Skip to content

Inventory

The inventory is the portal's record of the provider's infrastructure: the sites, the hardware, how it is wired, the platforms built on it and the services the sites run. Every installation starts from scratch: the portal is installed first (see Installation), and then the inventory is entered in it, item by item, through the Infrastructure pages. Nothing is imported from files.

From then on the portal's records are the only source. They are edited only in the portal (a list, the details of the selected item, add and edit in dialogs), and everything the portal does with the infrastructure reads them: the deployment, the configurations of the network functions, ordering and capacity, and the checks.

What it holds

flowchart TB
  site[Site] --> net[Networks<br/>ranges, BGP]
  site --> sub[IP subnets]
  site --> sto[Storage and volumes]
  site --> host[Hosts]
  site --> core[Core services]
  site --> plat[Platforms]
  prof[Server profiles] --> host
  plat --> ds[Datastores]
  plat --> pnet[Platform networks]
  plat --> cl[Clusters]
  cl -. hosts .-> host

The arrows show what belongs to what. Besides that, a datastore lives on a storage volume, a platform network on a site network, an IP subnet is carried by an exchange network, and every item that needs a password or a key refers to a secret.

Item What it is Refers to
Site a data center: its code, a site code (a number, fixed once set, part of the AS numbers of its networks), country, city, time zone, NTP (up to 4), DNS (up to 3) and green-field DNS servers (up to 2, used while deploying from scratch), IPMI settings, the management hosts and ports, jump hosts, log collectors secrets
Server profile a hardware model: chassis, CPU, RAM, the boot device and its volumes, the network cards and which ports make the nic-data and nic-storage bonds, GPUs a reference host
Host a server: its name, the address for ssh, the IPMI address, type and port, state, its server profile, and the hardware facts read from the IPMI and the OS a site, a server profile, the IPMI user (a secret)
Storage and volumes NFS, iSCSI or local storage of the site and the volumes on it a site
Network a network of the site: type (the core networks, an exchange network ix, the evpn underlay, storage, a public overlay), interface, VLAN, MTU, subnet, gateway, ranges and BGP (local AS, the IP subnets it carries, the peer groups) a site, the BGP secret of its exchange
IP subnet a public or corporate address range of the site that projects get their addresses from: the subnet, the largest block a project may get, how it is routed the exchange network that carries it
Platform an OpenNebula cloud of the site: kind (mgmt, shared, ext-managed), state, front end (single, local, primary or secondary), public endpoints of a shared platform, its service images a site, the OpenNebula and super admin secrets
Datastore a datastore of a platform: VMs, images, files or backups, the volume it lives on, the share kept free a platform, a volume
Cluster a cluster of a platform: kind (mgmt, shared, dedicated, platform, edge, ext-managed), state, its hosts, datastores and networks, vCores per physical core, spare hosts, reserved CPU and RAM a platform, hosts
Core service what the site itself runs (route reflectors, the site's gateways, name servers, logs, backup, …), on hosts that are there or on VMs made for it a site, platforms, an image, a template, secrets
Secret users (passwords, ssh keys), the BGP passwords of an exchange, the OpenNebula credentials of a front end, users' public keys; kept encrypted —

The fields of each item: Data model. The pages and their buttons: Portal reference, Infrastructure.

Before you start

Before the inventory is entered, the portal itself must be running:

  • a virtual machine with the operating system the installation script requires, and a user who can use sudo without a password;
  • outbound HTTPS from it to the licensing portal (the releases come from there as bundles), and to GitHub once to download the installation script;
  • a licence key;
  • a DNS name that resolves to that machine — needed for your users, and for the certificate authority if you want a real certificate rather than the self-signed one the installation always writes;
  • the first super admin, created from the one-time link the installation script prints;
  • mail configured in Settings, or nothing the portal sends will arrive.

Nothing of the provider's platform has to exist yet. The portal is installed first and the inventory is entered into it afterwards. See Installation.

Claugine builds the platform for you, from green field. The provider prepares the hardware and the landscape — storage presenting its resources, a network ready to carry traffic, delegated subnets, and hosts reachable over IPMI. Claugine installs the rest — the hosts' operating systems, OpenNebula, the network functions and the core services — keeps the record of it consistent, and deploys everything a project needs.

Entering it

From an empty portal to a running platform:

1. The site. Everything belongs to one: hosts, platforms, networks, storage.

2. The hosts. Each is added as planned, and the portal reads the machine over ssh and its management controller to record what it actually has, matching it to a server profile.

3. The platform. Added as planned. While it is planned its whole definition can be changed and it can be deleted outright — so this is the moment to get its front end right.

4. Claugine installs OpenNebula on it, with the hosts' operating system.

5. Its clusters, networks and datastores, entered against that platform. A host that serves only planned clusters is itself still planned, so the inventory stays honest about what is in service.

6. Mark the platform deployed. From here its front-end definition is fixed — name, mode and nodes — and it can no longer be deleted, only retired. A later change to one of its clusters, networks or datastores leaves it deployed.

