On Wed, Oct 2, 2013 at 9:40 AM, Thomas Eckert <thomas.r.w.eck...@gmail.com>wrote:
> Yann, although I do expect it to solve the issue discussed here, I don't > think simply flushing everything instantly is the right way to go. For > example, how do the proposed changes work with modules which scan the > request body like mod_security ? A lot of scanning/parsing can only be done > in a senseful way if the entity to be scanned/parsed is available in full. > Indeed, this patch won't help if an input filter in the chain does spool the body. Maybe mod_proxy_http can partially prefetch the request body before the "acquire/determine/connect backend" logic so that the is_socket_connected() runs just before the first byte is sent (regardless of the time taken by the first input filters call) ? The flushall flag would still (optionally) be welcome :p > > Passing through everything immediately will probably cause the SlowHTTP to > bybass the reverse proxy and harm the web server direclty. (I haven't had > time to appl the changes to a test environment) > SlowHTTP does currently eat Apache resources (the thread won't do anything else until all is forwarded to the backend, or even back to the client), though I don't know much about mpm_event still... IMHO that cannot be avoided without mod_reqtimeout (mod_security?). Regards, Yann.
