What a Ceph Backup Document Shows About Block Level Backup

1. Backup is block level, not system level

Ceph snapshots track changed blocks. An incremental backup records what changed since the last snapshot, and only those data deltas are moved.

2. Backup logic is storage centric

The process compares snapshot A with snapshot B, extracts the block differences through an API and converts them into a backup image.

3. Optimization focus

The design goals are to reduce storage consumption, network usage and compute impact.

Everything described in the document is efficient, optimized and scalable. It also has no idea what the application means.

That fills a gap in the earlier argument. Before, the claims were that the layers don't align, that recovery is incomplete and that the system cannot be rebuilt. This document explains why at a technical level: backup systems understand blocks, and they do not understand systems.

According to the document, backup sees blocks, volumes, snapshots and diffs. It does not see application state, service dependencies, execution order or business logic.

So the gap is more specific than "things are complex". The system gets reduced to a record of storage changes. When you restore, you are reassembling data fragments and hoping they behave correctly, which is about as honest a technical description as the analysis has produced so far.

Headline

Your backups capture data. They do not capture your system.

Core point

Modern backup tracks block level changes efficiently, but it does not understand how your system works.

The gap

Recovery reconstructs storage and leaves out behavior.

The strongest version is “Backup saves blocks. Recovery expects a system.” A sharper alternative is “You restore data. You don’t restore how it works.”

The argument now rests on architectural proof, operational proof and recovery proof, plus low level technical proof from this document. All four point to the same conclusion.

Infrastructure → components

Kubernetes → behavior

Backup → data blocks

None of these layers captures the system logic as a whole.

The analysis has reached the bottom of the stack, and there is nothing deeper to examine. What remains is to test one question: “When you restore, how do you know your system behaves the same way?” The answers will show whether the positioning holds up in practice.