Configuration¶
Configuration is how intent is expressed in JANUS. It is authored as declarative objects, versioned in Git, and compiled by the Controller into the desired state that each Gateway applies. Nothing is configured imperatively: an object is the whole statement of intent for the thing it describes.
This page is the complete reference for every configuration object kind. It assumes you have never read the JANUS source code; every field, type, required flag and validation rule below is the behavior of the shipping configuration loader and validator.
The configuration model¶
Every object, whatever its kind, has the same envelope:
apiVersion: janusaccess.net/v1alpha1
kind: <Kind>
metadata:
name: <lowercase-kebab-identifier>
uuid: <optional durable identity>
displayName: <optional presentation label>
spec:
...
| Field | Type | Required | Meaning |
|---|---|---|---|
apiVersion |
string | yes | Declarative schema version. Canonical value janusaccess.net/v1alpha1. |
kind |
string | yes | The object kind. Matched case-insensitively. |
metadata.name |
string | yes | Human-readable, scoped identifier and the reference key. |
metadata.uuid |
string | no | Durable, immutable machine identity. |
metadata.displayName |
string | no | Presentation-only label. |
metadata.deployment |
string | — | Never written in YAML. Derived from the object's file path. |
spec |
object | yes | Kind-specific body. |
apiVersion¶
The canonical declarative version is:
It is the product namespace, and every newly authored object — through the Management API, the WebUI or client enrolment — uses it.
The value used by older repositories is accepted as a supported, deprecated alias:
What the deprecation means for you as an operator:
- objects that use the alias still load, validate, compile and reconcile exactly like canonical objects;
- they are normalized in memory to the canonical value; the file on disk is never rewritten by the loader;
- a deprecation warning is collected for each such object, identifying the object, the version in use, the canonical version and the migration action. A warning never fails a load.
Any other value — an unknown namespace, a future version, or a missing
apiVersion — is a hard error. JANUS never guesses a schema, so an object
that declares a version it does not recognise fails closed rather than being
interpreted as the current one.
kind¶
Object decoding switches on the kind field. Matching is case-insensitive,
so kind: user and kind: User are the same object; unknown kinds fail closed.
The directory name a file lives in is not a reliable signal and is never
used to determine the kind.
metadata¶
metadata.name, metadata.uuid and metadata.displayName are described in
Identity and naming.
metadata.deployment is a derived field that never appears in YAML. It is the
first path segment of the object's path below the repository root, so
northwind-prod/users/alice.yaml is scoped to the deployment northwind-prod. A file with
fewer than three path segments (<name>.yaml, or <plural>/<name>.yaml) belongs
to the platform scope, written as the empty string "".
File layout and encoding¶
Deployment-scoped kinds live under a deployment directory:
<repository-root>/
tenants/<name>.yaml # platform scope
deployments/<name>.yaml # platform scope
gatewaynodes/<name>.yaml # platform scope
<deployment>/
gateways/<name>.yaml
sites/<name>.yaml
siteedges/<name>.yaml
prefixes/<name>.yaml
resources/<name>.yaml
users/<name>.yaml
groups/<name>.yaml
devices/<name>.yaml
accesspolicies/<name>.yaml
routepolicies/<name>.yaml
natpolicies/<name>.yaml
All YAML files must use:
- UTF-8 encoding
- LF line endings
- two-space indentation
- one object per file
Files and directories whose name begins with . or _ are ignored, as are
non-YAML files.
Platform-scoped versus deployment-scoped kinds¶
| Scope | Kinds | File location |
|---|---|---|
| Platform | Tenant, Deployment, GatewayNode | <plural>/<name>.yaml at the repository root |
| Deployment | Gateway, Site, SiteEdge, Prefix, Resource, User, Group, Device, AccessPolicy, RoutePolicy, NATPolicy | <deployment>/<plural>/<name>.yaml |
Platform-scoped kinds describe the multi-tenant control plane — ownership, entitlements, capability and placement. They must not live inside a deployment directory; the validator rejects that. Deployment-scoped kinds carry the deployment's own addressing, identities and policy. See Architecture overview.
Identity and naming¶
Every object carries three identity fields in metadata.
| Field | Purpose | Scope | Mutable | Unique |
|---|---|---|---|---|
metadata.name |
human-readable reference key | (deployment, kind); platform kinds: platform-wide |
renameable (preserves UUID) | within scope |
metadata.uuid |
durable machine identity | global | no (immutable) | globally |
metadata.displayName |
presentation | none | yes | no |
name versus uuid versus displayName¶
nameis lowercase kebab-case, matching^[a-z0-9]([a-z0-9-]*[a-z0-9])?$, with a maximum of 128 characters. It is the key that every declarative reference uses (spec.owner,spec.gateway.primary,spec.destination.resource,spec.site,spec.subject.user, …).uuidis a UUID (canonical8-4-4-4-12hexadecimal form). It is optional in a hand-authored Git object; the Management API assigns a time-ordered UUIDv7 on first write and never changes it afterwards. It is immutable — a rename preserves it, and delete-and-recreate produces a new one — and it is globally unique: two objects sharing a UUID fail validation. It is not a human-facing label and should not be typed by hand.displayNameis free text of at most 256 characters, used only for presentation. It is not a reference key, not unique, and never an authorization identity. When empty it falls back tometadata.name.
Identity-bearing references. A few security-relevant relationships do not
resolve by name. The owning-tenant relationships — Deployment.spec.tenant,
GatewayNode.spec.tenants (dedicated nodes) and MSP.spec.tenants — are
identity-bearing: the target's immutable metadata.uuid is authoritative and
the name is advisory display metadata:
tenant:
uuid: 0191f0aa-7c3e-7b21-9f4a-1a2b3c4d5e6f # authoritative
name: northwind # advisory, may be stale
A name-only reference (the legacy scalar form tenant: northwind) is accepted
only so a pre-existing repository keeps parsing, but it fails validation until
it is pinned to the target's durable identity — this is what the local
janus-controller config migrate-references does. The reason is security: if a
name were the durable link, re-using a name for a different tenant would silently
re-bind the relationship. When a uuid is present the name is never consulted, so
a rename (or a stale name) can never change which object a reference designates.
All other references (spec.owner, spec.gateway.primary, spec.site, …) remain
in-deployment name-based.
