Add prometheus metrics

Makefile.in

@@ -187,9 +187,9 @@ unittcpreuse.lo unitdoq.lo unitinfra.lo
 UNITTEST_OBJ_LINK=$(UNITTEST_OBJ) worker_cb.lo $(COMMON_OBJ) $(SLDNS_OBJ) \
 $(COMPAT_OBJ)
 DAEMON_SRC=daemon/acl_list.c daemon/cachedump.c daemon/daemon.c \
-daemon/remote.c daemon/stats.c daemon/unbound.c daemon/worker.c @WIN_DAEMON_SRC@
+daemon/remote.c daemon/stats.c daemon/metrics.c daemon/unbound.c daemon/worker.c @WIN_DAEMON_SRC@
 DAEMON_OBJ=acl_list.lo cachedump.lo daemon.lo \
-shm_main.lo remote.lo stats.lo unbound.lo \
+shm_main.lo remote.lo stats.lo metrics.lo unbound.lo \
 worker.lo @WIN_DAEMON_OBJ@
 DAEMON_OBJ_LINK=$(DAEMON_OBJ) $(COMMON_OBJ_ALL_SYMBOLS) $(SLDNS_OBJ) \
 $(COMPAT_OBJ) @WIN_DAEMON_OBJ_LINK@
@@ -721,6 +721,7 @@ depend:
 ipset.lo ipset.o: $(srcdir)/ipset/ipset.c
 doqclient.lo doqclient.o: $(srcdir)/testcode/doqclient.c
 unitdoq.lo unitdoq.o: $(srcdir)/testcode/unitdoq.c
+metrics.lo metrics.o: $(srcdir)/daemon/metrics.c
 
 # Dependencies
 dns.lo dns.o: $(srcdir)/services/cache/dns.c config.h $(srcdir)/iterator/iter_delegpt.h $(srcdir)/util/log.h \

config.h.in

@@ -237,6 +237,9 @@
 /* Define to 1 if you have the <event.h> header file. */
 #undef HAVE_EVENT_H
 
+/* Define to 1 if you have the `evhttp_free' function. */
+#undef HAVE_EVHTTP_FREE
+
 /* Define to 1 if you have the `EVP_aes_256_cbc' function. */
 #undef HAVE_EVP_AES_256_CBC
 
@@ -986,6 +989,9 @@
 /* define this to enable debug checks. */
 #undef UNBOUND_DEBUG
 
+/* Define the default metrics HTTP endpoint port. */
+#undef UNBOUND_METRICS_PORT
+
 /* Define to 1 to use cachedb support */
 #undef USE_CACHEDB
 
@@ -1033,6 +1039,10 @@
    distributions) the use of non-ephemeral ports. */
 #undef USE_LINUX_IP_LOCAL_PORT_RANGE
 
+/* Define this to expose Unbound statistics via a prometheus metrics HTTP
+   endpoint. */
+#undef USE_METRICS
+
 /* Define if you want to use internal select based events */
 #undef USE_MINI_EVENT
 

configure

@@ -22169,6 +22169,31 @@ else $as_nop
 fi
 printf "%s\n" "#define HAVE_DECL_EVSIGNAL_ASSIGN $ac_have_decl" >>confdefs.h
 
+	# prometheus metrics depend on libevent 2.0 and later, and is therefore
+	# only enabled when the required version is found and used
+
+  for ac_func in evhttp_free
+do :
+  ac_fn_c_check_func "$LINENO" "evhttp_free" "ac_cv_func_evhttp_free"
+if test "x$ac_cv_func_evhttp_free" = xyes
+then :
+  printf "%s\n" "#define HAVE_EVHTTP_FREE 1" >>confdefs.h
+
+
+printf "%s\n" "#define USE_METRICS /**/" >>confdefs.h
+
+
+printf "%s\n" "#define UNBOUND_METRICS_PORT 9101" >>confdefs.h
+
+
+else $as_nop
+
+		{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: disabling prometheus metrics" >&5
+printf "%s\n" "$as_me: disabling prometheus metrics" >&6;}
+
+fi
+
+done
         PC_LIBEVENT_DEPENDENCY="libevent"
 
 	if test -n "$BAK_LDFLAGS_SET"; then
@@ -22178,6 +22203,8 @@ else
 
 printf "%s\n" "#define USE_MINI_EVENT 1" >>confdefs.h
 
+	{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: Prometheus metrics are disabled with the builtin libevent alternative" >&5
+printf "%s\n" "$as_me: Prometheus metrics are disabled with the builtin libevent alternative" >&6;}
 fi
 
