Stats for discard-timeout and wait-limit (#1159)

wcawijngaards · web-flow · commit 75967e7f5eca · 2024-11-01T15:52:58.000+01:00
* - Stats num.queries_discard_timeout and num.queries_wait_limit are used
  instead of the mesh dropped that of requests exceeded.
diff --git a/daemon/remote.c b/daemon/remote.c
@@ -707,6 +707,10 @@ print_stats(RES* ssl, const char* nm, struct ub_stats_info* s)
 		(unsigned long)s->svr.num_queries_cookie_client)) return 0;
 	if(!ssl_printf(ssl, "%s.num.queries_cookie_invalid"SQ"%lu\n", nm,
 		(unsigned long)s->svr.num_queries_cookie_invalid)) return 0;
+	if(!ssl_printf(ssl, "%s.num.queries_discard_timeout"SQ"%lu\n", nm,
+		(unsigned long)s->svr.num_queries_discard_timeout)) return 0;
+	if(!ssl_printf(ssl, "%s.num.queries_wait_limit"SQ"%lu\n", nm,
+		(unsigned long)s->svr.num_queries_wait_limit)) return 0;
 	if(!ssl_printf(ssl, "%s.num.cachehits"SQ"%lu\n", nm,
 		(unsigned long)(s->svr.num_queries
 			- s->svr.num_queries_missed_cache))) return 0;
diff --git a/daemon/stats.c b/daemon/stats.c
@@ -281,6 +281,10 @@ server_stats_compile(struct worker* worker, struct ub_stats_info* s, int reset)
 		s->svr.rpz_action[i] += (long long)worker->env.mesh->rpz_action[i];
 	timehist_export(worker->env.mesh->histogram, s->svr.hist,
 		NUM_BUCKETS_HIST);
+	s->svr.num_queries_discard_timeout +=
+		(long long)worker->env.mesh->num_queries_discard_timeout;
+	s->svr.num_queries_wait_limit +=
+		(long long)worker->env.mesh->num_queries_wait_limit;
 	/* values from outside network */
 	s->svr.unwanted_replies = (long long)worker->back->unwanted_replies;
 	s->svr.qtcp_outgoing = (long long)worker->back->num_tcp_outgoing;
@@ -451,6 +455,9 @@ void server_stats_add(struct ub_stats_info* total, struct ub_stats_info* a)
 	total->svr.num_queries_cookie_valid += a->svr.num_queries_cookie_valid;
 	total->svr.num_queries_cookie_client += a->svr.num_queries_cookie_client;
 	total->svr.num_queries_cookie_invalid += a->svr.num_queries_cookie_invalid;
+	total->svr.num_queries_discard_timeout +=
+		a->svr.num_queries_discard_timeout;
+	total->svr.num_queries_wait_limit += a->svr.num_queries_wait_limit;
 	total->svr.num_queries_missed_cache += a->svr.num_queries_missed_cache;
 	total->svr.num_queries_prefetch += a->svr.num_queries_prefetch;
 	total->svr.num_queries_timed_out += a->svr.num_queries_timed_out;
diff --git a/doc/unbound-control.8.in b/doc/unbound-control.8.in
@@ -422,6 +422,12 @@ number of queries with a client part only DNS Cookie by thread
 .I threadX.num.queries_cookie_invalid
 number of queries with an invalid DNS Cookie by thread
 .TP
+.I threadX.num.queries_discard_timeout
+number of queries removed due to discard-timeout by thread
+.TP
+.I threadX.num.queries_wait_limit
+number of queries removed due to wait-limit by thread
+.TP
 .I threadX.num.cachehits
 number of queries that were successfully answered using a cache lookup
 .TP
@@ -511,6 +517,12 @@ summed over threads.
 .I total.num.queries_cookie_invalid
 summed over threads.
 .TP
+.I total.num.queries_discard_timeout
+summed over threads.
+.TP
+.I total.num.queries_wait_limit
+summed over threads.
+.TP
 .I total.num.cachehits
 summed over threads.
 .TP
diff --git a/libunbound/unbound.h b/libunbound/unbound.h
@@ -849,6 +849,10 @@ struct ub_server_stats {
 	long long mem_quic;
 	/** number of queries over (DNS over) QUIC */
 	long long qquic;
+	/** number of queries removed due to discard-timeout */
+	long long num_queries_discard_timeout;
+	/** number of queries removed due to wait-limit */
+	long long num_queries_wait_limit;
 };
 