Name uniqueness scope¶
Human-readable identity is (deployment, kind, name):
- a deployment-scoped name must be unique among objects of the same kind in
the same deployment — two deployments may each declare a
Usernamedalicewithout collision; - a platform-scoped name (
Tenant,Deployment,GatewayNode) must be unique platform-wide.
References resolve within the referencing object's own deployment only. There is no cross-deployment reference syntax; a reference that does not resolve in-deployment fails closed. See Relationships and ordering.
Identifier rules¶
| Field | Rule |
|---|---|
metadata.name |
^[a-z0-9]([a-z0-9-]*[a-z0-9])?$, max 128; unique per (deployment, kind), platform kinds platform-wide |
metadata.uuid |
UUIDv7, immutable, globally unique; 8-4-4-4-12 hex |
metadata.displayName |
free text, max 256, not a reference key, falls back to name |
| Capability / entitlement key | ^[a-z][a-z0-9-]*(:[a-z0-9][a-z0-9-]*)?$, max 128, e.g. remote-access, connector:hpe-aruba |
| WireGuard public key | base64 of exactly 32 bytes |
| Ed25519 identity key | base64 of exactly 32 bytes |
Gateway.spec.wireguard.endpoint |
host:port: DNS name, IPv4 literal or bracketed IPv6, port 1-65535 |
How configuration is authored¶
Git is the source of truth¶
The declarative repository is authoritative. Every object, however it is created, ends up as a versioned file:
- direct Git — commit the YAML file yourself under the paths above;
- Management API — write through the supported REST endpoints;
- WebUI — edit through the browser, which writes the same objects.
All three paths converge on the same validated objects. A direct Git change is validated with the same fail-closed rules as an API write, against the whole declarative repository, before compile and reconcile. A change that fails validation is never partially applied.
One object per file¶
Each file contains exactly one object and one object only. The desired-state
layout is <deployment>/<plural>/<name>.yaml for deployment-scoped kinds and
<plural>/<name>.yaml for platform-scoped kinds. The plural directory for each
kind is the one shown in the file-layout listing above and is the same directory
the Management API and WebUI use.
Validation is fail-closed¶
Validation is deliberately fail-closed. The presence of a contradiction — an unresolvable reference, an empty required list, a duplicate name, an invalid key — is always an error, never a warning or a silent default. Where a list must carry a meaning, an explicitly empty list is rejected rather than interpreted as "unset". A referencing object created before its referent fails validation; there is no partial resolution.
The Management API, the WebUI and the CLI surface these errors; see Reference: API and Reference: CLI.
Object reference¶
The subsections below are ordered Tenant, Deployment, GatewayNode, Gateway, Site, SiteEdge, Prefix, Resource, Device, User, Group, AccessPolicy, RoutePolicy, NATPolicy. Addresses in the examples use documentation and private-use ranges only.
Tenant¶
A Tenant is the administrative, security, ownership and commercial boundary.
It owns one or more deployments and is deliberately not itself a deployment. It
is a platform-scoped object.
apiVersion: janusaccess.net/v1alpha1
kind: Tenant
metadata:
name: northwind
uuid: 0191f0aa-7c3e-7b21-9f4a-1a2b3c4d5e6f
displayName: Acme Corp
spec:
enabled: true
entitlements:
- connector:hpe-aruba
- remote-access
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Locator. Unique across the platform. |
spec.enabled |
bool | yes | Tenant lifecycle. A disabled tenant owns its deployments for audit, but holds no entitlement in effect. |
spec.entitlements |
[]string | no | Capability keys the tenant is licensed to use. |
Entitlements. An entitlement is a licence to use a capability. It is not a
grant of access and it is never a throughput ceiling. A capability is in effect
only when the tenant is entitled to it and a placed, enabled node is capable
of it and the deployment requires it. A capability key is a lowercase token,
either feature or namespace:name; the vocabulary is open and extensible.
The reserved default tenant. The name default is reserved. It owns any
deployment scope that has no explicit Deployment object, so existing
installations keep working. The implicit default tenant holds no entitlement.
Declaring a Tenant named default materialises the same identity so
entitlements can be granted to it; ownership of its deployments does not change.
Validation
- name must be lowercase kebab-case, within the length limit, and unique across the platform;
- a duplicate entitlement is rejected;
- a malformed capability key is rejected;
- the object must be platform-scoped — placing it in a deployment directory is rejected;
- a duplicate
metadata.uuidanywhere is rejected.
Deployment¶
A Deployment declares the control-plane facts about one deployment scope: the
owning tenant, the required capabilities and where it may execute. It is a
platform-scoped object, and metadata.name must equal the deployment
name used as the first path segment of that deployment's objects.
apiVersion: janusaccess.net/v1alpha1
kind: Deployment
metadata:
name: northwind-prod
spec:
tenant:
uuid: 0191f0aa-7c3e-7b21-9f4a-1a2b3c4d5e6f
name: northwind
capabilities:
- connector:hpe-aruba
placement:
gateways:
- her-01
network:
internalPrefixes:
- 10.100.0.0/16
- 2001:db8:1234::/48
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | The deployment scope name. Must equal the deployment directory. Unique across the platform. |
spec.tenant |
object {uuid, name} |
yes | The owning tenant, as an identity-bearing reference: the immutable uuid is authoritative and name is advisory display metadata. An empty, name-only or unknown-uuid value fails closed. |
spec.capabilities |
[]string | no | Capability keys this deployment requires. |
spec.placement.gateways |
[]string | no | GatewayNode names this deployment may execute on. |
spec.selfService |
object | no | Centrally defined self-service policy. |
spec.network.internalPrefixes |
[]string | no | Extra prefixes classified as INTERNAL. |
Capability, entitlement and placement. A required capability is in effect only when all of these hold: the owning tenant is enabled and entitled to it; at least one GatewayNode the deployment is placed on is enabled and capable of it; and, for a dedicated node, the node admits the owning tenant. Capability alone never grants use, and entitlement alone never grants use.
spec.selfService. Declares the bounds of the self-service device
management portal. When it is absent, or present with enabled: false,
self-service is unavailable and the portal fails closed. A user manages only
their own devices, inside the bounds declared here — self-service never becomes
an authorization-policy editor.
| Field | Type | Required | Meaning |
|---|---|---|---|
enabled |
bool | no | Turns self-service device management on. |
gateway |
string | yes when enabled | The declared Gateway self-service devices attach to. |
clientIPv4Pool / clientIPv6Pool |
string | at least one when enabled | The client address pool(s). Valid CIDR, a network address with no host bits, and at least one usable host. |
maxDevicesPerUser |
int | no | Control-plane scale bound. Default 10; 0 means the built-in default; negative is rejected. |
spec.network.internalPrefixes. Additional deployment-scoped prefixes
classified as INTERNAL for Internet egress. These are classification overrides
only — never routes and never authorizations. A default route is rejected here,
because it would classify the whole address space as internal. See
Internet egress.
