nx_acceptq.nx

nx_acceptq.nx -- DETECTOR for the ACCEPT-LOOP WEDGE: a daemon that holds its listen socket but has stopped calling accept(). WHY THIS EXISTS (2026-07-31, measured). The mgmt API blocked in wait4 for the entire life of a forked build. Connections completed the TCP handshake -- the KERNEL accepts them into the backlog -- so every port-probe health check reported UP, for six minutes at a time, while no request was served. Three separate lanes filed it as "MGMT :18098 DOWN, ALL SEATS". :18098 was never down. It was blocked. ★★★ THE INSIGHT THIS ORGAN ENCODES: A SUCCESSFUL CONNECT PROVES THE KERNEL ACCEPTED, NOT THE PROCESS. The two are indistinguishable from outside -- unless you read the one number the kernel already keeps: on a LISTEN socket, /proc/net/tcp's rx_queue IS THE COUNT OF COMPLETED-BUT-UNACCEPTED CONNECTIONS. Non-zero means the app is behind. Sustained non-zero means the app is not accepting at all. Measured on the live wedge: rx_queue climbed 8 -> 34 while wedged, and returned to 0 once fixed. This is a pure READER: it opens /proc/net/tcp and nothing else. It cannot perturb the thing it measures, which matters because the failure it hunts is caused by blocking. nx_acceptq -> report every LISTEN socket with a non-zero accept backlog nx_acceptq <port> -> report just that port; exit 1 if it looks wedged expect_exit: 0 license_tier: ORIGINAL