2 * collectd - src/statsd.c
3 * Copyright (C) 2013 Florian octo Forster
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
24 * Florian octo Forster <octo at collectd.org>
30 #include "configfile.h"
31 #include "utils_avltree.h"
32 #include "utils_latency.h"
34 #include <sys/types.h>
38 /* AIX doesn't have MSG_DONTWAIT */
40 # define MSG_DONTWAIT MSG_NONBLOCK
43 #ifndef STATSD_DEFAULT_NODE
44 # define STATSD_DEFAULT_NODE NULL
47 #ifndef STATSD_DEFAULT_SERVICE
48 # define STATSD_DEFAULT_SERVICE "8125"
58 typedef enum metric_type_e metric_type_t;
60 struct statsd_metric_s
65 latency_counter_t *latency;
67 unsigned long updates_num;
69 typedef struct statsd_metric_s statsd_metric_t;
71 static c_avl_tree_t *metrics_tree = NULL;
72 static pthread_mutex_t metrics_lock = PTHREAD_MUTEX_INITIALIZER;
74 static pthread_t network_thread;
75 static _Bool network_thread_running = 0;
76 static _Bool network_thread_shutdown = 0;
78 static char *conf_node = NULL;
79 static char *conf_service = NULL;
81 static _Bool conf_delete_counters = 0;
82 static _Bool conf_delete_timers = 0;
83 static _Bool conf_delete_gauges = 0;
84 static _Bool conf_delete_sets = 0;
86 static double *conf_timer_percentile = NULL;
87 static size_t conf_timer_percentile_num = 0;
89 static _Bool conf_counter_sum = 0;
90 static _Bool conf_timer_lower = 0;
91 static _Bool conf_timer_upper = 0;
92 static _Bool conf_timer_sum = 0;
93 static _Bool conf_timer_count = 0;
95 /* Must hold metrics_lock when calling this function. */
96 static statsd_metric_t *statsd_metric_lookup_unsafe (char const *name, /* {{{ */
99 char key[DATA_MAX_NAME_LEN + 2];
101 statsd_metric_t *metric;
106 case STATSD_COUNTER: key[0] = 'c'; break;
107 case STATSD_TIMER: key[0] = 't'; break;
108 case STATSD_GAUGE: key[0] = 'g'; break;
109 case STATSD_SET: key[0] = 's'; break;
110 default: return (NULL);
114 sstrncpy (&key[2], name, sizeof (key) - 2);
116 status = c_avl_get (metrics_tree, key, (void *) &metric);
120 key_copy = strdup (key);
121 if (key_copy == NULL)
123 ERROR ("statsd plugin: strdup failed.");
127 metric = calloc (1, sizeof (*metric));
130 ERROR ("statsd plugin: calloc failed.");
136 metric->latency = NULL;
139 status = c_avl_insert (metrics_tree, key_copy, metric);
142 ERROR ("statsd plugin: c_avl_insert failed.");
149 } /* }}} statsd_metric_lookup_unsafe */
151 static int statsd_metric_set (char const *name, double value, /* {{{ */
154 statsd_metric_t *metric;
156 pthread_mutex_lock (&metrics_lock);
158 metric = statsd_metric_lookup_unsafe (name, type);
161 pthread_mutex_unlock (&metrics_lock);
165 metric->value = value;
166 metric->updates_num++;
168 pthread_mutex_unlock (&metrics_lock);
171 } /* }}} int statsd_metric_set */
173 static int statsd_metric_add (char const *name, double delta, /* {{{ */
176 statsd_metric_t *metric;
178 pthread_mutex_lock (&metrics_lock);
180 metric = statsd_metric_lookup_unsafe (name, type);
183 pthread_mutex_unlock (&metrics_lock);
187 metric->value += delta;
188 metric->updates_num++;
190 pthread_mutex_unlock (&metrics_lock);
193 } /* }}} int statsd_metric_add */
195 static void statsd_metric_free (statsd_metric_t *metric) /* {{{ */
200 if (metric->latency != NULL)
202 latency_counter_destroy (metric->latency);
203 metric->latency = NULL;
