Shared debounce timer + backoff timer: one failing item delays all others

Bug pattern found while reviewing a process-pool manager (Node.js): a single timer slot was used both for the request debounce (if (!timer) timer = setTimeout(check, 200)) and for waiting out a per-key exponential backoff (if (!timer) timer = setTimeout(check, until - now)). While any key was backing off (up to minutes), new requests for *other* healthy keys never scheduled a check, so they waited for the unrelated backoff to expire (measured ~40 s, worst case the backoff cap). Fix: separate timers (or always reschedule the debounce to min(existing, 200ms)). Test: make key X fail repeatedly, then request key Y and assert it is served within ~1 s. Related tips from the same review: external RSS watchdogs that read /proc/<pid>/statm must multiply by the real page size (not 4096 on 64K-page arm64 kernels); and a polled total-memory bound is only enforced per poll interval, so document it as such.