Validation
- empty
spec.tenantis rejected; - a name-only
spec.tenantis rejected — the owning tenant must be referenced by its immutableuuid(author the{uuid, name}mapping, or pin an existing name-only reference with the local reference migration). A name is advisory and must never be the durable link, so a re-used name can never silently re-bind the relationship; - a
spec.tenant.uuidthat does not resolve to a declared, durably-identified tenant is rejected (the reserveddefaultresolves implicitly); - a second
Deploymentobject for one deployment name is rejected — ownership would be ambiguous; - the object must be platform-scoped;
- a required capability that the tenant is not entitled to (or whose tenant is disabled) is rejected;
- a required capability for which no enabled, capable, placed node exists is rejected;
- an unknown, disabled or dedicated-ineligible node in
spec.placement.gatewaysis rejected; - a duplicate node or capability entry is rejected;
- an internal prefix that is a default route is rejected; a duplicate or malformed internal prefix is rejected;
- on an enabled self-service policy: a missing or unknown gateway, no client
pool, a pool with host bits set or no usable host, or a negative
maxDevicesPerUseris rejected.
GatewayNode¶
A GatewayNode is the infrastructure / execution node abstraction: what a node
can technically execute and how it may be shared. It is never owned by a
tenant and is distinct from the deployment-scoped Gateway. It is a
platform-scoped object.
apiVersion: janusaccess.net/v1alpha1
kind: GatewayNode
metadata:
name: her-01
spec:
enabled: true
country: DE
region: NRW
city: Herne
sharing: shared
capabilities:
- connector:hpe-aruba
- remote-access
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Node name. Unique across the platform; the name deployments place against. |
spec.enabled |
bool | yes | Node lifecycle. A disabled node satisfies no capability and cannot host a deployment. |
spec.country / spec.region / spec.city |
string | no | Operator-facing location metadata. |
spec.sharing |
string | no | shared (default) or dedicated. |
spec.tenants |
[]object {uuid, name} |
for dedicated only |
The tenants a dedicated node is reserved to, as identity-bearing references (immutable uuid authoritative; name advisory). |
spec.capabilities |
[]string | no | Capability keys the node can technically execute. |
Sharing. shared may host multiple eligible tenants and deployments and must
not carry a tenant list. dedicated restricts placement to the named tenants
and must name at least one; a deployment of any other tenant placed on it fails
closed. Shared versus dedicated is a placement/policy distinction on the node
only — both use the same data plane.
Validation
- an unknown
spec.sharingvalue is rejected; sharedwith a tenant list is rejected;dedicatedwith no tenant is rejected;- a
spec.tenantsentry that is unknown or name-only is rejected (each entry must carry the tenant's immutableuuid); a duplicate entry is rejected; - a malformed or duplicate capability key is rejected;
- a duplicate node name is rejected;
- the object must be platform-scoped.
Gateway¶
A Gateway declares one Access Concentrator gateway host / WireGuard endpoint
for one (deployment, node). It is a deployment-scoped kind, living at
<deployment>/gateways/<name>.yaml. spec.gatewayNode names the platform
GatewayNode this object runs on — the authoritative link between the
deployment-scoped Gateway and the infrastructure node.
apiVersion: janusaccess.net/v1alpha1
kind: Gateway
metadata:
name: fra-01-gw1
spec:
enabled: true
gatewayNode: her-01
location:
country: DE
region: NW
city: Herne
wireguard:
listenPort: 51820
endpoint: gw.example.net:51820
capabilities:
remoteAccess: true
siteToSite: true
bgp: false
publicPrefixes: true
nat: true
network:
allowedPublicIPv4:
- 203.0.113.0/24
fabric: {}
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Gateway identifier, deployment-scoped. |
spec.enabled |
bool | yes | Whether the gateway is active. |
spec.gatewayNode |
string | no | The platform GatewayNode this gateway runs on. When set it must name a real node; it links this object to the node's enrolled runtime identity. |
spec.location.country / region / city |
string | no | Location metadata. |
spec.wireguard.listenPort |
int | no | WireGuard UDP listen port (default 51820). |
spec.wireguard.endpoint |
string | no | Public host:port a client dials. A locator, not an identity. |
spec.capabilities.* |
bool | no | Feature flags; all default to false. |
spec.network.allowedPublicIPv4 / allowedPublicIPv6 |
[]string | no | Public address space this gateway may use. |
spec.fabric |
object | no | Declares this (deployment, node) a fabric participant. The block carries no key. |
Capabilities. remoteAccess, siteToSite, bgp, publicPrefixes and nat
are boolean feature flags for this gateway; each defaults to false.
WireGuard keys are automatic. The gateway's WireGuard public key and its
per-participant fabric public key are not authored here. The gateway
generates the matching private keys locally (they never leave the host) and
reports the public keys through its authenticated enrollment; the controller is
the single source of truth for those public identities. There is deliberately no
spec.wireguard.publicKey or spec.fabric.publicKey field: an operator never
types, pastes or copies a JANUS Gateway/Fabric key. (External keys — a Device's
own client key, a Site, a Site Edge's transport key — are still authored,
because JANUS does not own those private keys.)
Fabric participation. Declaring the spec.fabric block makes this
(deployment, node) a fabric participant; the participant's transport key is
reported automatically. A Gateway with no spec.fabric block is not a fabric
participant — it joins no fabric and no other participant installs a peer for it.
In a fabric with more than one participant, every participant must declare
spec.wireguard.endpoint. Fabric participation is required before a Gateway
can be a SiteEdge attachment target.
Validation
spec.gatewayNode, when set, must name a declaredGatewayNode;spec.wireguard.endpoint, when set, must behost:port(DNS name, IPv4 literal or bracketed IPv6), with a port in 1-65535;- in a multi-participant fabric, a participant with no declared transport endpoint host is rejected;
- a name colliding with another
Gatewayof the same deployment is rejected.
