ReadonlyconcurrencyThe maximum number of scheduled callbacks running at once, 1–4. A
timed-out callback that has not settled still occupies its slot, so a
hung indicator costs one slot and no more; the remaining slots keep
refreshing the other indicators. With as many hung callbacks as slots, no
scheduled check starts until one settles: each stalled alias keeps its
last observation with a growing ageMs (or stays never-observed), and
the snapshot's state turns stale once any retained observation ages
past staleAfterMs. Each cycle covers every scheduled indicator that is
not still in flight, rotating its starting point so no indicator is
starved.
ReadonlyindicatorsThe approved indicator names to collect on a schedule. Each must be a key
of HealthDiagnosticsOptions.indicators. At most 16 entries. A
name no indicator is registered under is skipped (it stays
never-observed), and the plugin logs a count-only warning at bootstrap.
ReadonlyintervalThe cadence between scheduled cycles, in milliseconds. 1,000–300,000.
ReadonlytimeoutThe per-check reporting deadline, in milliseconds. 1–30,000. A check
that has not settled within the deadline is reported as timed-out; the
deadline does not cancel the underlying callback.
The bounded, separately-controlled scheduled health collection (M98d).
Absent by default: an omitted
scheduledperforms no scheduled work. When present, it names a subset of the approved indicator names (the keys ofHealthDiagnosticsOptions.indicators) and supplies a cadence, a per-check reporting deadline, and a concurrency cap. A timeout here is a REPORTING bound, not a cancellation: a check that has not settled withintimeoutMsis reported astimed-out, but its raw callback remains in-flight until it actually settles, and no replacement check for it starts before then.Since
0.8.0