Struct PerInstanceReporter 

pub struct PerInstanceReporter { /* private fields */ }

Per-instance JSONL reporter.

No file-handle cache. The previous implementation kept every (metric_name, labels) combination’s File alive in a HashMap for the entire run, which monotonically grew the open-FD count — for workloads with many distinct label tuples (e.g. phase × profile × optimize_for × k × r), the process would exhaust its ulimit -n. Each report() call now opens, writes, and closes the target file per record. At the 30s cadence the metrics scheduler runs this on, the open() cost is dwarfed by the rest of the snapshot path.

Implementations

impl PerInstanceReporter

pub fn new(dir: impl AsRef<Path>) -> Result<Self, String>

Construct a reporter rooted at dir. The directory is created if missing; an existing directory is reused so consecutive sessions can co-locate output when desired.

Trait Implementations

impl Reporter for PerInstanceReporter

fn report(&mut self, snapshot: &MetricSet)

fn flush(&mut self)

fn finished(&self) -> bool

Self-termination signal. After a report that leaves this true, the subscriber’s cadence-feed dispatch worker exits its loop (calling flush on the way out) — the subscriber receives no further pulses. A one-shot subscriber — e.g. a settle / stop evaluator that has set a terminal phase disposition — uses this to unregister itself without a self-join deadlock (it runs on the worker thread, so it cannot call unsubscribe on itself directly). Default false (a long-lived subscriber that never self-terminates).