Subj : IRCd: 1 Hz blocking file_date() in the event loop turns a filesystem s
To   : GitLab note in main/sbbs
From : Rob Swindell
Date : Sat Sep 19 2026 23:49:14

https://gitlab.synchro.net/main/sbbs/-/work_items/1245#note_10389

## Correction to the attributed cause of the stall

The "What happened" section above attributes the CIFS stall to a web-scraping
storm saturating the Samba server. That attribution is not established, and a
related claim in #1246 (that a FastCGI busy-spin caused the wedge by write
volume) has been retracted there after measurement.

What is established, and what this issue is actually about, is unchanged:

- The IRCd's log output stopped for 91 seconds (17:00:33 to 17:02:04) and
  `sbbs.service` stopped for 82 seconds, both resuming in the same second.
- That second is when the CIFS client gave up and began returning `-EAGAIN`.
- The IRCd polls `file_date(system.ctrl_dir + "ircd.rehash")` once a second
  from its single-threaded event loop, so any stall of that `stat()` freezes
  the whole server.
- 91 seconds exceeds every configured ping interval, so about 50 server links
  expired at once when the loop resumed.

What put the CIFS client into that state is a separate question and remains
open. It does not affect the defect described here: any sufficiently long
filesystem stall, from any cause, produces the same network-wide split.

-- *Authored by Claude (Claude Code), on behalf of @rswindell*
--- SBBSecho 3.37-Linux
 * Origin: Vertrauen - [vert/cvs/bbs].synchro.net (1:103/705)

.