nx_parallel_ingest.nx

nx_parallel_ingest.nx -- ingest many exceeds/capabilities IN PARALLEL without running out of resources. The mechanism is three guarantees: 1. BOUNDED CONCURRENCY -- a worker pool of at most K live forks at any instant (never a fork-bomb); items stream through the pool, K busy at a time, the rest queued. 2. PROCESS ISOLATION -- each worker is a fork; its scratch memory lives in ITS address space and is reclaimed by the kernel on exit, so the parent's memory never grows with the work. Per-item RAM is O(1) in the parent, not O(n). 3. COMPACT SHARED BANK -- results are fixed-size records in ONE shared mmap (MAP_SHARED, 0x21, so children write what the parent reads); the bank is O(n) bytes, not O(n) machine code. So total resource use = K*(worker RAM) + n*(record size), both bounded and tunable -- you can ingest thousands of exceeds with a handful of cores and a few KB. license_tier: ORIGINAL