monitor.rates¶

Counter->rate conversion shared by shell parsers and the SNMP path.

Monotonic counters (network bytes, disk sectors, SNMP Counter32) become per-second rates here. One rule for both channels: a negative delta means the counter reset (device reboot) or wrapped — return None, re-baseline, and emit no point for that tick. Rates divide by actual elapsed time between samples, never the nominal interval, so they are correct at any cadence and across missed ticks.

otto.monitor.rates.compute_rate(prev_value: float, cur_value: float, dt: float) → float | None¶

Per-second rate over one sample interval, or None when undefined.

None on a non-positive dt (clock anomaly / duplicate tick) or a negative delta (counter reset or wrap — reboots are common on test beds, wraps are rare, and wrap-compensation would turn every reboot into one absurd spike; losing one tick is the better trade).

class otto.monitor.rates.RateTracker¶

Bases: object

Per-key previous-sample state for counter->rate conversion.

Shell rate parsers hold one as instance state (parser instances are per-target deep copies, so state never leaks across hosts); the SNMP path holds one per SnmpSource.

update(key: str, value: float, ts: datetime) → float | None¶

Record (value, ts) for key and return the rate since the previous sample.

Returns None on first sighting / reset (see compute_rate()).

prune(active: set[str]) → None¶

Drop state for keys not in active (e.g. a vanished interface).