Struct QuotaTracker 

pub struct QuotaTracker { /* private fields */ }

The QuotaTracker.

Implementations

impl QuotaTracker

pub fn new(config: TokenMaxingConfig) -> Self

Build a tracker from the user’s config. Providers not eligible under the metered-never rule are dropped from the budget.

pub fn reload(&self, config: TokenMaxingConfig)

Reload the config (e.g. on PUT /api/config). Preserves tokens_used for providers that remain in the config.

pub fn availability(&self, provider: &str) -> Availability

The verdict for a single provider.

pub fn snapshots(&self) -> Vec<QuotaTrackerSnapshot>

All providers (eligible or not), in declared order. Used by the orchestrator to build the live status panel and the report.

pub fn record_failure( &self, provider: &str, class: FailureClass, resets_at: Option<DateTime<Utc>>, )

Reactive override — record a failure class for provider from a real agent run.

Cooldown policy (kept separate so a single transient hiccup doesn’t drain the pool for an hour):

• QuotaExhausted → full reset-window cooldown. The provider’s plan limit has been hit; waiting makes sense.
• Transient → short cooldown (60s by default). classify.rs lumps 429/rate-limit into the same bucket as 500/timeout/ network blip. We don’t want a single upstream hiccup to mark a provider CooledDown for the whole reset window — that defeats the drain-by-rotating-across-providers loop. A short cooldown lets the next retry succeed on the same provider once the blip clears, while still preventing hot-looping.
• Anything else (auth failure, model unavailable, etc.) → no cooldown. Those are recovery-coordinator concerns, not pool concerns.

pub fn clear_cooldown(&self, provider: &str)

Clear the cooldown for a provider (e.g. when a window is observed to have actually reset).

pub fn apply_recalibration( &self, provider: &str, remaining_percent: Option<f64>, resets_at: Option<DateTime<Utc>>, outcome: RecalibrationOutcome, ) -> bool

Record a recalibration event and snap the budget to the new state. Returns true if the snap was applied.

pub fn reserve(&self, provider: &str, tokens: u64) -> Result<(), ReserveError>

Reserve tokens against the self-tracked counter. Used by the TokenMaxer after dispatching each unit of work.

pub fn release(&self, provider: &str, tokens: u64)

Release reserved tokens. Used on task failure / retry.

pub fn snapshot(&self, provider: &str) -> Option<ProviderSnapshot>

The full self-tracked state for a provider. Used by the report.

pub fn recalibration_history(&self) -> Vec<RecalibrationRecord>

Recalibration history.

pub fn cooldown_history(&self) -> Vec<CooldownRecord>

Cooldown records. Used by the report.

pub fn config(&self) -> TokenMaxingConfig

Read-only access to the current config.