On Mon, Jan 13, 2014 at 5:38 PM, Jim Jagielski <j...@jagunet.com> wrote: > This happens, at most, what, maybe 2 times? Is that > really an issue?
The worst case is "2 x number of balancer's workers" tries, when the request is a POST, ping is configured and all the balancer's wokers aren't connect()able. Whether it is or not a big deal depends on the backend's handling of Via/X-Forwarded-* headers duplicated values, or the log parser (since these values are there too). > And if so, since ap_proxy_http_request() > is local static, we could certainly pass the number of retries > to it and bypass the extra call to ap_proxy_create_hdrbrgd() > on retries, right? That would certainly help for the ping failure(s), but not for balancer's workers (recoverable) ones, where the whole process is replayed. Note the patch does not really add cycles since the copy of headers_in is already there to preserve "Connection" header (needed by output filter), so maybe it's not worth handle both cases differently (but why not, your proposal looks simple enough to handle the common case with no balancer). Regards.