Rook.io appears to be a rather popular object storage solution in the k8s ecosystem. I am wondering if anyone is using it successfully (or not) in production. I’d love to hear of the different kinds of use-cases people are using it for.
We will be running Rook with Ceph as from next week in PROD. We did extensive testing in sandbox environments and everything seems to work like a charm. Only faced some minor issues due to our version of the Linux kernels (strongly recommended to use latest 4.15+), but there was a workaround to bypass them.
Out configuration uses existing directories from all the nodes in the K8s cluster (network storage mounted and formatted as xfs).
Next thing we will add is local SSD disks as raw devices in bare metal servers.
Thanks, @emedina. Are you using a combination of block and shared file-system storage modes? If so, have you noticed any performance differences between the two?
I am assuming your storage is on bare-metal servers. Are you using any cloud-based storage with Rook?
Actually our first Rook cluster is using Ceph with “filestore” against directories from existing volumes (exported from storage cabin, so network storage).
Next, once we get the new bare metal servers, we will use local SSD, as already mentioned. So far for our use cases (Elasticsearch 6 and Kafka 1.1) we haven’t noticed any performance issues compared to our current solution where Rook is not present.