Site¶
A Site is a remote Site-to-Site VPN peer. It is a deployment-scoped kind at
<deployment>/sites/<name>.yaml.
apiVersion: janusaccess.net/v1alpha1
kind: Site
metadata:
name: leipzig-warehouse
spec:
enabled: true
gateway:
primary: fra-01-gw1
wireguard:
publicKey: AwoRGB8mLTQ7QklQV15lbHN6gYiPlp2kq7K5wMfO1dw=
tunnel:
ipv4:
local: 10.255.10.1/31
remote: 10.255.10.0/31
routing:
mode: static
advertisedPrefixes:
- 192.0.2.0/24
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Site identifier, deployment-scoped. |
spec.enabled |
bool | yes | Whether the site connection is active. |
spec.gateway.primary |
string | yes | Gateway this site connects to, in the same deployment. |
spec.wireguard.publicKey |
string | yes | The site's WireGuard public key. |
spec.tunnel.ipv4.local / .remote |
string | no | Tunnel IPv4 addresses (CIDR). |
spec.tunnel.ipv6.local / .remote |
string | no | Tunnel IPv6 addresses (CIDR). |
spec.routing.mode |
string | no | static or bgp (default static). |
spec.routing.advertisedPrefixes |
[]string | no | Prefixes the site advertises (CIDR). |
Routing modes. static compiles routes from advertisedPrefixes; bgp
exchanges routes via BGP and requires FRR on the gateway.
Validation
spec.gateway.primarymust resolve to aGatewayof the same deployment;spec.wireguard.publicKeymust be a valid WireGuard public key;- each tunnel address must be valid CIDR, and a tunnel IP must not be reused by another site in the same deployment;
spec.routing.modemust bestaticorbgp;- each advertised prefix must be valid CIDR;
- two enabled sites in a deployment must not advertise the same prefix, and must not advertise overlapping prefixes.
SiteEdge¶
A SiteEdge is one JANUS Site Edge: the customer-side software that terminates
JANUS connectivity on behalf of a Site. It is a deployment-scoped kind at
<deployment>/siteedges/<name>.yaml, and it references an existing Site in the
same deployment. One site may own several Site Edges.
A Site Edge routes; the provider authorizes. See Site Edge.
apiVersion: janusaccess.net/v1alpha1
kind: SiteEdge
metadata:
name: leipzig-edge1
spec:
enabled: true
site: leipzig-warehouse
roles: [ingress, egress]
attachment:
primary: fra-01-gw1
identity:
publicKey: CRYjMD1KV2RxfouYpbK/zNnm8wANGic0QU5baHWCj5w=
wireguard:
publicKey: AwoRGB8mLTQ7QklQV15lbHN6gYiPlp2kq7K5wMfO1dw=
lan:
prefixes:
- 192.0.2.0/24
- 2001:db8:50::/48
ha:
precedence: 10
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Name, scoped to (deployment, kind, name). |
spec.enabled |
bool | yes | Access state, never an activity state. |
spec.site |
string | yes | An existing Site in the same deployment. |
spec.roles |
[]string | yes | Non-empty subset of {ingress, egress}. |
spec.attachment.primary |
string | yes when enabled | The single provider Gateway this edge attaches to. |
spec.identity.publicKey |
string | yes | Pre-registered Ed25519 control-plane public key. |
spec.wireguard.publicKey |
string | no | Edge WireGuard transport public key (public material). |
spec.lan.prefixes |
[]string | no | Customer-local prefixes the edge declares. |
spec.ha.precedence |
int | no | Declarative redundancy preference. |
Roles. spec.roles is a set over the closed enum {ingress, egress}.
bidirectional is not a value: it is expressed as the two roles together.
Direction is enforced, not descriptive — a customer source reaching the Internet
is a customer-to-JANUS NEW connection and therefore needs ingress; an
egress-only edge denies that flow even when the subject holds a valid Internet
policy.
Attachment. spec.attachment.primary names exactly one provider
Gateway of the same deployment, and it is desired state rather than an
observation. The named gateway must exist in the edge's own deployment, be
enabled, and be a fabric participant (declare a spec.fabric block). A
disabled edge may omit the attachment, but a stated one must still resolve.
Identity. The pre-registered Ed25519 public key is the edge's durable control-plane identity. One identity key may back exactly one Site Edge everywhere; a second registration of the same key is rejected.
Declared versus accepted prefixes. Declaring a prefix in
spec.lan.prefixes does not make it reachable. A customer prefix becomes
provider reachability only when it is declared by both the edge's
spec.lan.prefixes and an enabled Resource bound to the edge via
spec.siteEdge:
acceptedStaticPrefixes(edge) =
edge.spec.lan.prefixes INTERSECT prefixes of enabled Resources bound to the edge
A declared prefix with no bound Resource is visible and reportable but never
becomes provider reachability. A default route (0.0.0.0/0, ::/0) is
representable but is never accepted for provider transit. Any prefix length is
representable, including /0, /32 and /128.
Validation
spec.siteis required and must resolve in the same deployment;spec.rolesmust be non-empty, contain no duplicates, and contain onlyingressandegress(notbidirectional);- on an enabled edge,
spec.attachment.primaryis required and must name an enabled, fabric-participantGatewayof the same deployment; spec.identity.publicKeyis required and must be a valid Ed25519 key; a key already pre-registered to another Site Edge is rejected;spec.wireguard.publicKey, when present, must be a valid WireGuard public key;- each LAN prefix must be valid CIDR, with no duplicate; two enabled Site Edges in one deployment must not accept the same prefix;
spec.ha.precedencemust not be negative;- a declared prefix with no bound enabled
Resource, or a declared default route, produces a warning (not accepted) rather than becoming reachability.
Prefix¶
A Prefix is an IP prefix assignable to a site or device, and usable as an
AccessPolicy source network. It is a deployment-scoped kind at
<deployment>/prefixes/<name>.yaml.
apiVersion: janusaccess.net/v1alpha1
kind: Prefix
metadata:
name: leipzig-public-v4
spec:
prefix: 203.0.113.0/24
family: IPv4
public: true
assignedTo: alice
site: leipzig-warehouse
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Prefix identifier, deployment-scoped. |
spec.prefix |
string | yes | CIDR notation. |
spec.family |
string | no | IPv4 or IPv6; auto-detected from the prefix when empty. |
spec.public |
bool | yes | Whether this is a public (Internet-routable) prefix. |
spec.assignedTo |
string | no | Free-text assignment label. |
spec.site |
string | no | Site this prefix is assigned to. |
Validation
spec.prefixmust be valid CIDR;- when
spec.familyis set it must match the prefix's address family (case-insensitive), otherwise the object is rejected; - when
spec.publicistrue, the prefix must fall within the allowed public aggregates configured on the Controller, otherwise it is rejected. Those aggregates are installation configuration, not desired state: the Controller configuration is the upper safety boundary, and no object in Git can widen it. The same aggregates bound a Gateway'sspec.network.allowedPublicIPv4/allowedPublicIPv6and therefore whichsnataddresses a deployment may use; spec.assignedTois a free-text label only.
