On Mon, 15 Aug 2016, Willem Jan Withagen wrote: > > A workaround might be to set a modest timeout (say, 1s) on the poll. > > Worst case, you wait up to 1s to shut down, which isn't to bad, and in the > > normal case you only wake up needlessly once per second (also not *too* > > bad). > > > >> The other solution would be to include a watch pipe in the poll fds. > >> And then write that pipe if you want to trigger the poll. Instead of > >> doing the shutdown. I think there is example code for this on the web, > >> it is called the self-pipe trick. > > > > That's probably the most elegant/correct solution... I'm not sure that teh > > current behavior where we call shutdown(2) on an unconnected socket and it > > wakes up poll(2) is well-defined behavior. > > I actually found more cases of the selfpipe solution, and it is a > portable ways of tackling the challenge. > But there are also more cases that use ::shutdown(). > Which I will have to review and see of they cause the same type of > errors. But perhaps should we just rewrite all the shutdown cases, just > to use the same code in all code. That would be my preference. Then we're not potentially abusing the API and FreeBSD isn't using a different path that isn't as well tested. Thanks! sage -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
