Struct AtlasRegistry 

pub struct AtlasRegistry { /* private fields */ }

In-memory state shared by all gRPC handlers (new + legacy). All ops go through one of the typed async methods below. Arc<AtlasRegistry> is cheap to clone; the interior RwLock serialises mutations.

contracts is loaded once at startup from <robonix_source>/capabilities/** and is read-only for the lifetime of the process — handlers serve QueryContract / ListContracts directly off it without locking.

Implementations

impl AtlasRegistry

pub fn with_contracts(contracts: ContractRegistry) -> Self

Construct a registry with an empty capability/channel state plus a pre-loaded contract registry. Use this from main.rs once the capabilities dir has been resolved.

pub fn contracts(&self) -> &ContractRegistry

impl AtlasRegistry

pub fn now_ms() -> u64

pub async fn register( &self, provider_id: &str, kind: Kind, namespace: &str, capability_md_path: &str, ) -> Result<String, Status>

Register a new capability instance, OR take over an existing provider_id slot whose previous provider is gone. Empty provider_id triggers Atlas-assigned ephemeral id. Returns the resolved id.

Takeover semantics — a re-Register on an existing provider_id is NOT an error. We assume the old process is dead (or about to be), so we drop its endpoints + state and reset last_heartbeat to now. The caller is then expected to redeclare capabilities with its own fresh endpoints. Without this, an orphan provider (heartbeat eviction hasn’t fired yet — 60s default) blocks every subsequent boot of the same package: rbnx waits for a “new” provider to appear in atlas, the python framework only retries register once and then quietly keeps going, and atlas keeps pointing consumers at the dead process’s endpoints. The previous “ALREADY_EXISTS” failure mode caught accidental dual deployments but the cure was worse than the disease — silent boot failures are nearly impossible to diagnose. If two live processes claim the same id, both will heartbeat into the same record and the latest declare wins; that case shows up as scattered/mysterious endpoint flips, which is at least visible.

pub async fn unregister(&self, provider_id: &str) -> bool

Idempotent: returns true if a record was removed, false if the id was unknown. Also drops any channels where this provider was the provider — consumers will get NOT_FOUND on their next call and can re-discover.

pub async fn set_lifecycle_state( &self, provider_id: &str, new_state: LifecycleState, detail: &str, ) -> Result<LifecycleState, Status>

Update the provider’s lifecycle state. Returns the previous value (or the inferred fallback when nothing’s been pushed yet) so callers can log “X went INACTIVE → ACTIVE” without a separate query. Validation is soft: illegal transitions log a warn but the new state is still stored. Strict validation will land later once telemetry confirms there are no spurious illegal transitions during atlas/provider startup-race conditions.

pub async fn heartbeat(&self, provider_id: &str) -> Result<u64, Status>

Updates last_heartbeat_ms to now. Returns the timestamp it set.

pub async fn declare( &self, provider_id: &str, contract_id: &str, transport: Transport, proposed: &str, params: TransportParams, description: &str, ) -> Result<String, Status>

Declare ONE transport for ONE contract on a registered provider. Returns the authoritative endpoint string (may differ from proposed when Atlas rewrote it to disambiguate on a mintable transport).

pub async fn query( &self, provider_id: &str, kind: Kind, contract: &str, transport: Transport, ) -> Vec<CapabilityProvider>

Snapshot of registered Providers matching the given filters. Empty provider_id / empty contract / Transport::Unspecified / Kind::Unspecified mean “no filter on that field”. Each returned record carries only the Capabilities that satisfy the contract + transport filters.

pub async fn query_with_prefix( &self, provider_id: &str, kind: Kind, contract: &str, namespace_prefix: &str, transport: Transport, ) -> Vec<CapabilityProvider>

pub async fn connect( &self, consumer_id: &str, provider_id: &str, contract_id: &str, transport: Transport, ) -> Result<(String, String, TransportParams), Status>

Open a channel to one (provider provider, contract, transport). Atlas only providers the edge — the consumer dials the returned endpoint itself (each transport has its own connect protocol; atlas can’t dial generically). Returns the allocated channel handle and the full binding the consumer needs.

pub async fn disconnect(&self, channel_id: &str) -> bool

Idempotent: returns true if a channel was removed, false if the id was unknown (already released, or auto-dropped when its provider went away).

pub async fn capability_md(&self, provider_id: &str) -> Result<String, Status>

Read the provider’s CAPABILITY.md content. Returns “” when the provider registered without a path.

pub async fn inspect_json(&self) -> Result<String, Status>

Debug-only JSON dump of the entire registry. Schema is unstable.

Trait Implementations

impl Debug for AtlasRegistry

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for AtlasRegistry

fn default() -> AtlasRegistry

Returns the “default value” for a type.