Admin Debug Memory

In-process leak diagnostic: RSS + tracemalloc growth vs baseline.

WARNING: tracemalloc adds per-allocation CPU overhead. On a busy hub prefer the low-overhead gc path (no ?enable), and keep frames small.

Params (runtime-controllable, no restart): ?gc=1 one-shot gc object-type histogram (heaviest read; run it twice a few min apart and diff gc_top_types) ?enable=1 start tracemalloc + baseline (default frames=1) ?frames=N tracemalloc stack depth, 1-10 (higher = more CPU) ?disable=1 stop tracemalloc (turn overhead back off) ?reset_baseline=1 re-zero the growth baseline (measure from now) ?top=N cap reported sites (default 20)

Recommended low-risk run: GET ?gc=1 twice ~5 min apart, compare the type whose count grew. Only if inconclusive, GET ?enable=1&frames=1 for a SHORT window (2-5 min), GET (read top_growth), then ?disable=1.