 # check for libexpat

configure.ac

@@ -1498,13 +1498,22 @@ large outgoing port ranges.  ])
 #  include "event2/event.h"
 #endif
 	])
+	# prometheus metrics depend on libevent 2.0 and later, and is therefore
+	# only enabled when the required version is found and used
+	AC_CHECK_FUNCS([evhttp_free], [
+		AC_DEFINE_UNQUOTED([USE_METRICS], [], [Define this to expose Unbound statistics via a prometheus metrics HTTP endpoint.])
+		AC_DEFINE_UNQUOTED([UNBOUND_METRICS_PORT], [9101], [Define the default metrics HTTP endpoint port.])
+	], [
+		AC_MSG_NOTICE([disabling prometheus metrics])
+	])
         PC_LIBEVENT_DEPENDENCY="libevent"
         AC_SUBST(PC_LIBEVENT_DEPENDENCY)
 	if test -n "$BAK_LDFLAGS_SET"; then
 		LDFLAGS="$BAK_LDFLAGS"
 	fi
 else
 	AC_DEFINE(USE_MINI_EVENT, 1, [Define if you want to use internal select based events])
+	AC_MSG_NOTICE([Prometheus metrics are disabled with the builtin libevent alternative])
 fi
 
 # check for libexpat

contrib/metrics.awk

@@ -2,7 +2,7 @@
 # and output prometheus metrics style output.
 # use these options:
 #	server:		extended-statistics: yes
-#			statistics-cumulative: no
+#			statistics-cumulative: yes
 #			statistics-interval: 0
 #	remote-control: control-enable: yes
 # Can use it like unbound-control stats | awk -f "metrics.awk"
@@ -17,7 +17,7 @@ BEGIN {
 # print the output metrics
 END {
 	print "# HELP unbound_hits_queries Unbound DNS traffic and cache hits"
-	print "# TYPE unbound_hits_queries gauge"
+	print "# TYPE unbound_hits_queries counter"
 	print "unbound_hits_queries{type=\"total.num.queries\"} " val["total.num.queries"];
 	for (x=0; x<99; x++) {
 		if(val["thread" $x ".num.queries"] != "") {
@@ -70,7 +70,7 @@ END {
 	print ""
 
 	print "# HELP unbound_by_type_queries Unbound DNS queries by type"
-	print "# TYPE unbound_by_type_queries gauge"
+	print "# TYPE unbound_by_type_queries counter"
 	for(x in val) {
 		if(x ~ /^num.query.type./) {
 			if(val[x] != "") {
@@ -82,7 +82,7 @@ END {
 	print ""
 
 	print "# HELP unbound_by_class_queries Unbound DNS queries by class"
-	print "# TYPE unbound_by_class_queries gauge"
+	print "# TYPE unbound_by_class_queries counter"
 	for(x in val) {
 		if(x ~ /^num.query.class./) {
 			if(val[x] != "") {
@@ -94,7 +94,7 @@ END {
 	print ""
 
 	print "# HELP unbound_by_opcode_queries Unbound DNS queries by opcode"
-	print "# TYPE unbound_by_opcode_queries gauge"
+	print "# TYPE unbound_by_opcode_queries counter"
 	for(x in val) {
 		if(x ~ /^num.query.opcode./) {
 			if(val[x] != "") {
@@ -106,7 +106,7 @@ END {
 	print ""
 
 	print "# HELP unbound_by_rcode_queries Unbound DNS answers by rcode"
-	print "# TYPE unbound_by_rcode_queries gauge"
+	print "# TYPE unbound_by_rcode_queries counter"
 	for(x in val) {
 		if(x ~ /^num.answer.rcode./) {
 			if(val[x] != "") {
@@ -118,7 +118,7 @@ END {
 	print ""
 
 	print "# HELP unbound_by_flags_queries Unbound DNS queries by flags"
-	print "# TYPE unbound_by_flags_queries gauge"
+	print "# TYPE unbound_by_flags_queries counter"
 	for(x in val) {
 		if(x ~ /^num.query.flags./) {
 			if(val[x] != "") {
@@ -136,7 +136,7 @@ END {
 	print ""
 
 	print "# HELP unbound_histogram_seconds Unbound DNS histogram of reply time"
-	print "# TYPE unbound_histogram_seconds gauge"
+	print "# TYPE unbound_histogram_seconds counter"
 	print "unbound_histogram_seconds{bucket=\"000000.000000.to.000000.000001\"} " val["histogram.000000.000000.to.000000.000001"];
 	print "unbound_histogram_seconds{bucket=\"000000.000001.to.000000.000002\"} " val["histogram.000000.000001.to.000000.000002"];
 	print "unbound_histogram_seconds{bucket=\"000000.000002.to.000000.000004\"} " val["histogram.000000.000002.to.000000.000004"];

