lib/nsrep: some changes in NS selection algorythm

daemon/lua/kres-gen.lua

@@ -96,6 +96,7 @@ struct kr_qflags {
 	_Bool FORWARD : 1;
 	_Bool DNS64_MARK : 1;
 	_Bool CACHE_TRIED : 1;
+	_Bool NO_NS_FOUND : 1;
 };
 typedef struct {
 	knot_rrset_t **at;

daemon/worker.c

@@ -1329,7 +1329,7 @@ static void on_udp_timeout(uv_timer_t *timer)
 				VERBOSE_MSG(qry, "=> server: '%s' flagged as 'bad'\n", addr_str);
 			}
 			kr_nsrep_update_rtt(&qry->ns, choice, KR_NS_TIMEOUT,
-					    worker->engine->resolver.cache_rtt, KR_NS_UPDATE);
+					    worker->engine->resolver.cache_rtt, KR_NS_RESET);
 		}
 	}
 	task->timeouts += 1;

lib/nsrep.c

@@ -162,10 +162,10 @@ static int eval_nsrep(const char *k, void *v, void *baton)
 	 * The fastest NS is preferred by workers until it is depleted (timeouts or degrades),
 	 * at the same time long distance scouts probe other sources (low probability).
 	 * Servers on TIMEOUT (depleted) can be probed by the dice roll only */
-	if (score <= ns->score && (qry->flags.NO_THROTTLE || score < KR_NS_TIMEOUT)) {
+	if (score <= ns->score && (score < KR_NS_LONG  || (qry->flags.NO_THROTTLE && (score < KR_NS_TIMEOUT)))) {
 		update_nsrep_set(ns, (const knot_dname_t *)k, addr_choice, score);
 		ns->reputation = reputation;
-	} else {
+	} else if (score < KR_NS_TIMEOUT) {
 		/* With 10% chance, probe server with a probability given by its RTT / MAX_RTT */
 		if ((kr_rand_uint(100) < 10) && (kr_rand_uint(KR_NS_MAX_SCORE) >= score)) {
 			/* If this is a low-reliability probe, go with TCP to get ICMP reachability check. */

lib/resolve.c

@@ -863,7 +863,7 @@ static void update_nslist_score(struct kr_request *request, struct kr_query *qry
 		}
 	/* Penalise resolution failures except validation failures. */
 	} else if (!(qry->flags.DNSSEC_BOGUS)) {
-		kr_nsrep_update_rtt(&qry->ns, src, KR_NS_TIMEOUT, ctx->cache_rtt, KR_NS_RESET);
+		kr_nsrep_update_rtt(&qry->ns, src, KR_NS_TIMEOUT, ctx->cache_rtt, KR_NS_UPDATE);
 		WITH_VERBOSE(qry) {
 			char addr_str[INET6_ADDRSTRLEN];
 			inet_ntop(src->sa_family, kr_inaddr(src), addr_str, sizeof(addr_str));
@@ -1433,8 +1433,12 @@ int kr_resolve_produce(struct kr_request *request, struct sockaddr **dst, int *t
 			} else {
 				VERBOSE_MSG(qry, "=> no valid NS left\n");
 			}
-			ITERATE_LAYERS(request, qry, reset);
-			kr_rplan_pop(rplan, qry);
+			if (!qry->flags.NO_NS_FOUND) {
+				qry->flags.NO_NS_FOUND = true;
+			} else {
+				ITERATE_LAYERS(request, qry, reset);
+				kr_rplan_pop(rplan, qry);
+			}
 			return KR_STATE_PRODUCE;
 		}
 	}

lib/rplan.h

@@ -62,6 +62,7 @@ struct kr_qflags {
 	bool FORWARD : 1;        /**< Forward all queries to upstream; validate answers. */
 	bool DNS64_MARK : 1;     /**< Internal mark for dns64 module. */
 	bool CACHE_TRIED : 1;    /**< Internal to cache module. */
+	bool NO_NS_FOUND : 1;    /**< No valid NS found during last PRODUCE stage. */
 };
 
 /** Combine flags together.  This means set union for simple flags. */

modules/serve_stale/serve_stale.lua

@@ -26,8 +26,8 @@ M.layer = {
 
 		local now = ffi.C.kr_now()
 		local deadline = qry.creation_time_mono + M.timeout
-		if now > deadline then
-			--log('[     ][stal]   => deadline has passed')
+		if now > deadline or qry.flags.NO_NS_FOUND then
+			log('[     ][stal]   => deadline has passed')
 			qry.stale_cb = M.callback
 			-- TODO: probably start the same request that doesn't stale-serve,
 			-- but first we need some detection of non-interactive / internal requests.