X-Git-Url: https://git.octo.it/?p=collectd.git;a=blobdiff_plain;f=src%2Fwrite_tsdb.c;h=f8f4cb9104efb1753a875402510ecabc0eb1adc3;hp=9cb3c97de611323b3c731ba7f5bef3883cd2c866;hb=54619dc85fd308b21ed09a0271e5c7383c7921b9;hpb=35a6c9c5fcd87cb78451a974c4d5b5707926845c

diff --git a/src/write_tsdb.c b/src/write_tsdb.c
index 9cb3c97d..f8f4cb91 100644
--- a/src/write_tsdb.c
+++ b/src/write_tsdb.c
@@ -43,10 +43,10 @@
 
 #include "collectd.h"
 
-#include "common.h"
 #include "plugin.h"
-
+#include "utils/common/common.h"
 #include "utils_cache.h"
+#include "utils_random.h"
 
 #include <netdb.h>
 
@@ -71,14 +71,16 @@
  * Private variables
  */
 struct wt_callback {
+  struct addrinfo *ai;
+  cdtime_t ai_last_update;
   int sock_fd;
 
   char *node;
   char *service;
   char *host_tags;
 
-  _Bool store_rates;
-  _Bool always_append_ds;
+  bool store_rates;
+  bool always_append_ds;
 
   char send_buf[WT_SEND_BUF_SIZE];
   size_t send_buf_free;
@@ -86,8 +88,15 @@ struct wt_callback {
   cdtime_t send_buf_init_time;
 
   pthread_mutex_t send_lock;
+
+  bool connect_failed_log_enabled;
+  int connect_dns_failed_attempts_remaining;
+  cdtime_t next_random_ttl;
 };
 
+static cdtime_t resolve_interval;
+static cdtime_t resolve_jitter;
+
 /*
  * Functions
  */
@@ -102,10 +111,9 @@ static int wt_send_buffer(struct wt_callback *cb) {
   ssize_t status = 0;
 
   status = swrite(cb->sock_fd, cb->send_buf, strlen(cb->send_buf));
-  if (status < 0) {
-    char errbuf[1024];
+  if (status != 0) {
     ERROR("write_tsdb plugin: send failed with status %zi (%s)", status,
-          sstrerror(errno, errbuf, sizeof(errbuf)));
+          STRERRNO);
 
     close(cb->sock_fd);
     cb->sock_fd = -1;
@@ -121,7 +129,7 @@ static int wt_flush_nolock(cdtime_t timeout, struct wt_callback *cb) {
   int status;
 
   DEBUG("write_tsdb plugin: wt_flush_nolock: timeout = %.3f; "
-        "send_buf_fill = %zu;",
+        "send_buf_fill = %" PRIsz ";",
         (double)timeout, cb->send_buf_fill);
 
   /* timeout == 0  => flush unconditionally */
@@ -144,9 +152,16 @@ static int wt_flush_nolock(cdtime_t timeout, struct wt_callback *cb) {
   return status;
 }
 
+static cdtime_t new_random_ttl(void) {
+  if (resolve_jitter == 0)
+    return 0;
+
+  return (cdtime_t)cdrand_range(0, (long)resolve_jitter);
+}
+
 static int wt_callback_init(struct wt_callback *cb) {
-  struct addrinfo *ai_list;
   int status;
+  cdtime_t now;
 
   const char *node = cb->node ? cb->node : WT_DEFAULT_NODE;
   const char *service = cb->service ? cb->service : WT_DEFAULT_SERVICE;
@@ -154,28 +169,68 @@ static int wt_callback_init(struct wt_callback *cb) {
   if (cb->sock_fd > 0)
     return 0;
 
-  struct addrinfo ai_hints = {.ai_family = AF_UNSPEC,
-                              .ai_flags = AI_ADDRCONFIG,
-                              .ai_socktype = SOCK_STREAM};
+  now = cdtime();
+  if (cb->ai) {
+    /* When we are here, we still have the IP in cache.
+     * If we have remaining attempts without calling the DNS, we update the
+     * last_update date so we keep the info until next time.
+     * If there is no more attempts, we need to flush the cache.
+     */
+
+    if ((cb->ai_last_update + resolve_interval + cb->next_random_ttl) < now) {
+      cb->next_random_ttl = new_random_ttl();
+      if (cb->connect_dns_failed_attempts_remaining > 0) {
+        /* Warning : this is run under send_lock mutex.
+         * This is why we do not use another mutex here.
+         * */
+        cb->ai_last_update = now;
+        cb->connect_dns_failed_attempts_remaining--;
+      } else {
+        freeaddrinfo(cb->ai);
+        cb->ai = NULL;
+      }
+    }
+  }
 
