Struct ClusterSupervisor 

pub struct ClusterSupervisor { /* private fields */ }

The cluster supervisor: health scoring + automatic range failover planning.

Holds only the HealthPolicy; all live state is read through ClusterSignals at scan time, so one supervisor instance serves the whole cluster lifetime.

Implementations

impl ClusterSupervisor

pub fn new(policy: HealthPolicy) -> ClusterSupervisor

A supervisor with the given health policy.

pub fn policy(&self) -> &HealthPolicy

pub fn assess(&self, signals: &MemberSignals) -> HealthScore

Score a single member’s health under the policy. The building block of degraded-member detection: an operator surface calls this for every authorized member to render a health view.

pub fn assess_members( &self, membership: &MembershipCatalog, signals: &impl ClusterSignals, ) -> BTreeMap<NodeIdentity, HealthScore>

Score every authorized member of membership, in stable identity order. Includes healthy, degraded, and failed members alike — the input to a cluster health dashboard.

pub fn plan_failovers( &self, membership: &MembershipCatalog, ownership: &ShardOwnershipCatalog, signals: &impl ClusterSignals, ) -> FailoverPlan

Plan automatic failovers across the whole ownership catalog without mutating it. For each range whose owner is failover-eligible (Failed and past the grace period), pick the safest caught-up replica candidate and produce a PlannedPromotion; if no replica is safe, record a BlockedFailover. Owners that are healthy, merely degraded, or still inside their grace period produce nothing.

pub fn run_failovers( &self, membership: &MembershipCatalog, ownership: &mut ShardOwnershipCatalog, signals: &impl ClusterSignals, ) -> (Vec<Result<TransitionOutcome, TransitionError>>, FailoverPlan)

Plan failovers and immediately run the safe promotions through the ownership transition machine, fencing each failed owner via the epoch bump. Returns the activated TransitionOutcomes and the surviving FailoverPlan (whose blocked entries still need attention; its promotions are the requests that were run).

Each promotion is an independent catalog entry, so running them in sequence never invalidates another’s CAS. A promotion whose CAS lost a race (the catalog moved between planning and activation) surfaces as a TransitionError in the returned vector rather than aborting the rest.

Trait Implementations

impl Clone for ClusterSupervisor

fn clone(&self) -> ClusterSupervisor

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ClusterSupervisor

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

Formats the value using the given formatter.

impl Default for ClusterSupervisor

fn default() -> ClusterSupervisor

Returns the “default value” for a type.