206 if (metric->set != NULL)
211 while (c_avl_pick (metric->set, &key, &value) == 0)
214 assert (value == NULL);
217 c_avl_destroy (metric->set);
222 } /* }}} void statsd_metric_free */
224 static int statsd_parse_value (char const *str, value_t *ret_value) /* {{{ */
228 ret_value->gauge = (gauge_t) strtod (str, &endptr);
229 if ((str == endptr) || ((endptr != NULL) && (*endptr != 0)))
233 } /* }}} int statsd_parse_value */
235 static int statsd_handle_counter (char const *name, /* {{{ */
236 char const *value_str,
243 if ((extra != NULL) && (extra[0] != '@'))
249 status = statsd_parse_value (extra + 1, &scale);
253 if (!isfinite (scale.gauge) || (scale.gauge <= 0.0) || (scale.gauge > 1.0))
258 status = statsd_parse_value (value_str, &value);
262 /* Changes to the counter are added to (statsd_metric_t*)->value. ->counter is
263 * only updated in statsd_metric_submit_unsafe(). */
264 return (statsd_metric_add (name, (double) (value.gauge / scale.gauge),
266 } /* }}} int statsd_handle_counter */
268 static int statsd_handle_gauge (char const *name, /* {{{ */
269 char const *value_str)
275 status = statsd_parse_value (value_str, &value);
279 if ((value_str[0] == '+') || (value_str[0] == '-'))
280 return (statsd_metric_add (name, (double) value.gauge, STATSD_GAUGE));
282 return (statsd_metric_set (name, (double) value.gauge, STATSD_GAUGE));
283 } /* }}} int statsd_handle_gauge */
285 static int statsd_handle_timer (char const *name, /* {{{ */
286 char const *value_str,
289 statsd_metric_t *metric;
295 if ((extra != NULL) && (extra[0] != '@'))
301 status = statsd_parse_value (extra + 1, &scale);
305 if (!isfinite (scale.gauge) || (scale.gauge <= 0.0) || (scale.gauge > 1.0))
310 status = statsd_parse_value (value_str, &value_ms);
314 value = MS_TO_CDTIME_T (value_ms.gauge / scale.gauge);
316 pthread_mutex_lock (&metrics_lock);
318 metric = statsd_metric_lookup_unsafe (name, STATSD_TIMER);
321 pthread_mutex_unlock (&metrics_lock);
325 if (metric->latency == NULL)
326 metric->latency = latency_counter_create ();
327 if (metric->latency == NULL)
329 pthread_mutex_unlock (&metrics_lock);
333 latency_counter_add (metric->latency, value);
334 metric->updates_num++;
336 pthread_mutex_unlock (&metrics_lock);
338 } /* }}} int statsd_handle_timer */
340 static int statsd_handle_set (char const *name, /* {{{ */
341 char const *set_key_orig)
343 statsd_metric_t *metric = NULL;
347 pthread_mutex_lock (&metrics_lock);
349 metric = statsd_metric_lookup_unsafe (name, STATSD_SET);
352 pthread_mutex_unlock (&metrics_lock);
356 /* Make sure metric->set exists. */
357 if (metric->set == NULL)
358 metric->set = c_avl_create ((int (*) (const void *, const void *)) strcmp);
360 if (metric->set == NULL)
362 pthread_mutex_unlock (&metrics_lock);
363 ERROR ("statsd plugin: c_avl_create failed.");
367 set_key = strdup (set_key_orig);
370 pthread_mutex_unlock (&metrics_lock);
371 ERROR ("statsd plugin: strdup failed.");
375 status = c_avl_insert (metric->set, set_key, /* value = */ NULL);
378 pthread_mutex_unlock (&metrics_lock);
380 ERROR ("statsd plugin: c_avl_insert (\"%s\") failed with status %i.",
385 else if (status > 0) /* key already exists */
390 metric->updates_num++;
392 pthread_mutex_unlock (&metrics_lock);
