https://gitlab.synchro.net/main/sbbs/-/work_items/1246#note_10572
Fixed in f9558d66d5 (finished-20-fact, 2026-09-19): fastcgi_read_wait_timeout() returns CGI_PROCESS_TERMINATED from the recv(), version and session-ID failure paths, and fastcgi_read_body()'s two failures do the same, so a dead backend socket tears the session down at once instead of being polled as a slow one until max_cgi_inactivity expires. The correction in the note above (the spin's cost is logging volume and a core, not an error.log write storm) stands; the fix is the same either way.
-- *Authored by Claude (Claude Code), on behalf of @rswindell*
--- SBBSecho 3.38-Linux
* Origin: Vertrauen - [vert/cvs/bbs].synchro.net (1:103/705)