7. Give the platform its endpoint and credentials — the address of its OpenNebula API and the account the portal uses. Until these are set the portal cannot reach the platform: its health stays unknown and the next step has no effect.

8. Set its mode to execution. Until this is done the portal reads the platform and writes nothing to it. A platform can be complete, deployed, reachable and still provision nothing — this step is the one people miss.

9. The storage, its systems and their volumes.

10. The core services. Added by type, or the site's standard set created in one action with Add the standard layout; the rest of each definition is calculated. Check the definition, then deploy. These need a platform already in execution mode, and run from the active portal node. Add the standard layout never changes services that already exist.

11. Projects can now be ordered, and their gateways deployed onto the platform.

Steps 1, 2, 3, 5 and 9 are a working order rather than a rule — the portal will take them in another sequence, but each thing has to exist before something else can point at it. Steps 4, 6, 8, 10 and 11 are forced: the rules of the platform's state and mode make any other order impossible. Step 7 is in between: the endpoint and credentials can be given any time after the platform exists, but step 8 does nothing until they are there.

A record can only refer to records that already exist (every reference is chosen from a list of them), so what a record points at is entered first, wherever it falls in the order above:

  • Secrets the other items use, beyond a platform's own credentials (step 7): the platforms' super admins, the IPMI users, the jump hosts, the portal's own service user, the BGP passwords of each exchange and EVPN network.
  • The site's networks, with their ranges and BGP, before a platform network on one of them; then the IP subnets, each carried by one exchange network of its type.
  • A storage volume before a datastore that lives on it.
  • Server profiles are made by the hardware read in step 2 rather than entered. When a host is added, its profile is detected from the hardware read through its IPMI (Redfish) and its OS over ssh, and a new profile is made when none matches. Check compares every host of a profile with it; a profile whose hosts are checked and the same is verified.

States

A platform's state is set by an operator on the platform's form, not computed by the portal:

State What it means
planned planned (not installed)
deployed installed and in service
retired out of service

A platform starts as planned. While it is planned its whole definition can be edited, and the platform can be deleted outright — nothing has been built from it yet. Everything belonging to it is planned too: its clusters, and any host that serves only planned clusters.

When the platform has been installed, an operator sets its state to deployed. From that moment the front-end definition is fixed: the name, the mode and the nodes can no longer be changed, only the state. A deployed platform cannot be deleted — it is retired instead, so the record of what existed survives.

Changing a deployed platform's clusters, datastores or networks in Infrastructure leaves it deployed.

  • An installed host or cluster is active. Removing one marks it retired until it is taken down; removing a planned one deletes it.
  • Only active hosts give capacity, and only active shared clusters of an active shared platform can be ordered.

Read-only or execution: the platform's mode

Separate from the state, a platform has a mode that says whether the portal may write to the OpenNebula installation at all:

  • Read-only — the portal reads the platform and writes nothing to it, except into its own test projects.
  • Execution — the portal provisions projects on the platform.

A platform is read-only until someone changes it. Operations that build anything — deploying a core service, creating a project's gateway — are refused on a read-only platform, and only the active portal node may perform them at all. So a platform that is fully entered and marked deployed still provisions nothing until its mode is changed: that is a deliberate guard.

Rules the portal keeps

A step that cannot be taken is refused with the reason, and nothing is changed. Deleting an installed platform, for example, is refused with "Only a planned platform can be deleted; mark an installed one retired instead."

  • A record cannot be deleted while another refers to it: a volume under a datastore, a network under a platform network, a host in a cluster, an IP subnet in use. A secret in use is replaced, not deleted; one that nothing uses can be archived.
  • A host belongs to one site, and its clusters to one platform. A secondary front end has a primary.
  • Shared front ends are separate: no exchange network and no host may be used by two of them.
  • Objects made in OpenNebula carry the site in their name (payload1-vms01_dc1), so federated zones never clash.
  • An exchange network carries IP subnets of its own type only, and an IP subnet is carried by one exchange network only. A subnet cannot be edited after it is added.
  • BGP peers are inside their network: route reflectors by address, igw and egw by address or subnet.
  • Sizes are in GiB; only the boot disks of server profiles are in GB, as vendors give them.

What reads it

  • Deployment: the platforms, the core services and every project's network functions are made from the inventory. Their configurations come from the templates filled with its addresses, names, AS numbers, peer ranges and users, and from Secrets.
  • Ordering: the sites and clusters a project can order, the GPU bundles still free on a cluster, and the prefix lengths offered from the site's IP subnets.
  • Capacity, in the details of a cluster: its hosts and their server profile against the projects' quotas (vCPU as an overcommit ratio, RAM, GPUs never overcommitted).
  • Checks: the platforms' health, reading the core services and their hosts and VMs, the IPMI settings of the hosts, and the server profile check.

Who edits it: resource managers and support engineers; some actions (showing secrets, redeploying a core service) are the super admin's. See Roles.