions. And it allows me to evade vendor lock-in when writing my solution. – mohamed emad Aug 12 '22 at 15:40

1 Answers1

2

Well you don't mention specifics about the exact service or guarantees involved but my first question would be why aren't you just using the public cloud provider's native storage - most of them offer all you'd need; block, file, object.

But if you're dead-set on doing this then no, do not skip replication, not if you like your data anyway - for a start sometimes public cloud services just go off air for a while, if this happens with no replication then you have just lost access to data and it may be corrupt when it comes back. So yes, please use replication.

Chopper3
  • 100,240
  • 9
  • 106
  • 238
  • > it may be corrupt when it comes back doesn't the Public cloud SLA guarantee this doesn't happen? i.e: no data corruption from their end is allowed? because they use replication and what not? – mohamed emad Aug 11 '22 at 11:12
  • Oh man I wish that was true for what you're suggesting, for their own-managed file/object storage yes, for just block storage the onus is on the customer to make sure their data is backed up as it's just blocks of data to them. – Chopper3 Aug 11 '22 at 12:18
  • yes but the blocks I put in are the same blocks I read out. i.e: they guarantee the blocks won't change unless **I** change them. I wonder is there any other source of "block corruption" here other than just normal hardware failures? – mohamed emad Aug 11 '22 at 14:09