394 } /* }}} int statsd_handle_set */
396 static int statsd_parse_line (char *buffer) /* {{{ */
403 type = strchr (name, '|');
409 value = strrchr (name, ':');
415 extra = strchr (type, '|');
422 if (strcmp ("c", type) == 0)
423 return (statsd_handle_counter (name, value, extra));
424 else if (strcmp ("ms", type) == 0)
425 return (statsd_handle_timer (name, value, extra));
427 /* extra is only valid for counters and timers */
431 if (strcmp ("g", type) == 0)
432 return (statsd_handle_gauge (name, value));
433 else if (strcmp ("s", type) == 0)
434 return (statsd_handle_set (name, value));
437 } /* }}} void statsd_parse_line */
439 static void statsd_parse_buffer (char *buffer) /* {{{ */
441 while (buffer != NULL)
447 next = strchr (buffer, '\n');
460 sstrncpy (orig, buffer, sizeof (orig));
462 status = statsd_parse_line (buffer);
464 ERROR ("statsd plugin: Unable to parse line: \"%s\"", orig);
468 } /* }}} void statsd_parse_buffer */
470 static void statsd_network_read (int fd) /* {{{ */
476 status = recv (fd, buffer, sizeof (buffer), /* flags = */ MSG_DONTWAIT);
481 if ((errno == EAGAIN) || (errno == EWOULDBLOCK))
484 ERROR ("statsd plugin: recv(2) failed: %s",
485 sstrerror (errno, errbuf, sizeof (errbuf)));
489 buffer_size = (size_t) status;
490 if (buffer_size >= sizeof (buffer))
491 buffer_size = sizeof (buffer) - 1;
492 buffer[buffer_size] = 0;
494 statsd_parse_buffer (buffer);
495 } /* }}} void statsd_network_read */
497 static int statsd_network_init (struct pollfd **ret_fds, /* {{{ */
500 struct pollfd *fds = NULL;
503 struct addrinfo *ai_list;
504 struct addrinfo *ai_ptr;
507 char const *node = (conf_node != NULL) ? conf_node : STATSD_DEFAULT_NODE;
508 char const *service = (conf_service != NULL)
509 ? conf_service : STATSD_DEFAULT_SERVICE;
511 struct addrinfo ai_hints = {
512 .ai_family = AF_UNSPEC,
513 .ai_flags = AI_PASSIVE | AI_ADDRCONFIG,
514 .ai_socktype = SOCK_DGRAM
517 status = getaddrinfo (node, service, &ai_hints, &ai_list);
520 ERROR ("statsd plugin: getaddrinfo (\"%s\", \"%s\") failed: %s",
521 node, service, gai_strerror (status));
525 for (ai_ptr = ai_list; ai_ptr != NULL; ai_ptr = ai_ptr->ai_next)
530 char dbg_node[NI_MAXHOST];
531 char dbg_service[NI_MAXSERV];
533 fd = socket (ai_ptr->ai_family, ai_ptr->ai_socktype, ai_ptr->ai_protocol);
537 ERROR ("statsd plugin: socket(2) failed: %s",
538 sstrerror (errno, errbuf, sizeof (errbuf)));
542 getnameinfo (ai_ptr->ai_addr, ai_ptr->ai_addrlen,
543 dbg_node, sizeof (dbg_node), dbg_service, sizeof (dbg_service),
544 NI_DGRAM | NI_NUMERICHOST | NI_NUMERICSERV);
545 DEBUG ("statsd plugin: Trying to bind to [%s]:%s ...", dbg_node, dbg_service);
547 status = bind (fd, ai_ptr->ai_addr, ai_ptr->ai_addrlen);
551 ERROR ("statsd plugin: bind(2) failed: %s",
552 sstrerror (errno, errbuf, sizeof (errbuf)));
557 tmp = realloc (fds, sizeof (*fds) * (fds_num + 1));
560 ERROR ("statsd plugin: realloc failed.");
568 memset (tmp, 0, sizeof (*tmp));
570 tmp->events = POLLIN | POLLPRI;
573 freeaddrinfo (ai_list);
577 ERROR ("statsd plugin: Unable to create listening socket for [%s]:%s.",
578 (node != NULL) ? node : "::", service);
583 *ret_fds_num = fds_num;
585 } /* }}} int statsd_network_init */
587 static void *statsd_network_thread (void *args) /* {{{ */