Resource¶
A Resource is an internal resource — a VM, service or network — that clients or
sites can access. It is a deployment-scoped kind at
<deployment>/resources/<name>.yaml.
apiVersion: janusaccess.net/v1alpha1
kind: Resource
metadata:
name: frankfurt-fileserver
spec:
type: host
addresses:
- 10.20.0.10/32
services:
- name: ssh
protocol: tcp
port: 22
gateway:
primary: fra-01-gw1
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Resource identifier, deployment-scoped. |
spec.type |
string | yes | Resource type. host, network or service. |
spec.prefixes |
[]string | no | Network prefixes (CIDR). |
spec.addresses |
[]string | no | Individual addresses (CIDR). |
spec.services |
[]object | no | Service definitions (see below). |
spec.gateway.primary |
string | no | Gateway that provides reachability to, and originates, this resource. |
spec.siteEdge |
string | no | A SiteEdge through which this resource is originated and reached. |
Reachability ownership. spec.gateway.primary names the provider Gateway
that provides reachability to the resource inside the deployment fabric; the
resource's addresses are then originated by that participant. spec.siteEdge
instead names a SiteEdge: the resource sits behind a customer edge, is reached
through it, and the edge's spec.attachment.primary determines the provider
Gateway path. The two are mutually exclusive — a resource behind a Site
Edge is reached through that edge, so spec.gateway.primary must be empty when
spec.siteEdge is set. A resource with neither field is neither advertised into
the fabric nor bound to an edge.
A resource bound to a SiteEdge contributes to that edge's accepted prefixes:
a customer prefix is accepted for provider transit only when it is declared by
both the edge's spec.lan.prefixes and a bound, enabled Resource.
Service definition (shared with AccessPolicy):
| Field | Type | Required | Meaning |
|---|---|---|---|
name |
string | no | Human-readable service name. |
protocol |
string | yes | tcp, udp or sctp. |
port |
int | yes | Port number, 1-65535. |
Validation
spec.typeis required; when it ishost(case-insensitive), every address must be a/32(IPv4) or/128(IPv6) prefix;- every
spec.prefixesandspec.addressesentry must be valid CIDR; - every service
protocolmust betcp,udporsctp, and everyportmust be 1-65535; spec.gateway.primary, when set, must resolve to aGatewayof the same deployment;spec.siteEdgemust resolve to a deployment-scopedSiteEdge, and is mutually exclusive withspec.gateway.primary;- binding to a disabled Site Edge is a warning: the resource's prefixes are not accepted for provider transit.
Device¶
A Device is a VPN client device — a laptop or phone. It is a
deployment-scoped kind at <deployment>/devices/<name>.yaml.
apiVersion: janusaccess.net/v1alpha1
kind: Device
metadata:
name: alice-laptop
spec:
owner: alice
enabled: true
wireguard:
publicKey: AwoRGB8mLTQ7QklQV15lbHN6gYiPlp2kq7K5wMfO1dw=
ipv4: 10.255.0.10/32
ipv6: 2001:db8:10::10/128
gateway:
primary: fra-01-gw1
eligible:
- fra-01-gw1
- fra-01-gw2
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Device identifier, deployment-scoped. |
spec.owner |
string | yes | Owning User, in the same deployment. |
spec.enabled |
bool | yes | Whether the device is active. |
spec.wireguard.publicKey |
string | yes | The client's own WireGuard public key. |
spec.wireguard.ipv4 / ipv6 |
string | no | Assigned overlay addresses (CIDR). |
spec.gateway.primary |
string | yes* | Preferred gateway, in the same deployment. |
spec.gateway.eligible |
[]string | no | Closed set of gateways this device may attach to. |
* primary may be omitted only when eligible is also omitted; the historical
"any enabled gateway of the deployment" behavior is then preserved.
Client-owned keys. spec.wireguard.publicKey is the client's own public
key. JANUS never generates, receives or stores a client private key. A device is
normally created by the client-enrollment flow — which authors this object — but
may equally be created by an administrator directly through the API or by
committing the YAML. Both paths produce the same declarative object.
Eligibility semantics. eligible is the closed set of gateways this device
is permitted to attach to. It is desired state: an administrator declares it.
primary = the PREFERRED gateway (where the device attaches when reachable)
eligible = the SET the device may attach to; runtime failover selects from it
eligibleabsent → effective eligibility is exactly[primary](single-gateway behavior). Ifprimaryis also absent, the device may attach at any enabled gateway of the deployment — the historical behavior.eligiblepresent, non-empty → that set is the eligibility, andprimary(which must be one of them) is the preferred member.eligiblepresent, empty → validation error. An explicitly empty list is never silently treated as[primary]and never silently widened.
Which gateway a device is currently using is runtime state and never produces a
Git commit; only an administrator changes primary or eligible. A runtime
observation may select only among gateways that are already eligible.
Validation
spec.ownermust resolve to aUserof the same deployment;spec.wireguard.publicKeymust be a valid WireGuard public key;spec.wireguard.ipv4andspec.wireguard.ipv6, when set, must be valid CIDR, and an overlay IP must not be reused by another device in the same deployment;spec.gateway.primary, when set, must resolve to aGatewayof the same deployment;spec.gateway.eligible, when present, must be non-empty, contain no empty or duplicate member, name only declaredGateways of the same deployment, and containgateway.primary;- when
eligibleis set,primaryis required; whenprimaryis set andeligibleis set,primarymust be ineligible.
User¶
A User is a human user identity. It is a deployment-scoped kind at
<deployment>/users/<name>.yaml.
apiVersion: janusaccess.net/v1alpha1
kind: User
metadata:
name: alice
spec:
displayName: Thomas Example
email: alice@example.com
enabled: true
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | User identifier, deployment-scoped. |
spec.displayName |
string | no | Human-readable display name. |
spec.email |
string | no | Email address. |
spec.enabled |
bool | yes | Whether the user account is active. |
Note. spec.displayName is a pre-existing user-profile field. It is distinct
from metadata.displayName — the two are separate fields on User, not aliases.
