Caller-owned map of checks that have not settled yet, keyed by probe name.
Abandoning a check at the deadline does not cancel it. Against a
partitioned Redis the driver holds the PING in its offline queue until
reconnect, so without this every scrape adds another pending command —
unbounded during a long outage, and released as a burst on recovery. With
it, a probe whose previous check is still running is joined rather than
re-issued: one in-flight command per dependency no matter how often the
endpoint is scraped. Each caller still applies its own deadline, so a
joining request does not inherit how long the shared check has already
been waiting.
Pass the same map across calls (the readiness router holds one for its
lifetime). Omit it and every call issues its own check.
ReadonlytimeoutMs
timeoutMs:number
Per-probe deadline. A probe that has not settled by then is reported as
failed. This has to be shorter than whatever scrape interval the
orchestrator uses, or an unready pod reads as a slow one.
Caller-owned map of checks that have not settled yet, keyed by probe name.
Abandoning a check at the deadline does not cancel it. Against a partitioned Redis the driver holds the
PINGin its offline queue until reconnect, so without this every scrape adds another pending command — unbounded during a long outage, and released as a burst on recovery. With it, a probe whose previous check is still running is joined rather than re-issued: one in-flight command per dependency no matter how often the endpoint is scraped. Each caller still applies its own deadline, so a joining request does not inherit how long the shared check has already been waiting.Pass the same map across calls (the readiness router holds one for its lifetime). Omit it and every call issues its own check.