589 struct pollfd *fds = NULL;
594 status = statsd_network_init (&fds, &fds_num);
597 ERROR ("statsd plugin: Unable to open listening sockets.");
598 pthread_exit ((void *) 0);
601 while (!network_thread_shutdown)
603 status = poll (fds, (nfds_t) fds_num, /* timeout = */ -1);
608 if ((errno == EINTR) || (errno == EAGAIN))
611 ERROR ("statsd plugin: poll(2) failed: %s",
612 sstrerror (errno, errbuf, sizeof (errbuf)));
616 for (i = 0; i < fds_num; i++)
618 if ((fds[i].revents & (POLLIN | POLLPRI)) == 0)
621 statsd_network_read (fds[i].fd);
624 } /* while (!network_thread_shutdown) */
627 for (i = 0; i < fds_num; i++)
632 } /* }}} void *statsd_network_thread */
634 static int statsd_config_timer_percentile (oconfig_item_t *ci) /* {{{ */
636 double percent = NAN;
640 status = cf_util_get_double (ci, &percent);
644 if ((percent <= 0.0) || (percent >= 100))
646 ERROR ("statsd plugin: The value for \"%s\" must be between 0 and 100, "
647 "exclusively.", ci->key);
651 tmp = realloc (conf_timer_percentile,
652 sizeof (*conf_timer_percentile) * (conf_timer_percentile_num + 1));
655 ERROR ("statsd plugin: realloc failed.");
658 conf_timer_percentile = tmp;
659 conf_timer_percentile[conf_timer_percentile_num] = percent;
660 conf_timer_percentile_num++;
663 } /* }}} int statsd_config_timer_percentile */
665 static int statsd_config (oconfig_item_t *ci) /* {{{ */
669 for (i = 0; i < ci->children_num; i++)
671 oconfig_item_t *child = ci->children + i;
673 if (strcasecmp ("Host", child->key) == 0)
674 cf_util_get_string (child, &conf_node);
675 else if (strcasecmp ("Port", child->key) == 0)
676 cf_util_get_service (child, &conf_service);
677 else if (strcasecmp ("DeleteCounters", child->key) == 0)
678 cf_util_get_boolean (child, &conf_delete_counters);
679 else if (strcasecmp ("DeleteTimers", child->key) == 0)
680 cf_util_get_boolean (child, &conf_delete_timers);
681 else if (strcasecmp ("DeleteGauges", child->key) == 0)
682 cf_util_get_boolean (child, &conf_delete_gauges);
683 else if (strcasecmp ("DeleteSets", child->key) == 0)
684 cf_util_get_boolean (child, &conf_delete_sets);
685 else if (strcasecmp ("CounterSum", child->key) == 0)
686 cf_util_get_boolean (child, &conf_counter_sum);
687 else if (strcasecmp ("TimerLower", child->key) == 0)
688 cf_util_get_boolean (child, &conf_timer_lower);
689 else if (strcasecmp ("TimerUpper", child->key) == 0)
690 cf_util_get_boolean (child, &conf_timer_upper);
691 else if (strcasecmp ("TimerSum", child->key) == 0)
692 cf_util_get_boolean (child, &conf_timer_sum);
693 else if (strcasecmp ("TimerCount", child->key) == 0)
694 cf_util_get_boolean (child, &conf_timer_count);
695 else if (strcasecmp ("TimerPercentile", child->key) == 0)
696 statsd_config_timer_percentile (child);
698 ERROR ("statsd plugin: The \"%s\" config option is not valid.",
703 } /* }}} int statsd_config */
705 static int statsd_init (void) /* {{{ */
707 pthread_mutex_lock (&metrics_lock);
708 if (metrics_tree == NULL)
709 metrics_tree = c_avl_create ((int (*) (const void *, const void *)) strcmp);
711 if (!network_thread_running)
715 status = pthread_create (&network_thread,
717 statsd_network_thread,
722 pthread_mutex_unlock (&metrics_lock);
723 ERROR ("statsd plugin: pthread_create failed: %s",
