Back to Schedule

RWO and RWX Replicated Block Storage with LINBIT SDS (sponsored)

Philipp Reisner

10:55 - 11:10

La Cave

Every stateful workload on Kubernetes eventually runs into the same question: does this need RWO or RWX, and what does that choice cost? RWO maps cleanly onto replicated block storage. DRBD replicates a block device synchronously across nodes in the Linux kernel; LINSTOR places the replicas and hands the volume to the pod through its CSI driver. The pod mounts a local device, reads happen at local-disk speed, and the data survives the node. RWX does not map cleanly onto block storage at all — a filesystem needs a single writer. In Piraeus Datastore and LINBIT SDS we solve it by putting an NFS server container in front of a replicated volume: the block layer stays highly available, and the NFS server becomes a failover unit on top of it. This talk shows both paths, what the NFS layer costs you in latency and in failover time, and how to decide which of your workloads actually needs RWX.