Validation
- the name must satisfy the identifier rules and be unique within its
(deployment, kind); - an
AccessPolicyreferencing a non-existent user is a validation error.
Group¶
A Group is a set of users, used as an AccessPolicy subject. It is a
deployment-scoped kind at <deployment>/groups/<name>.yaml.
apiVersion: janusaccess.net/v1alpha1
kind: Group
metadata:
name: engineering
spec:
members:
- alice
- alice
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Group identifier, deployment-scoped. |
spec.members |
[]string | no | Names of the users in this group. |
Validation
- the name must satisfy the identifier rules and be unique within its
(deployment, kind); - a group referenced by an
AccessPolicymust exist in the same deployment.
AccessPolicy¶
An AccessPolicy grants a user, group, device or prefix access to a resource or
to the Internet. It is a deployment-scoped kind at
<deployment>/accesspolicies/<name>.yaml.
The default is deny: only explicitly enabled policies grant access.
apiVersion: janusaccess.net/v1alpha1
kind: AccessPolicy
metadata:
name: alice-fileserver-ssh
spec:
subject:
user: alice
destination:
resource: frankfurt-fileserver
services:
- protocol: tcp
port: 22
enabled: true
expires: 2026-09-11T18:00:00+02:00
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Policy identifier, deployment-scoped. |
spec.subject |
object | yes | The one subject this policy applies to. |
spec.destination |
object | yes | The one destination: a resource or the Internet. |
spec.services |
[]object | no | Protocol/port constraints; the ServiceDef shape above. |
spec.enabled |
bool | yes | Whether the policy is active. |
spec.expires |
string | no | Expiry timestamp. |
Subjects. spec.subject sets exactly one of:
| Subject | Field | Meaning |
|---|---|---|
| User | spec.subject.user |
the user's devices |
| Group | spec.subject.group |
the member users' devices |
| Device | spec.subject.device |
exactly that Device — it does not grant the owner's other devices |
| Prefix | spec.subject.prefix |
an existing deployment-scoped Prefix used as an authoritative source network |
Setting none, or more than one, is a validation error. A raw CIDR is never a
subject — a Prefix subject must name an existing Prefix of the same
deployment. A Prefix subject must not be a default route or a SPECIAL range,
and must have a valid ownership/routing relationship: either it declares a
site that exists and actually covers the prefix (advertised or tunnel
prefixes), or the prefix must be JANUS-known space (declared by an enabled
Resource, a Site advertised prefix, or a Site Edge). A Device subject resolves
to the device's canonical tunnel addresses from trusted desired state; a
client-supplied address is never accepted as the meaning of a policy.
Destination. spec.destination is a discriminated pair — exactly one of
resource or internet:
destination.resourcenames aResourcein the same deployment;destination.internet: truemeans the deployment's INTERNET destination class — never0.0.0.0/0or::/0. INTERNAL and SPECIAL destinations are excluded, so an Internet grant can never authorize an internal JANUS or customer destination. See Internet egress.
Setting both, or neither, is a validation error.
Services. The service model applies unchanged. A policy with no services grants nothing, and a policy is not automatically all protocols and ports: DNS is reachable only if the services allow it.
Temporary access. spec.expires states when the grant should be withdrawn;
the Controller reconciles and removes the access without a new Git commit.
Validation
- exactly one of
subject.user,subject.group,subject.device,subject.prefixmust be set, and must reference an existing object of the same deployment; - exactly one of
destination.resourceordestination.internetmust be set; - a
subject.prefixmust be an existing deployment-scopedPrefix, must not be a default route or a SPECIAL range, and must be either site-covered or JANUS-known space; - each service
protocolmust betcp,udporsctp, and eachportmust be 1-65535.
RoutePolicy¶
A RoutePolicy controls which prefixes a site may import and export. It is a
deployment-scoped kind at <deployment>/routepolicies/<name>.yaml.
apiVersion: janusaccess.net/v1alpha1
kind: RoutePolicy
metadata:
name: leipzig-routes
spec:
site: leipzig-warehouse
import:
- 192.0.2.0/24
export:
- prefix: 203.0.113.0/24
- prefix: 2001:db8:5::/48
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Policy identifier, deployment-scoped. |
spec.site |
string | yes | The Site this policy applies to. |
spec.import |
[]string | no | Prefixes the site may import (CIDR). |
spec.export |
[]object | no | Prefixes the site may export. |
spec.export[].prefix |
string | yes (per entry) | A prefix CIDR the site may export. |
Validation
spec.sitemust resolve to an existingSiteof the same deployment;- each
spec.importentry must be a valid CIDR; - each
spec.export[].prefixvalue must be a valid CIDR. APrefixobject name is not accepted here: the validator parses the value as a prefix.
Effect in this release
RoutePolicy is part of the schema and is validated, but it does not
change the compiled data plane in this release: the compiler deliberately
produces no routing peers from it, and an empty import/export set means deny.
Routing is static and is derived from the Site's own declarations —
spec.routing.advertisedPrefixes, its tunnel pair, and the ownership
declared by Resources, Prefixes and Site Edges. Declare those to make a
prefix reachable; a RoutePolicy alone will not.
NATPolicy¶
A NATPolicy controls source NAT behavior between a site and destination
networks. It is a deployment-scoped kind at
<deployment>/natpolicies/<name>.yaml.
NAT is never implicit. NAT occurs only when an administrator explicitly
declares a NATPolicy. Zero NAT policies means zero NAT rules.
apiVersion: janusaccess.net/v1alpha1
kind: NATPolicy
metadata:
name: branch-special
spec:
site: leipzig-warehouse
mode: snat
source:
prefixes:
- 10.50.0.0/16
destination:
prefixes:
- 10.10.20.0/24
snat:
address: 192.0.2.10
| Field | Type | Required | Meaning |
|---|---|---|---|
metadata.name |
string | yes | Policy identifier, deployment-scoped. |
spec.site |
string | yes | The Site this policy belongs to, in the same deployment. |
spec.mode |
string | yes | none, masquerade or snat. |
spec.source.prefixes |
[]string | no | Optional source scope. Empty means any source. |
spec.destination.prefixes |
[]string | one of | Destination prefixes this policy matches (CIDR). |
spec.destination.internet |
bool | one of | Matches the deployment's Internet destinations. |
spec.snat.address |
string | for snat |
The explicit source address to translate to. |
Mode semantics.
| Mode | Behavior |
|---|---|
none |
No source NAT for matching traffic. The original routed source address is preserved. |
masquerade |
The kernel selects the source address from the outgoing interface/route. IPv4 only. |
snat |
Translate the source address to the explicitly configured snat.address. IPv4 only. |
none is an operational rule, not merely the absence of one: it is applied
as an explicit no-NAT escape so a later, less-specific masquerade or SNAT rule
cannot still translate the traffic.