 /**
diff --git a/services/mesh.c b/services/mesh.c
@@ -214,6 +214,8 @@ mesh_create(struct module_stack* stack, struct module_env* env)
 	mesh->stats_dropped = 0;
 	mesh->ans_expired = 0;
 	mesh->ans_cachedb = 0;
+	mesh->num_queries_discard_timeout = 0;
+	mesh->num_queries_wait_limit = 0;
 	mesh->max_reply_states = env->cfg->num_queries_per_thread;
 	mesh->max_forever_states = (mesh->max_reply_states+1)/2;
 #ifndef S_SPLINT_S
@@ -424,7 +426,7 @@ void mesh_new_client(struct mesh_area* mesh, struct query_info* qinfo,
 		verbose(VERB_ALGO, "Too many queries waiting from the IP. "
 			"dropping incoming query.");
 		comm_point_drop_reply(rep);
-		mesh->stats_dropped++;
+		mesh->num_queries_wait_limit++;
 		return;
 	}
 	if(!unique)
@@ -1540,7 +1542,7 @@ void mesh_query_done(struct mesh_state* mstate)
 				http2_stream_remove_mesh_state(r->h2_stream);
 			comm_point_drop_reply(&r->query_reply);
 			mstate->reply_list = reply_list;
-			mstate->s.env->mesh->stats_dropped++;
+			mstate->s.env->mesh->num_queries_discard_timeout++;
 			continue;
 		}
 
@@ -2042,6 +2044,8 @@ mesh_stats_clear(struct mesh_area* mesh)
 	memset(&mesh->ans_rcode[0], 0, sizeof(size_t)*UB_STATS_RCODE_NUM);
 	memset(&mesh->rpz_action[0], 0, sizeof(size_t)*UB_STATS_RPZ_ACTION_NUM);
 	mesh->ans_nodata = 0;
+	mesh->num_queries_discard_timeout = 0;
+	mesh->num_queries_wait_limit = 0;
 }
 
 size_t
@@ -2270,7 +2274,7 @@ mesh_serve_expired_callback(void* arg)
 				http2_stream_remove_mesh_state(r->h2_stream);
 			comm_point_drop_reply(&r->query_reply);
 			mstate->reply_list = reply_list;
-			mstate->s.env->mesh->stats_dropped++;
+			mstate->s.env->mesh->num_queries_discard_timeout++;
 			continue;
 		}
 
diff --git a/services/mesh.h b/services/mesh.h
@@ -132,6 +132,10 @@ struct mesh_area {
 	size_t ans_nodata;
 	/** (extended stats) type of applied RPZ action */
 	size_t rpz_action[UB_STATS_RPZ_ACTION_NUM];
+	/** stats, number of queries removed due to discard-timeout */
+	size_t num_queries_discard_timeout;
+	/** stats, number of queries removed due to wait-limit */
+	size_t num_queries_wait_limit;
 
 	/** backup of query if other operations recurse and need the
 	 * network buffers */
diff --git a/smallapp/unbound-control.c b/smallapp/unbound-control.c
@@ -222,6 +222,9 @@ static void pr_stats(const char* nm, struct ub_stats_info* s)
 		s->svr.num_queries_cookie_client);
 	PR_UL_NM("num.queries_cookie_invalid",
 		s->svr.num_queries_cookie_invalid);
+	PR_UL_NM("num.queries_discard_timeout",
+		s->svr.num_queries_discard_timeout);
+	PR_UL_NM("num.queries_wait_limit", s->svr.num_queries_wait_limit);
 	PR_UL_NM("num.cachehits",
 		s->svr.num_queries - s->svr.num_queries_missed_cache);
 	PR_UL_NM("num.cachemiss", s->svr.num_queries_missed_cache);
