If APR_POLLSET_WAKEABLE was more universal and, therefore, more widely tested, I'd be +1... as it is, let's see what the feedback is.
> On Aug 5, 2016, at 8:58 AM, Luca Toscano <toscano.l...@gmail.com> wrote: > > > > 2016-08-04 17:56 GMT+02:00 Luca Toscano <toscano.l...@gmail.com>: > Hi Apache Devs, > > there is an interesting bugzilla ticket about mpm_event and frequent > wake-ups: https://bz.apache.org/bugzilla/show_bug.cgi?id=57399 > > Would it be possible to avoid them adding APR_POLLSET_WAKEABLE to the > event_pollset flags and calling apr_pollset_wakeup right after > apr_skiplist_insert? > > I might have completely misunderstood the code, but I wanted to bring up the > issue to have more expert opinions from the list. The skiplist use case is > still a bit unclear to me even if we discussed it a while ago in another > email thread. > > Very high level idea: http://apaste.info/MKE > > Remarks from APR: > > * @remark If flags contains APR_POLLSET_WAKEABLE, then a pollset is > * created with additional internal pipe object used for the > * apr_pollset_wakeup() call. The actual size of pollset is > * in that case size + 1. This feature is only supported on some > * platforms; the apr_pollset_create_ex() call will fail with > * APR_ENOTIMPL on platforms where it is not supported. > > Moreover for apr_pollset_wakeup: "If the pollset was not created with > APR_POLLSET_WAKEABLE the return value is APR_EINIT." > > This is only an idea of course, I am probably missing a lot of pieces, but > let me know if you have any feedback. > > Regards, > > Luca >