Destination and precedence. NAT selection is based on the destination prefix. Where several policies match a destination, the most-specific destination prefix wins; for equal destination specificity the most specific source prefix wins, then policy name. The compiler resolves this once and emits the rules in evaluation order. List order in Git is never packet-rule precedence.
| Traffic to | Result |
|---|---|
10.20.1.1 |
masquerade |
10.10.30.1 |
none (no NAT) |
10.10.20.5 |
snat to 192.0.2.10 |
Same-prefix conflicts. Two enabled NAT policies in the same deployment and
site that share an identical effective (masked) destination prefix and the
same source scope are ambiguous — whichever were evaluated first would win.
This is rejected at validation time, whether the effects are identical
("duplicate NAT intent") or different ("different NAT effects"). Overlapping but
distinct prefixes (for example 10.0.0.0/8 and 10.10.0.0/16) are valid:
specificity resolves them deterministically.
IPv4 only; no NAT66. none, masquerade and snat are fully supported for
IPv4. masquerade and snat may not target an IPv6 destination; such a policy
fails validation. Public IPv6 stays routed, never translated — no NAT66 is ever
synthesized as a fallback. none remains valid for IPv6 (it is a no-NAT rule).
A single policy must not mix IPv4 and IPv6 destination prefixes, and its source
and destination prefixes must be the same address family.
Internet NAT. With destination.internet, an snat address must be a public
IPv4 address the deployment owns — inside an enabled Gateway's allowed public
address space. An unowned address is rejected. Internet NAT is the
least-specific NAT entry, so it can never pre-empt an explicit, more specific
rule.
NAT is not authorization.
AccessPolicy decides whether traffic is allowed.
NATPolicy decides how already-authorized traffic is translated.
A NAT rule never grants access to a destination the firewall policy denies. If
no AccessPolicy allows the traffic, a matching NATPolicy changes nothing: the
traffic remains denied.
Validation
spec.sitemust resolve to an existingSiteof the same deployment;spec.modemust benone,masqueradeorsnat;spec.destination.prefixesandspec.destination.internetare mutually exclusive, and at least one destination prefix is required;spec.snat.addressis required formode: snat; it must be a single IPv4 address (never a prefix); it is rejected with a non-snatmode;- a malformed, undecodable or empty destination prefix is rejected;
- an IPv6 destination with
masqueradeorsnatis rejected (no NAT66); - mixed IPv4/IPv6 destinations, or a cross-family source/destination pair, is rejected;
- a same-deployment, same-site, same-source, identical-destination conflict is rejected;
- an Internet
snataddress that is not within an enabled Gateway's allowed public address space is rejected.
Relationships and ordering¶
Reference resolution is within-deployment, fail-closed. Creating a referencing object before its referent exists produces a validation error; there is no partial or silent resolution. In practice you author in dependency order:
| Referencing object | Referent that must exist first |
|---|---|
Deployment.spec.tenant |
Tenant, by its immutable uuid (identity-bearing reference; a name-only reference is refused) |
Deployment.spec.placement.gateways |
GatewayNode (enabled; dedicated nodes must admit the tenant) |
Deployment.spec.capabilities |
node capability and tenant entitlement and placement |
Gateway (fabric target) |
must be enabled and declare a spec.fabric block before it can be a SiteEdge attachment |
Device.spec.owner |
User |
Group.spec.members |
User |
AccessPolicy.spec.subject.user |
User |
AccessPolicy.spec.subject.group |
Group |
AccessPolicy.spec.subject.device |
Device |
AccessPolicy.spec.subject.prefix |
Prefix (site-covered or JANUS-known space) |
Device.spec.gateway.primary/eligible, Site.spec.gateway.primary, Resource.spec.gateway.primary |
Gateway |
Resource.spec.siteEdge |
SiteEdge (plus its Site and an enabled fabric Gateway) |
AccessPolicy.spec.destination.resource |
Resource |
RoutePolicy.spec.site, NATPolicy.spec.site, SiteEdge.spec.site |
Site |
RoutePolicy.spec.export[].prefix |
Prefix |
Cross-object references that fail closed. A reference that does not resolve
in the referencing object's own deployment is always an error. There is no
cross-deployment reference syntax: an object in deployment A can never name an
object in deployment B, even if the name exists there. The reserved default
tenant is the one implicit resolution — a deployment scope with no Deployment
object is owned by default.
Concurrent uniqueness and conflict constraints. Besides per-scope name uniqueness, these cross-object constraints are enforced:
metadata.uuidis unique globally;- a destination prefix must be originated by exactly one fabric participant of a deployment;
- two enabled sites must not advertise the same or overlapping prefix;
- two enabled Site Edges must not accept the same prefix;
- two NAT policies with the same site, source scope and identical destination prefix conflict.
A worked example¶
A fictional operator, Northwind Logistics, runs one deployment northwind-prod
owned by tenant northwind. The provider has one node fra-01 running one
gateway fra-01-gw1. A warehouse site in Leipzig is connected both as a
Site-to-Site peer and through a Site Edge, and a Frankfurt file server is an
internal resource. All addressing below is documentation or private-use space.