724 sstrerror (errno, errbuf, sizeof (errbuf)));
728 network_thread_running = 1;
730 pthread_mutex_unlock (&metrics_lock);
733 } /* }}} int statsd_init */
735 /* Must hold metrics_lock when calling this function. */
736 static int statsd_metric_clear_set_unsafe (statsd_metric_t *metric) /* {{{ */
741 if ((metric == NULL) || (metric->type != STATSD_SET))
744 if (metric->set == NULL)
747 while (c_avl_pick (metric->set, &key, &value) == 0)
754 } /* }}} int statsd_metric_clear_set_unsafe */
756 /* Must hold metrics_lock when calling this function. */
757 static int statsd_metric_submit_unsafe (char const *name, statsd_metric_t *metric) /* {{{ */
760 value_list_t vl = VALUE_LIST_INIT;
764 sstrncpy (vl.host, hostname_g, sizeof (vl.host));
765 sstrncpy (vl.plugin, "statsd", sizeof (vl.plugin));
767 if (metric->type == STATSD_GAUGE)
768 sstrncpy (vl.type, "gauge", sizeof (vl.type));
769 else if (metric->type == STATSD_TIMER)
770 sstrncpy (vl.type, "latency", sizeof (vl.type));
771 else if (metric->type == STATSD_SET)
772 sstrncpy (vl.type, "objects", sizeof (vl.type));
773 else /* if (metric->type == STATSD_COUNTER) */
774 sstrncpy (vl.type, "derive", sizeof (vl.type));
776 sstrncpy (vl.type_instance, name, sizeof (vl.type_instance));
778 if (metric->type == STATSD_GAUGE)
779 values[0].gauge = (gauge_t) metric->value;
780 else if (metric->type == STATSD_TIMER)
783 _Bool have_events = (metric->updates_num > 0);
785 /* Make sure all timer metrics share the *same* timestamp. */
788 ssnprintf (vl.type_instance, sizeof (vl.type_instance),
790 values[0].gauge = have_events
791 ? CDTIME_T_TO_DOUBLE (latency_counter_get_average (metric->latency))
793 plugin_dispatch_values (&vl);
795 if (conf_timer_lower) {
796 ssnprintf (vl.type_instance, sizeof (vl.type_instance),
798 values[0].gauge = have_events
799 ? CDTIME_T_TO_DOUBLE (latency_counter_get_min (metric->latency))
801 plugin_dispatch_values (&vl);
804 if (conf_timer_upper) {
805 ssnprintf (vl.type_instance, sizeof (vl.type_instance),
807 values[0].gauge = have_events
808 ? CDTIME_T_TO_DOUBLE (latency_counter_get_max (metric->latency))
810 plugin_dispatch_values (&vl);
813 if (conf_timer_sum) {
814 ssnprintf (vl.type_instance, sizeof (vl.type_instance),
816 values[0].gauge = have_events
817 ? CDTIME_T_TO_DOUBLE (latency_counter_get_sum (metric->latency))
819 plugin_dispatch_values (&vl);
822 for (i = 0; i < conf_timer_percentile_num; i++)
824 ssnprintf (vl.type_instance, sizeof (vl.type_instance),
825 "%s-percentile-%.0f", name, conf_timer_percentile[i]);
826 values[0].gauge = have_events
827 ? CDTIME_T_TO_DOUBLE (latency_counter_get_percentile (metric->latency, conf_timer_percentile[i]))
829 plugin_dispatch_values (&vl);
832 /* Keep this at the end, since vl.type is set to "gauge" here. The
833 * vl.type's above are implicitly set to "latency". */
834 if (conf_timer_count) {
835 sstrncpy (vl.type, "gauge", sizeof (vl.type));
836 ssnprintf (vl.type_instance, sizeof (vl.type_instance),
838 values[0].gauge = latency_counter_get_num (metric->latency);
839 plugin_dispatch_values (&vl);
842 latency_counter_reset (metric->latency);
845 else if (metric->type == STATSD_SET)
847 if (metric->set == NULL)
848 values[0].gauge = 0.0;
850 values[0].gauge = (gauge_t) c_avl_size (metric->set);
