On Fri, Mar 6, 2015 at 5:24 PM, Plüm, Rüdiger, Vodafone Group <[email protected]> wrote: > > >> -----Ursprüngliche Nachricht----- >> Von: Yann Ylavic [mailto:[email protected]] >> Gesendet: Freitag, 6. März 2015 17:02 >> An: httpd >> Betreff: Re: error_time reset in proxy_util.c >> >> >> If the worker is *not a BalancerMember*, couldn't we simply mark the >> connection as close = 1 here, and continue with it until the end of >> this request? >> This is to avoid responding 503 whereas it succeed for this particular >> request (since this is a race, the worker was put in error state a >> very little time ago, not at ap_proxy_acquire_connection() time for >> this request at least). > > IMHO the situation does not really differ between BalancerMember and non > BalancerMembers. > Another process / thread detected an error condition that might not be caused > by > a failed connect on TCP layer, but by a failonstatus / failontimeout by a > previous > request. We should respect this here as we would when this worker was marked > as down > a few moments earlier which would have prevented us to the connection at all.
Actually since r1650028 (in trunk), non balancer-member workers are not put in error state by default (PROXY_WORKER_IGNORE_ERRORS). This has not been backported (reviewed?) yet, but comes from PR 48388. So this should not affect them (by default), +1. Regards, Yann.