daemon/daemon.c

@@ -69,6 +69,7 @@
 #include "daemon/daemon.h"
 #include "daemon/worker.h"
 #include "daemon/remote.h"
+#include "daemon/metrics.h"
 #include "daemon/acl_list.h"
 #include "util/log.h"
 #include "util/config_file.h"
@@ -547,10 +548,25 @@ daemon_init(void)
 	if(gettimeofday(&daemon->time_boot, NULL) < 0)
 		log_err("gettimeofday: %s", strerror(errno));
 	daemon->time_last_stat = daemon->time_boot;
+#ifdef USE_METRICS
+	if(!(daemon->metrics = daemon_metrics_create())) {
+		acl_list_delete(daemon->acl_interface);
+		acl_list_delete(daemon->acl);
+		tcl_list_delete(daemon->tcl);
+		edns_known_options_delete(daemon->env);
+		free(daemon->env);
+		free(daemon);
+		return NULL;
+	}
+	daemon->metrics_port = -1;
+#endif /* USE_METRICS */
 	if((daemon->env->auth_zones = auth_zones_create()) == 0) {
 		acl_list_delete(daemon->acl_interface);
 		acl_list_delete(daemon->acl);
 		tcl_list_delete(daemon->tcl);
+#ifdef USE_METRICS
+		daemon_metrics_delete(daemon->metrics);
+#endif
 		edns_known_options_delete(daemon->env);
 		free(daemon->env);
 		free(daemon);
@@ -561,6 +577,9 @@ daemon_init(void)
 		acl_list_delete(daemon->acl_interface);
 		acl_list_delete(daemon->acl);
 		tcl_list_delete(daemon->tcl);
+#ifdef USE_METRICS
+		daemon_metrics_delete(daemon->metrics);
+#endif
 		edns_known_options_delete(daemon->env);
 		free(daemon->env);
 		free(daemon);
@@ -686,6 +705,19 @@ daemon_open_shared_ports(struct daemon* daemon)
 			return 0;
 		daemon->rc_port = daemon->cfg->control_port;
 	}
+#ifdef USE_METRICS
+	if(!daemon->cfg->metrics_enable && daemon->metrics_port != -1) {
+		daemon_metrics_close_ports(daemon->metrics);
+		daemon->metrics_port = -1;
+	}
+	if(daemon->cfg->metrics_enable &&
+		daemon->cfg->metrics_port != daemon->metrics_port) {
+		daemon_metrics_close_ports(daemon->metrics);
+		if(!daemon_metrics_open_ports(daemon->metrics, daemon->cfg))
+			return 0;
+		daemon->metrics_port = daemon->cfg->metrics_port;
+	}
+#endif /* USE_METRICS */
 	return 1;
 }
 
@@ -1171,6 +1203,9 @@ daemon_cleanup(struct daemon* daemon)
 		auth_zones_cleanup(daemon->env->auth_zones);
 	/* key cache is cleared by module deinit during next daemon_fork() */
 	daemon_remote_clear(daemon->rc);
+#ifdef USE_METRICS
+	daemon_metrics_detach(daemon->metrics);
+#endif
 	if(daemon->fast_reload_thread)
 		fast_reload_thread_stop(daemon->fast_reload_thread);
 	if(daemon->fast_reload_printq_list)
@@ -1213,6 +1248,9 @@ daemon_delete(struct daemon* daemon)
 	modstack_call_destartup(&daemon->mods, daemon->env);
 	modstack_free(&daemon->mods);
 	daemon_remote_delete(daemon->rc);
+#ifdef USE_METRICS
+	daemon_metrics_delete(daemon->metrics);
+#endif
 	for(i = 0; i < daemon->num_ports; i++)
 		listening_ports_free(daemon->ports[i]);
 	free(daemon->ports);

daemon/daemon.h

@@ -56,6 +56,7 @@ struct local_zones;
 struct views;
 struct ub_randstate;
 struct daemon_remote;
+struct daemon_metrics;
 struct respip_set;
 struct shm_main_info;
 struct doq_table;
@@ -99,6 +100,10 @@ struct daemon {
 	struct listen_port* rc_ports;
 	/** remote control connections management (for first worker) */
 	struct daemon_remote* rc;
+	/** port number for metrics that has ports opened. */
+	int metrics_port;
+	/** metrics endpoint connections management (for first worker) */
+	struct daemon_metrics* metrics;
 	/** ssl context for listening to dnstcp over ssl */
 	void* listen_dot_sslctx;
 	/** ssl context for connecting to dnstcp over ssl */