852 else { /* STATSD_COUNTER */
853 gauge_t delta = nearbyint (metric->value);
855 /* Etsy's statsd writes counters as two metrics: a rate and the change since
856 * the last write. Since collectd does not reset its DERIVE metrics to zero,
857 * this makes little sense, but we're dispatching a "count" metric here
858 * anyway - if requested by the user - for compatibility reasons. */
859 if (conf_counter_sum)
861 sstrncpy (vl.type, "count", sizeof (vl.type));
862 values[0].gauge = delta;
863 plugin_dispatch_values (&vl);
865 /* restore vl.type */
866 sstrncpy (vl.type, "derive", sizeof (vl.type));
869 /* Rather than resetting value to zero, subtract delta so we correctly keep
870 * track of residuals. */
871 metric->value -= delta;
872 metric->counter += (derive_t) delta;
874 values[0].derive = metric->counter;
877 return (plugin_dispatch_values (&vl));
878 } /* }}} int statsd_metric_submit_unsafe */
880 static int statsd_read (void) /* {{{ */
882 c_avl_iterator_t *iter;
884 statsd_metric_t *metric;
886 char **to_be_deleted = NULL;
887 size_t to_be_deleted_num = 0;
890 pthread_mutex_lock (&metrics_lock);
892 if (metrics_tree == NULL)
894 pthread_mutex_unlock (&metrics_lock);
898 iter = c_avl_get_iterator (metrics_tree);
899 while (c_avl_iterator_next (iter, (void *) &name, (void *) &metric) == 0)
901 if ((metric->updates_num == 0)
902 && ((conf_delete_counters && (metric->type == STATSD_COUNTER))
903 || (conf_delete_timers && (metric->type == STATSD_TIMER))
904 || (conf_delete_gauges && (metric->type == STATSD_GAUGE))
905 || (conf_delete_sets && (metric->type == STATSD_SET))))
907 DEBUG ("statsd plugin: Deleting metric \"%s\".", name);
908 strarray_add (&to_be_deleted, &to_be_deleted_num, name);
912 /* Names have a prefix, e.g. "c:", which determines the (statsd) type.
913 * Remove this here. */
914 statsd_metric_submit_unsafe (name + 2, metric);
916 /* Reset the metric. */
917 metric->updates_num = 0;
918 if (metric->type == STATSD_SET)
919 statsd_metric_clear_set_unsafe (metric);
921 c_avl_iterator_destroy (iter);
923 for (i = 0; i < to_be_deleted_num; i++)
927 status = c_avl_remove (metrics_tree, to_be_deleted[i],
928 (void *) &name, (void *) &metric);
931 ERROR ("stats plugin: c_avl_remove (\"%s\") failed with status %i.",
932 to_be_deleted[i], status);
937 statsd_metric_free (metric);
940 pthread_mutex_unlock (&metrics_lock);
942 strarray_free (to_be_deleted, to_be_deleted_num);
945 } /* }}} int statsd_read */
947 static int statsd_shutdown (void) /* {{{ */
952 if (network_thread_running)
954 network_thread_shutdown = 1;
955 pthread_kill (network_thread, SIGTERM);
956 pthread_join (network_thread, /* retval = */ NULL);
958 network_thread_running = 0;
960 pthread_mutex_lock (&metrics_lock);
962 while (c_avl_pick (metrics_tree, &key, &value) == 0)
965 statsd_metric_free (value);
967 c_avl_destroy (metrics_tree);
971 sfree (conf_service);
973 pthread_mutex_unlock (&metrics_lock);
976 } /* }}} int statsd_shutdown */
978 void module_register (void)
980 plugin_register_complex_config ("statsd", statsd_config);
981 plugin_register_init ("statsd", statsd_init);
982 plugin_register_read ("statsd", statsd_read);
983 plugin_register_shutdown ("statsd", statsd_shutdown);
986 /* vim: set sw=2 sts=2 et fdm=marker : */