The five infrastructure families
What belongs in Infrastructure and what stays in the server list, where the rows come from, and what the five screens share.
Infrastructure holds five families, one screen each: virtualization, storage systems, cloud accounts, container clusters, and backup & protection.
| Family | What a row is | What is not a row |
|---|---|---|
| Virtualization | A cluster, and the management plane that runs it | A single host — that is a server |
| Storage systems | The system | A volume, a LUN or a capacity pool |
| Cloud accounts | An account, subscription or project, and the link that attaches it | A cloud instance — that is a server |
| Container clusters | The cluster and its node pools | A pod |
| Backup & protection | The protection service and the coverage it provides | The backup appliance — that is a storage system whose role is Backup |
What belongs here and what stays in the server list
The rule is not "does it have an operating system". An object is a server when the server inventory is already the authority for it: general-purpose compute with a DNS name, patched on the OS vendor's cycle, whose tier produces a meaningful patch cadence and next-due date.
A virtual machine is a server. A container worker node is a server. A cloud instance is a server. A firewall, a switch, a storage controller and a database instance all have an operating system, are all patched, and none of them is a server here.
Where the rows come from
Every row has two halves:
- The discovered half — what an inventory the product already reads knows about the object. Virtualization clusters are derived from the cluster field in the server inventory, and storage systems are detected automatically from the equipment register when the model names a system family the vendor publishes. The other three families are discovered from nowhere.
- The recorded half — what nothing discovers: the support end date, the owning team, the management plane, the HA policy. Those are entered on the record card.
A discovered value is never overwritten by a recorded one. A cluster's platform is read from the hosts actually in it; a typed value beside it would be a second authority for one fact. The card fills in only what no inventory answers.
The Status column — five values, not four
| Value | What it means |
|---|---|
| Fault | The source reports a fault |
| Warning | The source reports a degraded or partial state |
| In maintenance | Deliberately out of service — suspended, disabled, in maintenance mode |
| OK | The source reports everything is well |
| Unknown | The source said nothing, or said something the product does not recognize |
That is also the sort order: worst first, Unknown last. "We cannot say" is not "this is fine", but it is not a finding either, and putting it at the top would bury the real ones.
"In maintenance" and "Unknown" are two different facts. A host deliberately taken out for maintenance has not gone missing, and telling an on-call engineer that its state is unknown wastes the first ten minutes of an incident.
A value the product does not recognize falls to Unknown and the original word stays visible beside it. A source that starts reporting a sixth state must show its own word, not a confident and wrong glyph.
"Support ends" is a date, not a verdict
The column shows the date the source gave and turns red once it has passed. It does not assert that the product is out of support: a red badge on a version still receiving security errata is a false compliance finding the product itself would have produced.
What the five screens share
Full-text search, per-column filters, the quick-filter rail, saved views, column choice and order, and export to CSV and Excel — the same toolkit every table in the product has. The export carries every field the source ships and every custom column that has been added, not only the visible ones.
Next to the refresh button sits a data-freshness chip. The five families are fed by a sync rather than by a live view, so "when was this read" is a question that comes before "refresh it".
Clicking a row opens the record card, and there is no side drawer. A row with no name does not open at all, and the row says why rather than looking inert for no visible reason.
Permissions
| Family | View | Edit | Export |
|---|---|---|---|
| Virtualization | virt.view | virt.edit | virt.export |
| Storage systems | storage.view | storage.edit | storage.export |
| Cloud accounts | cloud.view | cloud.edit | cloud.export |
| Container clusters | containers.view | containers.edit | containers.export |
| Backup & protection | backup.view | backup.edit | backup.export |
One key per family rather than one key for all of infrastructure — an organization that bought storage and did not buy containers has to be expressible. The five view keys and the five export keys are an operator's default; no edit key is.
Editing here means editing the inventory inside the product. It changes nothing in the management plane, in the storage system, or at the cloud provider.
On the connection map
Infrastructure records are nodes on the connection map, and the edges are named: Stores on, Replicates to, Member of, Runs on, Backs up, Terminates on.
A storage system that is also in the equipment register carries the same record key in both registers, so the map folds it into one node and the two cards link to each other. Nobody should meet two cards for one machine with nothing saying it is the same machine.
The product reads from inventory sources and never writes back — see Architecture.
Updated
This page is the file content/docs/en/v1/infra/overview.mdx