Struct MemoryDeviceStore 

pub struct MemoryDeviceStore { /* private fields */ }

Available on crate feature device only.

In-memory device store backed by DashMap.

For testing and single-node development only. Not suitable for production: data is lost on restart, no encryption at rest, no cross-node visibility.

Time injection

record_sighting and set_trust_level accept now as a parameter so the store itself does not own a Clock; callers thread the injected clock from the surrounding service. This mirrors how SessionStore::cycle accepts caller-supplied state for DST testability.

Implementations

impl MemoryDeviceStore

pub fn new() -> Self

Create an empty in-memory device store.

pub fn with_sweep_config(self, config: SweepConfig) -> Self

Override the SweepConfig driving the retention ladder.

pub fn len(&self) -> usize

Number of devices currently held. Useful in tests; not part of the trait surface.

pub fn is_empty(&self) -> bool

true when no devices are held.

Trait Implementations

impl Clone for MemoryDeviceStore

fn clone(&self) -> MemoryDeviceStore

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Default for MemoryDeviceStore

fn default() -> MemoryDeviceStore

Returns the “default value” for a type.

impl DeviceStore for MemoryDeviceStore

type Error = MemoryDeviceStoreError

The error type returned by storage operations.

async fn load( &self, tenant_id: &TenantId, id: &DeviceId, ) -> Result<Option<Device>, Self::Error>

Load the device by id. Returns None if absent.

async fn find_by_fingerprint( &self, tenant_id: &TenantId, hash: &FingerprintHash, ) -> Result<Option<Device>, Self::Error>

Look up a device by its keyed fingerprint within a tenant. Used as the fast-path during request handling when no device_id cookie is present yet. Collisions must be vanishingly rare given a per-tenant HMAC key, but implementations MUST scope the query by tenant_id to prevent cross-tenant correlation.

async fn find_for_user( &self, tenant_id: &TenantId, user_id: &UserId, limit: usize, ) -> Result<Vec<Device>, Self::Error>

List active devices for a user, newest-sighted first. limit caps the result so a high-cardinality user doesn’t blow up device-management UIs.

async fn find_by_refresh_family( &self, tenant_id: &TenantId, family_id: &str, ) -> Result<Vec<Device>, Self::Error>

Find every device in tenant carrying a DeviceBinding::Refresh { family_id } matching the supplied family_id. Used by the refresh-cascade path to convert “this refresh-token family was compromised” into the list of Devices to revoke.

async fn save(&self, device: &Device) -> Result<(), Self::Error>

Persist a new device row, or overwrite an existing one. Idempotent.

async fn record_sighting( &self, tenant_id: &TenantId, id: &DeviceId, now: DateTime<Utc>, ) -> Result<(), Self::Error>

Touch the last_seen_at timestamp for the given device. Implementations SHOULD perform this as a single UPDATE rather than a load-modify-save round trip.

async fn set_trust_level( &self, tenant_id: &TenantId, id: &DeviceId, level: DeviceTrustLevel, now: DateTime<Utc>, ) -> Result<(), Self::Error>

Set the trust level. Drives transitions across the DeviceTrustLevel ladder. Setting to DeviceTrustLevel::Revoked also stamps revoked_at = now.

async fn delete( &self, tenant_id: &TenantId, id: &DeviceId, ) -> Result<(), Self::Error>

Hard-delete the row. Idempotent. Used by the retention sweep once a Revoked device has aged out of the configured grace window, and by Art 17 erasure cascades.

async fn sweep( &self, tenant_id: &TenantId, now: DateTime<Utc>, ) -> Result<SweepCounts, Self::Error>

Drive the three-stage retention ladder: Trusted → Seen (after the configured Trusted-idle window), Seen → Revoked (after the configured Seen-idle window), Revoked → purged (after the configured grace window). Returns the count of rows changed at each stage.

fn find_active_for_user( &self, tenant_id: &TenantId, user_id: &UserId, limit: usize, ) -> impl Future<Output = Result<Vec<Device>, Self::Error>> + Send

Same as find_for_user but excludes Revoked rows. Default implementation filters in-memory after a full find_for_user call; SQL backends SHOULD override with a WHERE trust_level != 'Revoked' clause that the optimiser can use against the trust-level index.

impl HealthCheck for MemoryDeviceStore

fn check(&self) -> Pin<Box<dyn Future<Output = HealthStatus> + Send + '_>>

Probe the component and return its current health.