Managed DB storage limit



GKE (on GCP) has a limit of 60TB per node.

Does your managed service on GCP have a similar limit? Ideally we would like to start off small and then add storage when needed without any limits.



(Raphael 'kena' Poss) #2

Hi @batman
thank you for your inquiry!
There is a limit on storage per node (the particular limit depends on the package you choose), but of course it is configurable and we’ll be happy to accommodate your growing size needs over time.
Does this answer? Is there anything specific you’d like to know?



Thanks for the fast reply.

I always thought data is replicated on every node? So say we had a DB of size100 TB, I thought that was 100 TB/node or 300 TB of storage for a 3 node cluster? Is that incorrect?

One of our internal systems has 1 Peta byte of storage. I am not sure how much we will need in the next 5 years but I need to know if we wanted to have a DB that needs storage to scale, we can achieve that with a managed solution?


(Raphael 'kena' Poss) #4

CockroachDB uses N:M replication - increasing the number of nodes and increasing the replication factor are two independent operations.

  • replication factor is about how many node failures you wish to tolerate
  • number of nodes is about overall horizontal scalability (performance)

It is common/customary to use a replication factor of 3 with many more nodes.

So if you have, say, 10 nodes with each node able to store 500GB of data, with replication factor 3 the total “useful” storage is 10x500/3 = ~1.6TB. With 20 nodes it would be ~3.3TB.

We commonly advise to increase storage/node if you have a lot of “cold” data (rarely accessed) relative to warm data. As the amount of warm data increases, we recommend increasing the number of nodes instead (at constant storage/node) so that you can serve the warm data more effectively.

Overall, serving 1PB of data through a CockroachDB cluster should be fine. The remaining question is what cluster size you’d need. For this I recommend you contact one of the sales reps.



Thanks, that was a very useful explanation. I missed that the replication factor and number of nodes were independent.

GKE can support 1000 nodes per cluster which means that with a limit of 64 TB/node and default 3 way replication, the useful storage is 1000*64/3 ~= 21,333 TB.

1 Like

(Raphael 'kena' Poss) #6

We’re not going to recommend you run CockroachDB with 1000 nodes at this point. To achieve 1PB of useful storage we’ll recommend to bump storage/node and vcpus/node instead.

But I’m glad the situation is now clarified. Will be curious to know what you end up choosing in the end.