On Fri, Jul 5, 2019 at 10:37 AM Ruediger Pluem <rpl...@apache.org> wrote: > > Wouldn't that bio_filter_out_flush in the patch cause a blocking write, > although we were > non blocking?
That's what we already do in ssl_io_filter_output() (e.g. EAGAIN from reading the bucket just above). We never set (only clear) the bio_retry_flag in the bio_filter_out_*() functions which always flush, so I don't thing that openssl will ever return WANT_WRITE from SSL_write() with our mechanism. Anyway, if it happens (!) we are allowed to block here in the output filter, so FLUSH-ing is probably what makes more sense directly in this worker thread. A prossibly better option in trunk (and its pending/aside data mechanism) would be to break out and leave this bucket with the aside ones, then let WRITE_COMPLETION do the job when possible. Not for 2.4.x though.. Regards, Yann.
