Fix rtt assignement for low values of infra-cache-max-rtt (#1070)

gthess · web-flow · commit 3ff5c7a74d10 · 2024-05-16T13:42:32.000+02:00
* Fix rtt assignement for still useful servers when a low value for
  infra-cache-max-rtt is configured.
diff --git a/iterator/iter_utils.c b/iterator/iter_utils.c
@@ -279,9 +279,10 @@ iter_filter_unsuitable(struct iter_env* iter_env, struct module_env* env,
 		name, namelen, qtype, &lame, &dnsseclame, &reclame,
 		&rtt, now)) {
 		log_addr(VERB_ALGO, "servselect", &a->addr, a->addrlen);
-		verbose(VERB_ALGO, "   rtt=%d%s%s%s%s", rtt,
+		verbose(VERB_ALGO, "   rtt=%d%s%s%s%s%s", rtt,
 			lame?" LAME":"",
 			dnsseclame?" DNSSEC_LAME":"",
+			a->dnsseclame?" ADDR_DNSSEC_LAME":"",
 			reclame?" REC_LAME":"",
 			a->lame?" ADDR_LAME":"");
 		if(lame)
diff --git a/services/cache/infra.c b/services/cache/infra.c
@@ -60,6 +60,16 @@
  * can do this number of packets (until those all timeout too) */
 #define TIMEOUT_COUNT_MAX 3
 
+/** Minus 1000 because that is outside of the RTTBAND, so
+ * blacklisted servers stay blacklisted if this is chosen.
+ * If USEFUL_SERVER_TOP_TIMEOUT is below 1000 (configured via RTT_MAX_TIMEOUT,
+ * infra-cache-max-rtt) change it to just above the RTT_BAND. */
+#define STILL_USEFUL_TIMEOUT (				\
+	USEFUL_SERVER_TOP_TIMEOUT < 1000 ||		\
+	USEFUL_SERVER_TOP_TIMEOUT - 1000 <= RTT_BAND	\
+		?RTT_BAND + 1				\
+		:USEFUL_SERVER_TOP_TIMEOUT - 1000)
+
 /** ratelimit value for delegation point */
 int infra_dp_ratelimit = 0;
 
@@ -656,7 +666,7 @@ infra_update_tcp_works(struct infra_cache* infra,
 	if(data->rtt.rto >= RTT_MAX_TIMEOUT)
 		/* do not disqualify this server altogether, it is better
 		 * than nothing */
-		data->rtt.rto = RTT_MAX_TIMEOUT-1000;
+		data->rtt.rto = STILL_USEFUL_TIMEOUT;
 	lock_rw_unlock(&e->lock);
 }
 
@@ -796,34 +806,31 @@ infra_get_lame_rtt(struct infra_cache* infra,
 		&& infra->infra_keep_probing) {
 		/* single probe, keep probing */
 		if(*rtt >= USEFUL_SERVER_TOP_TIMEOUT)
-			*rtt = USEFUL_SERVER_TOP_TIMEOUT-1000;
+			*rtt = STILL_USEFUL_TIMEOUT;
 	} else if(host->rtt.rto >= PROBE_MAXRTO && timenow < host->probedelay
 		&& rtt_notimeout(&host->rtt)*4 <= host->rtt.rto) {
 		/* single probe for this domain, and we are not probing */
 		/* unless the query type allows a probe to happen */
 		if(qtype == LDNS_RR_TYPE_A) {
 			if(host->timeout_A >= TIMEOUT_COUNT_MAX)
 				*rtt = USEFUL_SERVER_TOP_TIMEOUT;
-			else	*rtt = USEFUL_SERVER_TOP_TIMEOUT-1000;
+			else	*rtt = STILL_USEFUL_TIMEOUT;
 		} else if(qtype == LDNS_RR_TYPE_AAAA) {
 			if(host->timeout_AAAA >= TIMEOUT_COUNT_MAX)
 				*rtt = USEFUL_SERVER_TOP_TIMEOUT;
-			else	*rtt = USEFUL_SERVER_TOP_TIMEOUT-1000;
+			else	*rtt = STILL_USEFUL_TIMEOUT;
 		} else {
 			if(host->timeout_other >= TIMEOUT_COUNT_MAX)
 				*rtt = USEFUL_SERVER_TOP_TIMEOUT;
-			else	*rtt = USEFUL_SERVER_TOP_TIMEOUT-1000;
+			else	*rtt = STILL_USEFUL_TIMEOUT;
 		}
 	}
 	/* expired entry */
 	if(timenow > host->ttl) {
-
 		/* see if this can be a re-probe of an unresponsive server */
-		/* minus 1000 because that is outside of the RTTBAND, so
-		 * blacklisted servers stay blacklisted if this is chosen */
 		if(host->rtt.rto >= USEFUL_SERVER_TOP_TIMEOUT) {
 			lock_rw_unlock(&e->lock);
-			*rtt = USEFUL_SERVER_TOP_TIMEOUT-1000;
+			*rtt = STILL_USEFUL_TIMEOUT;
 			*lame = 0;
 			*dnsseclame = 0;
 			*reclame = 0;
