upssched v2.8.0 hogs CPU after a while (Debian 12 package)

[jimklimov]

Raised in https://alioth-lists.debian.net/pipermail/nut-upsuser/2023-June/013338.html

Per initial analysis in the thread, it seems that the tight loop runs when select() returns quickly but read() has nothing to report. Why - that is uncertain now. Possibly we miss the detection of a closed socket (disconnected client) and the raised "I am ready" flag is an attempt to convey that?..

Might be or not be related to #1274 which landed a couple of months before NUT v2.8.0 release: not on first sight, but need a repro case to say for sure.

One easy-pickings solution (to CPU hogging) would be to detect that we spend almost no time in the loop to read nothing, and induce a delay (e.g. select() would have waited a second... might as well try to usleep() the rest of it).

However it is best to reproduce the problem and understand why it happens - if we lost the ability to forget disconnected clients, it is a path to resource leak and crash later on. And might be somehow related to other upssched issues reported over the past year or two...