Struct FieldObserver 

pub struct FieldObserver { /* private fields */ }

Capped, opt-in field-name counter shared across producers (the daemon’s event task, the eval streaming loop) and consumers (the daemon’s HTTP handlers, the eval report writer).

Implementations

impl FieldObserver

pub fn new(max_keys: usize) -> Self

Create a new observer with the given upper bound on distinct keys.

A max_keys of 0 is allowed and disables tracking entirely; every observed field counts as overflow. Callers wanting “no cap” should pick a large finite number (e.g. usize::MAX / 2).

pub fn observe<E: Event + ?Sized>(&self, event: &E)

Walk the event’s field keys and update the per-field counters.

Insertion of a new key is skipped once the observer is at capacity; already-tracked keys keep counting. The method takes &self, so the observer can be shared behind an Arc without locking from the caller’s side.

pub fn snapshot(&self) -> FieldObservation

Snapshot the current counts. Entries are sorted by descending count, then by ascending name for deterministic output.

Cheap relative to the cardinality of the observer: each entry only refcount-clones the Arc<str> key rather than copying the key bytes, so a 10 000-key snapshot costs ~10 000 atomic increments plus one Vec allocation, not 10 000 String allocations.

pub fn reset(&self) -> (usize, u64)

Reset every counter and the overflow tally. Returns the previous (unique_keys, events_observed) pair so the API endpoint can report what was cleared.

pub fn events_observed(&self) -> u64

Total events observed since the observer was created or last reset.

pub fn lifetime_events_observed(&self) -> u64

Lifetime total of events observed since the observer was constructed, ignoring resets. Monotonic; suitable for driving Prometheus counters.

pub fn unique_keys(&self) -> usize

Distinct keys currently tracked (does not include overflow drops).

pub fn overflow_dropped(&self) -> u64

Insert attempts dropped because the observer was at capacity since the last reset.

pub fn lifetime_overflow_dropped(&self) -> u64

Lifetime total of insert attempts dropped because the observer was at capacity, ignoring resets. Monotonic.

pub fn max_keys(&self) -> usize

Configured per-observer ceiling for distinct keys.