Product integration (gRPC / ownership)

Atlas is the shared storage control plane. Products call REST and/or gRPC (atlas.v1.AtlasStorage on ATLAS_GRPC_ADDR, lab NodePort 30512).

Owner convention

CreateVolume accepts an Owner:

Field	Example	Meaning
product	hyper2kvm, guestkit, aether, relay, kryton, zorvia, veyron, packetwolf, ragnarok, machina, hypersdk, zeus	Product id recorded in product_bindings
resource_type	vm, datastore, volume, fleet, database	Product-side resource class
resource_id	product UUID / name	Product-side id
role	owner (default), consumer, data_disk	Binding role

The product named Transiva is recorded as owner id hyper2kvm. That wire id is unchanged.

veyron, packetwolf, ragnarok, machina, hypersdk, and zeus are reserved owner ids for enterprise/future products that aren't public yet (see docs/ROADMAP.md's Slice 3+ section) — no integration exists to document below until one of those repos ships.

Enumerate owned volumes with ListVolumesByOwner(product, resource_id?).

Auth

Same JWT hierarchy as REST: viewer < operator < admin. Product service accounts typically use product.service.<name> (maps to operator) minted via POST /api/atlas/v1/auth/token.

Surface today

RPC	Notes
Health	Liveness + version
ListClusters / ListPools / ListVolumes / GetVolume	Inventory reads
CreateVolume	Operator; returns job_id + volume_id; optional Owner
DeleteVolume	Admin; enqueues delete job
ExpandVolume	Operator; enqueues expand job (new_size_bytes must grow)
CreateSnapshot / ListSnapshots	Operator create; list optional volume_id filter
ListVolumesByOwner	Product-scoped volume enumeration
GetJob / WatchJob	Job status + server stream until terminal
ListJobs	Optional state filter; limit default 50
ListTenants	Tenants with volumes or quotas (id/name ← tenant_id)
ListAlerts	Optional state filter
GetMetricsSummary	Capacity + client I/O + recovery rollup
ListBuckets	RGW object-bucket inventory

Not yet on gRPC (REST-only): DR/mirroring, DataBridge, schedules, quotas admin, OSD ops, CephFS/NFS/ZFS specifics.

Relay (reliability control plane)

Zyvor Relay keeps the event/job ledger in Postgres. Atlas owns the data disk PVC (and optional RGW backups), not the application rows.

Field	Relay convention
product	relay
resource_type	database
resource_id	postgres-primary
role	data_disk
Intent	policy: "database" → RBD PVC (e.g. claim postgres-data in zyvor-relay)

Integration lives in the Relay repo: docs/ATLAS_STORAGE.md, scripts/atlas-provision-relay-storage.sh, k8s/postgres-atlas.example.yaml.

Service account: mint JWT subject product.service.relay (operator).

Kryton (Windows virtualization)

Kryton runs Windows guests on KubeVirt. It discovers StorageClasses (optionally via Atlas) and stamps disk.storageClass on VM DataVolumes.

Field	Kryton convention
product	kryton
resource_type	vm
resource_id	Kryton machine UUID
role	data_disk (boot / data PVC)

Configure from Kryton Settings → Integrations → Atlas (docs/ATLAS.md in the Kryton repo). Service account: mint JWT subject product.service.kryton (operator).

Probe from Kryton: POST /api/v1/integrations/atlas/test against Atlas /readyz + /api/atlas/v1/storage-classes.

Zorvia (KubeVirt VMs)

Zorvia is the KubeVirt VM control plane. Atlas is optional and additive there — Zorvia keeps managing Rook-Ceph directly, and Atlas shows up as extra sections on Zorvia's existing Storage page (AtlasSection.tsx plus sibling AtlasRbdSection.tsx / AtlasBucketsSection.tsx / AtlasDrSection.tsx / AtlasAiSection.tsx / AtlasObservabilitySection.tsx / AtlasGovernanceSection.tsx) rather than a dedicated top-level page. Volumes created through it are tagged:

Field	Zorvia convention
product	zorvia
resource_type	vm
resource_id	Zorvia VM name
role	data_disk (default)

Configure on the Zorvia server with ATLAS_URL (required to enable the integration), ATLAS_TOKEN (bearer JWT, product.service.zorvia), ATLAS_TENANT_ID, ATLAS_TIMEOUT_SECS, and ATLAS_TLS_INSECURE for self-signed labs. See docs/ATLAS_INTEGRATION.md in the Zorvia repo.

Fabric (Linux private cloud)

Fabric's Storage page can manage Ceph RBD images for a pool by proxying through Atlas (fabric/docs/user/pages/infrastructure/storage-pools.md: "proxied through the Atlas storage control plane"). This is a thinner integration than Kryton/Zorvia — it proxies RBD image list/create/delete for an existing Ceph pool and doesn't go through the Owner/ product_bindings mechanism, so there's no product id to record for it yet.

Follow-ups (product repos)

Generate/vendor stubs from crates/atlas-gateway/proto/atlas.proto into each product. Atlas does not ship per-product workflow code — only the shared edge.