repository-root/
tenants/northwind.yaml
gatewaynodes/fra-01.yaml
deployments/northwind-prod.yaml
northwind-prod/
gateways/fra-01-gw1.yaml
sites/leipzig-warehouse.yaml
siteedges/leipzig-edge1.yaml
prefixes/leipzig-public-v4.yaml
resources/frankfurt-fileserver.yaml
resources/leipzig-scada.yaml
users/alice.yaml
groups/logistics.yaml
devices/alice-laptop.yaml
accesspolicies/alice-fileserver-ssh.yaml
accesspolicies/logistics-internet.yaml
accesspolicies/alice-scada-https.yaml
routepolicies/leipzig-routes.yaml
natpolicies/leipzig-internet-nat.yaml
Tenant — tenants/northwind.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: Tenant
metadata:
name: northwind
uuid: 0191f0aa-7c3e-7b21-9f4a-1a2b3c4d5e6f
displayName: Northwind Logistics
spec:
enabled: true
entitlements:
- remote-access
- connector:hpe-aruba
GatewayNode — gatewaynodes/fra-01.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: GatewayNode
metadata:
name: fra-01
spec:
enabled: true
country: DE
region: HE
city: Frankfurt
sharing: shared
capabilities:
- remote-access
- connector:hpe-aruba
Deployment — deployments/northwind-prod.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: Deployment
metadata:
name: northwind-prod
spec:
tenant:
uuid: 0191f0aa-7c3e-7b21-9f4a-1a2b3c4d5e6f
name: northwind
capabilities:
- remote-access
placement:
gateways:
- fra-01
network:
internalPrefixes:
- 10.0.0.0/8
Gateway — northwind-prod/gateways/fra-01-gw1.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: Gateway
metadata:
name: fra-01-gw1
spec:
enabled: true
location:
country: DE
region: HE
city: Frankfurt
wireguard:
listenPort: 51820
publicKey: AwoRGB8mLTQ7QklQV15lbHN6gYiPlp2kq7K5wMfO1dw=
endpoint: gw.example.net:51820
capabilities:
remoteAccess: true
siteToSite: true
bgp: false
publicPrefixes: true
nat: true
network:
allowedPublicIPv4:
- 203.0.113.0/24
fabric:
publicKey: BRAbJjE8R1JdaHN+iZSfqrXAy9bh7PcCDRgjLjlET1o=
Site — northwind-prod/sites/leipzig-warehouse.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: Site
metadata:
name: leipzig-warehouse
spec:
enabled: true
gateway:
primary: fra-01-gw1
wireguard:
publicKey: AwoRGB8mLTQ7QklQV15lbHN6gYiPlp2kq7K5wMfO1dw=
tunnel:
ipv4:
local: 10.255.10.1/31
remote: 10.255.10.0/31
routing:
mode: static
advertisedPrefixes:
- 198.51.100.0/24
SiteEdge — northwind-prod/siteedges/leipzig-edge1.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: SiteEdge
metadata:
name: leipzig-edge1
spec:
enabled: true
site: leipzig-warehouse
roles: [ingress, egress]
attachment:
primary: fra-01-gw1
identity:
publicKey: CRYjMD1KV2RxfouYpbK/zNnm8wANGic0QU5baHWCj5w=
wireguard:
publicKey: AwoRGB8mLTQ7QklQV15lbHN6gYiPlp2kq7K5wMfO1dw=
lan:
prefixes:
- 198.51.100.0/24
ha:
precedence: 10
Prefix — northwind-prod/prefixes/leipzig-public-v4.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: Prefix
metadata:
name: leipzig-public-v4
spec:
prefix: 203.0.113.0/24
family: IPv4
public: true
site: leipzig-warehouse
Resource — northwind-prod/resources/frankfurt-fileserver.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: Resource
metadata:
name: frankfurt-fileserver
spec:
type: host
addresses:
- 10.0.5.10/32
services:
- name: ssh
protocol: tcp
port: 22
gateway:
primary: fra-01-gw1
Resource — northwind-prod/resources/leipzig-scada.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: Resource
metadata:
name: leipzig-scada
spec:
type: network
prefixes:
- 198.51.100.0/24
services:
- name: https
protocol: tcp
port: 443
siteEdge: leipzig-edge1
leipzig-scada declares 198.51.100.0/24 and binds to leipzig-edge1, whose
spec.lan.prefixes also declares 198.51.100.0/24 — so that prefix is
accepted for provider transit. It sets no spec.gateway.primary, which is
required: the two fields are mutually exclusive.
User — northwind-prod/users/alice.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: User
metadata:
name: alice
spec:
displayName: Alice Example
email: alice@example.net
enabled: true
Group — northwind-prod/groups/logistics.yaml¶
Device — northwind-prod/devices/alice-laptop.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: Device
metadata:
name: alice-laptop
spec:
owner: alice
enabled: true
wireguard:
publicKey: AwoRGB8mLTQ7QklQV15lbHN6gYiPlp2kq7K5wMfO1dw=
ipv4: 10.255.0.10/32
gateway:
primary: fra-01-gw1
eligible:
- fra-01-gw1
AccessPolicy — northwind-prod/accesspolicies/alice-fileserver-ssh.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: AccessPolicy
metadata:
name: alice-fileserver-ssh
spec:
subject:
user: alice
destination:
resource: frankfurt-fileserver
services:
- protocol: tcp
port: 22
enabled: true
AccessPolicy — northwind-prod/accesspolicies/logistics-internet.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: AccessPolicy
metadata:
name: logistics-internet
spec:
subject:
group: logistics
destination:
internet: true
services:
- protocol: udp
port: 53
- protocol: tcp
port: 443
enabled: true
AccessPolicy — northwind-prod/accesspolicies/alice-scada-https.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: AccessPolicy
metadata:
name: alice-scada-https
spec:
subject:
device: alice-laptop
destination:
resource: leipzig-scada
services:
- protocol: tcp
port: 443
enabled: true
RoutePolicy — northwind-prod/routepolicies/leipzig-routes.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: RoutePolicy
metadata:
name: leipzig-routes
spec:
site: leipzig-warehouse
import:
- 198.51.100.0/24
export:
- prefix: 203.0.113.0/24
The warehouse prefix that is actually reachable is declared on the Site
(spec.routing.advertisedPrefixes) and by the Prefix and SiteEdge objects
that own it; the RoutePolicy records the import/export intent and is validated,
but it does not itself create reachability in this release.
NATPolicy — northwind-prod/natpolicies/leipzig-internet-nat.yaml¶
apiVersion: janusaccess.net/v1alpha1
kind: NATPolicy
metadata:
name: leipzig-internet-nat
spec:
site: leipzig-warehouse
mode: snat
source:
prefixes:
- 198.51.100.0/24
destination:
internet: true
snat:
address: 203.0.113.10
203.0.113.10 lies within fra-01-gw1's allowed public space
(203.0.113.0/24), so the Internet SNAT address is owned by the deployment as
required. The policy translates already-authorized traffic only: the
logistics-internet policy decides whether the flow is allowed.
Related¶
- Operations — observing desired versus applied state
- WebUI — the browser authoring surface
- Troubleshooting
- Reference: API
- Reference: CLI
- Architecture overview
- Internet egress
- Site Edge