Imagine the administrator of a busy hospital office the morning after the records server fails. Nobody is hurt, nobody’s care is affected by this story, but the front desk cannot see tomorrow’s appointment list and the billing team is staring at a blank screen. Three colleagues each offer a different rescue. One says “just roll back the snapshot”. One says “restore last night’s backup”. One says “rebuild the machine from the written recipe and reload the data”. All three are reasonable, and all three are different. This article compares them as a SysOps question: the everyday discipline of keeping servers healthy, seen here purely as office-systems administration with no clinical content.
The comparison leans on the Ops Atlas reference for SysOps, linked below, which says plainly that a snapshot is not a backup. Let’s find out why, and when each route is the right one.
Route one: the snapshot
A snapshot freezes a disk exactly as it stood at one instant. It is cheap to create, quick to roll back and wonderfully convenient before a risky change. If an update breaks the web application, rolling back to the picture taken ten minutes earlier can save an afternoon.
The weakness is where the picture lives. A snapshot normally sits on the same storage system as the data it describes. If the storage fails, is deleted by mistake or is hit by ransomware that reaches the same volume, the snapshot goes with it. The Atlas lists “snapshot is not a backup” among its anti-patterns for this reason. Use snapshots as a short-term safety net, never as the only copy.
Strengths and limits
- Speed: minutes to roll back, measured in seconds on many platforms.
- Scope: the whole disk, including mistakes made since the last good point.
- Risk: shares fate with the original storage, and rolling back also discards good data written after the picture.
Route two: the backup
A backup is a separate copy, kept elsewhere, that has been proven by a restore. The proof matters more than the copy. The Atlas stresses that a restore sample belongs in the first ninety days of any improvement plan, and that records of restores are what lift a team onto the top rung of its maturity ladder.
Backups are slower. Restoring a large volume can take hours, and data written since the last run is gone unless the system also keeps logs of changes. On the other hand a backup survives the loss of the machine, the storage and sometimes the whole site. If the host is a virtual machine on rented infrastructure, the Atlas adds a reminder: the guest operating system and its data are the customer’s responsibility, so check what your provider actually backs up.
Strengths and limits
- Durability: independent of the failure that triggered the need for it.
- Granularity: a single file or table can be restored to a scratch location without touching the live system.
- Risk: a backup nobody has tested may be incomplete, corrupt or impossible to read.
Route three: the rebuild
The third colleague’s route relies on a written recipe, which in operations language is a baseline. If a host can be rebuilt from documented settings and its data restored from a backup, then the machine itself is replaceable. The Atlas names this exact idea as its top maturity level: replaceable, backed by records of drills, console access and restores.
A rebuild is the cleanest cure for a machine you no longer trust, for example after unexplained changes or a suspected compromise. It is also the most demanding, because it needs the baseline to exist and to be current. A configuration tool whose runs are safe to repeat helps here; the Atlas observes that tools whose repeated runs change nothing new allow a baseline to be applied hourly without harm, which keeps the recipe honest.
Strengths and limits
- Trust: you end with a machine whose state you can explain.
- Cost: upfront effort to write and maintain the recipe.
- Risk: a stale recipe produces a machine that differs from the one it replaced.
What each route cannot do
It helps to name the blind spots as well. A snapshot cannot rescue you from the loss of the storage beneath it. A backup cannot tell you what the machine’s settings were, so on its own it leaves you guessing how to reassemble the service around the restored data. A rebuild cannot recreate data that was never copied. Each route covers part of the problem, and the combination covers it all, which is why careful teams write down which route handles which scenario before any scenario arrives.
Which route wins on which morning?
The honest answer is that they answer different questions.
- A change just went wrong and the machine is otherwise healthy. Roll back the snapshot. It is the fastest and least disruptive option.
- Data has been deleted or damaged, and you need yesterday’s version of one item. Restore from backup into a scratch area and copy back only what you need.
- The storage or machine has failed or cannot be trusted. Rebuild from the baseline and restore data from backup.
Notice that the second and third routes both depend on a backup that has been tested. That is why the Atlas gives restore evidence so much weight, and why it asks teams to try one before anything goes wrong.
A short word on measuring readiness
Readiness can be reduced to a few questions you can answer with dates. When did we last restore a sample, and who watched? How long does a rebuild take on the written recipe? How many days of free storage do our volumes have? The Atlas’s worked example shows how to compute that last figure: a 500 GiB volume holding 410 GiB, growing 63 GiB in a week, sits near a ceiling where the apps stop at about 475 GiB once the system’s reserved share is excluded, leaving about 7.2 days of runway. Numbers like these turn a vague worry into a date on a calendar.
Why the story matters to non-technical staff
Front-desk and records teams may never touch a server, but they decide how much downtime is tolerable and how much data loss is acceptable. Those two answers tell the technical team which route to invest in. Ask your IT contact for a one-page summary: what is snapshotted, what is backed up and where, when a restore was last tested and how long a rebuild would take. If the page is empty, you have found your first task. For another planning perspective, an article on preparing well ahead of a scheduled visit makes the general point that preparation beats improvisation, even though its subject is far removed from servers.
If you are drawn to the technical side, a practice plan for operations interviews shows how these topics come up in hiring conversations.
Back to the morning after
In our opening story, the office recovers by combining the routes: the broken update is rolled back from the snapshot, the damaged appointment table is restored from last night’s backup into a scratch area and compared, and a rebuild is scheduled for the weekend. For the complete reference, including scope boundaries and a thirty-sixty-ninety day plan, see the Atlas SysOps entry. The three routes are not rivals. They are three tools, and a calm team knows which one to pick.