On 2024/11/27 13:52, Laurence Tratt wrote: > On Wed, Nov 27, 2024 at 01:08:51PM +0000, Stuart Henderson wrote: > > Hello Stuart, > > >> One minor thing: I wonder if suggesting mosquitto in `pkg/DESCR` is the > >> right thing on OpenBSD? > >> > >> There is a very long-standing bug (I first experienced it a few years > >> back, and it's still present as of a week or two back) on some > >> platforms, including OpenBSD, where mosquitto uses 100% CPU when clients > >> disconnect [1]. It baffled me for ages! > >> > >> Personally, as someone experimenting with zigbee2mqtt in the last couple > >> of weeks, I switched to rabbitmq, which seems to have no such issues. > >> > >> [1] https://github.com/eclipse-mosquitto/mosquitto/issues/2807 > > Interesting, I've been running mosquitto since at least 2022 and so far > > bit's been behaving itself quite well. > > The problem seems to be solely about whether a client with id X > disconnects: if it stays connected (including reconnects), there's no > problem; if it disconnects, CPU usage can go to 100%. I've run a default > config mosquitto and seen this problem on two machines (both with low > mosquitto usage), though it took me ages to realise what the cause was! If > your clients stay connected all, or at least most, of the time, I doubt > you'd see this problem. It also seems like it might only affect BSD > platforms, though I'm not totally certain about that.
mosquitto_sub uses a different id each time you run it (unless overridden with -i), mosquitto_sub_$pid, and I've definitely run that a few separate times recently. I've just had a play with a bunch of mosquitto_pub (just sending one message each run, and with a different id each time), not hit it there either. Are you using TLS?
