Kubernetes Storage Carries the Data and Gets the Least Coverage

Search any cloud-native storage survey and etcd sits near the top. That result is not a measurement of adoption. It is a measurement of installation. etcd ships inside every Kubernetes cluster as the default datastore. It arrives whether an operator wants it or not.

Longhorn is the project actually holding persistent data, and it gets a fraction of the coverage. That gap is the story.

Storage is invisible until it fails, and then it is the only thing anyone talks about. That makes it a hard subject to cover on a schedule. Control plane outages make news. A slow volume that quietly corrupts a backup does not.

The gap matters for buyers. Teams pick storage from blog posts and rankings, then live with the choice for years. Weak reporting at the top turns into expensive mistakes at the bottom.

Survey Rankings Reward What Ships by Default

A default beats a choice every time in a survey. etcd requires no decision, so every cluster counts as a user. That inflates its number toward the whole install base. Projects people have to install on purpose will always look smaller.

Read any adoption chart and the shape repeats. A flat line at the top for anything bundled, a long tail for everything installed by hand.

The ranking also hides a split in the work. etcd keeps control plane state. Persistent volume drivers keep application data. They solve different problems, and only one of them takes your database down when it fails.

etcd sits at CNCF Graduated, and it earned that. The point is not that etcd is weak. The point is that it answers a different question than a volume driver does.

Four Projects Cover Four Different Jobs

Longhorn is a CNCF Incubating project, accepted in 2019 and promoted in 2021. It gives Kubernetes distributed block storage with replication and snapshots, and it runs well on modest clusters.

Rook is CNCF Graduated, reaching that level in 2020. It orchestrates Ceph inside Kubernetes rather than replacing it. You get Ceph’s depth and Ceph’s operational weight, which makes Rook the most capable option here and the hardest to run.

OpenEBS sits at the CNCF Sandbox level. It leans on per-workload engines instead of one cluster-wide system, which keeps the blast radius small. That design appeals to teams that want isolation over throughput.

TopoLVM is not a CNCF project. It is a standalone CSI driver that carves local disks into LVM volumes and exposes them as persistent volumes. It is fast because the storage is local, and it is fragile for the same reason.

Maturity level describes governance, not quality. Sandbox projects can be stable and widely used. Graduated projects can still be hard to operate. Read the label as a signal of process, not a grade on the code.

The four also split on where the data sits. Longhorn replicates blocks across nodes. Rook and Ceph pool disks into a shared cluster. OpenEBS lets each workload keep its own engine. TopoLVM keeps everything on one node.

None of the four replaces the others. A cluster can run Longhorn for applications and TopoLVM for a database on the same day.

The Quiet Projects Carry the Production Data

Choose by job, not by ranking. Control plane state goes to etcd. Application data goes to a volume driver.

For most mid-size clusters, Longhorn wins on effort. Rook and Ceph win when you need serious storage and have the staff for it. OpenEBS suits teams that prize isolation. TopoLVM suits databases that want local speed and can tolerate node loss.

That last difference decides failure behavior. Local storage is fast and simple, and it ties a volume to one machine. A node loss then means a restore, not a failover.

Staffing is the other filter. Longhorn and OpenEBS a small team can run. Rook and Ceph want an engineer who understands placement groups and rebalancing. TopoLVM needs almost no operator, and it needs perfect node hygiene.

Start with the workload. Ask whether a node can die without the data going with it. That one question narrows the list faster than any ranking.

Coverage shapes which projects survive a budget cycle. A well-covered default keeps winning renewals even when it is the wrong tool. The quiet projects keep carrying the data anyway.

Related reading. The Most-Used Storage on Kubernetes Is the One You Get for Free. Your Kubernetes Bill Is an Architecture Problem, Not a FinOps Problem. Platform Lock-In Is a Team Decision, Not a Vendor One. The 2026 State of Enterprise Infrastructure.