On Thu, Sep 24, 2015 at 5:57 PM, Eric Covener <[email protected]> wrote: > > I initially thought this as well, and it resembles something > potentially fishy I thought I saw in mod_h2, but I convinced myself > it's actually the EOR bucket being read/written in order (but our > response bytes are still buffered in openssl)
The EOR is supposed to live until the response is flushed (all buffered/previous buckets sent). Until then, the request is still alive and waiting to be logged, yet all its response bytes are accounted with ap_logio_add_bytes() *after* the corresponding apr_socket_send*(). The problem is that mod_logio's cfg->r is not (anymore) reliable when r is in the flush queue, since it may have been replaced by the next request's post_read_request()...
