Struct Manager 

pub struct Manager {
    pub incidents: BTreeMap<IncidentId, Incident>,
    pub(crate) tracks: BTreeMap<UnitId, Track>,
    pub(crate) next_id: u64,
    pub startup_grace: Millis,
}

Fields

incidents: BTreeMap<IncidentId, Incident>

tracks: BTreeMap<UnitId, Track>

next_id: u64

startup_grace: Millis

How long after coming up a failing health probe is not yet a fault, for units that do not set their own (UnitDef::startup_grace).

Implementations

impl Manager

pub fn restore(incidents: impl IntoIterator<Item = Incident>) -> Manager

Rebuild from the incidents the journal holds (open and cooling ones).

pub fn started(&mut self, unit: &UnitId, now: Timestamp)

Planarian (or a user through it) just started or restarted unit: its startup grace begins again.

pub fn active_for(&self, unit: &UnitId) -> Option<&Incident>

pub(crate) fn active_id(&self, unit: &UnitId) -> Option<IncidentId>

pub(crate) fn close( &mut self, id: IncidentId, now: Timestamp, changed: &mut Vec<Incident>, )

pub fn tick( &mut self, now: Timestamp, state: &SystemState, defs: &UnitDefs, ) -> Vec<Incident>

Observe one tick. Returns every incident that opened, changed or closed.

pub fn tick_at( &mut self, now: Timestamp, decided: Timestamp, state: &SystemState, defs: &UnitDefs, ) -> Vec<Incident>

tick, with the decision’s own time: now is the observation’s (what the evidence is timed by), decided the clock read after it, once the probes have run. A hold set here (a third regression’s escalation) runs from decided, as every cooldown and backoff does (§13), so a slow observation never shortens it.

pub(crate) fn open( &mut self, target: &UnitId, kind: IncidentKind, divergence: (DesiredStatus, UnitStatus), now: Timestamp, next_attempt: Timestamp, regressions: u8, ) -> Incident

pub fn actionable(&self, now: Timestamp) -> Vec<&Incident>

Repair incidents the diagnoser may act on now. An incident whose target currently meets its desired state is stabilizing, not actionable: acting on it would find nothing to do and wrongly exhaust it.

pub fn diagnostic(&self) -> Vec<&Incident>

Diagnostic incidents (re-observe through another Readiness probe).

pub fn record_attempt( &mut self, id: IncidentId, txn: TxnId, outcome: Outcome, now: Timestamp, ) -> Option<Incident>

Record an attempt’s transaction. Spends budget, sets the next backoff (longer after regressions), and enters cooldown when the budget is gone.

pub(crate) fn cool(i: &mut Incident, now: Timestamp)

pub fn exhausted(&mut self, id: IncidentId, now: Timestamp) -> Option<Incident>

The diagnoser found nothing (A: report to the user). The incident goes quiet with a cooldown rather than retrying forever (§1 step 4).

pub fn escalate( &mut self, id: IncidentId, reason: EscalationReason, now: Timestamp, ) -> Option<Incident>

Hand the incident to a person, with why (§13): no autonomous action until the hold ends or someone acts on the target.

pub fn human_acted( &mut self, target: &UnitId, now: Timestamp, ) -> Option<Incident>

A human acted on the target (release, explicit request): lift any hold.

Trait Implementations

impl Clone for Manager

fn clone(&self) -> Manager

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Manager

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

Formats the value using the given formatter.

impl Default for Manager

fn default() -> Manager

Returns the “default value” for a type.