-  status = getaddrinfo(node, service, &ai_hints, &ai_list);
-  if (status != 0) {
-    ERROR("write_tsdb plugin: getaddrinfo (%s, %s) failed: %s", node, service,
-          gai_strerror(status));
-    return -1;
+  if (cb->ai == NULL) {
+    if ((cb->ai_last_update + resolve_interval + cb->next_random_ttl) >= now) {
+      DEBUG("write_tsdb plugin: too many getaddrinfo(%s, %s) failures", node,
+            service);
+      return -1;
+    }
+    cb->ai_last_update = now;
+    cb->next_random_ttl = new_random_ttl();
+
+    struct addrinfo ai_hints = {
+        .ai_family = AF_UNSPEC,
+        .ai_flags = AI_ADDRCONFIG,
+        .ai_socktype = SOCK_STREAM,
+    };
+
+    status = getaddrinfo(node, service, &ai_hints, &cb->ai);
+    if (status != 0) {
+      if (cb->ai) {
+        freeaddrinfo(cb->ai);
+        cb->ai = NULL;
+      }
+      if (cb->connect_failed_log_enabled) {
+        ERROR("write_tsdb plugin: getaddrinfo(%s, %s) failed: %s", node,
+              service, gai_strerror(status));
+        cb->connect_failed_log_enabled = 0;
+      }
+      return -1;
+    }
   }
 
-  assert(ai_list != NULL);
-  for (struct addrinfo *ai_ptr = ai_list; ai_ptr != NULL;
-       ai_ptr = ai_ptr->ai_next) {
-    cb->sock_fd =
-        socket(ai_ptr->ai_family, ai_ptr->ai_socktype, ai_ptr->ai_protocol);
+  assert(cb->ai != NULL);
+  for (struct addrinfo *ai = cb->ai; ai != NULL; ai = ai->ai_next) {
+    cb->sock_fd = socket(ai->ai_family, ai->ai_socktype, ai->ai_protocol);
     if (cb->sock_fd < 0)
       continue;
 
     set_sock_opts(cb->sock_fd);
 
-    status = connect(cb->sock_fd, ai_ptr->ai_addr, ai_ptr->ai_addrlen);
+    status = connect(cb->sock_fd, ai->ai_addr, ai->ai_addrlen);
     if (status != 0) {
       close(cb->sock_fd);
       cb->sock_fd = -1;
@@ -185,16 +240,19 @@ static int wt_callback_init(struct wt_callback *cb) {
     break;
   }
 
-  freeaddrinfo(ai_list);
-
   if (cb->sock_fd < 0) {
-    char errbuf[1024];
     ERROR("write_tsdb plugin: Connecting to %s:%s failed. "
           "The last error was: %s",
-          node, service, sstrerror(errno, errbuf, sizeof(errbuf)));
+          node, service, STRERRNO);
     return -1;
   }
 
+  if (0 == cb->connect_failed_log_enabled) {
+    WARNING("write_tsdb plugin: Connecting to %s:%s succeeded.", node, service);
+    cb->connect_failed_log_enabled = 1;
+  }
+  cb->connect_dns_failed_attempts_remaining = 1;
+
   wt_reset_buffer(cb);
 
   return 0;
@@ -255,7 +313,7 @@ static int wt_flush(cdtime_t timeout,
 
 static int wt_format_values(char *ret, size_t ret_len, int ds_num,
                             const data_set_t *ds, const value_list_t *vl,
-                            _Bool store_rates) {
+                            bool store_rates) {
   size_t offset = 0;
   int status;
   gauge_t *rates = NULL;
@@ -266,7 +324,7 @@ static int wt_format_values(char *ret, size_t ret_len, int ds_num,
 
 #define BUFFER_ADD(...)                                                        \
   do {                                                                         \
-    status = ssnprintf(ret + offset, ret_len - offset, __VA_ARGS__);           \
+    status = snprintf(ret + offset, ret_len - offset, __VA_ARGS__);            \
     if (status < 1) {                                                          \
       sfree(rates);                                                            \
       return -1;                                                               \
@@ -289,7 +347,7 @@ static int wt_format_values(char *ret, size_t ret_len, int ds_num,
     }
     BUFFER_ADD(GAUGE_FORMAT, rates[ds_num]);
   } else if (ds->ds[ds_num].type == DS_TYPE_COUNTER)
-    BUFFER_ADD("%llu", vl->values[ds_num].counter);
+    BUFFER_ADD("%" PRIu64, (uint64_t)vl->values[ds_num].counter);
   else if (ds->ds[ds_num].type == DS_TYPE_DERIVE)
     BUFFER_ADD("%" PRIi64, vl->values[ds_num].derive);
   else if (ds->ds[ds_num].type == DS_TYPE_ABSOLUTE)
@@ -329,36 +387,36 @@ static int wt_format_name(char *ret, int ret_len, const value_list_t *vl,
   if (ds_name != NULL) {
     if (vl->plugin_instance[0] == '\0') {
       if (vl->type_instance[0] == '\0') {
-        ssnprintf(ret, ret_len, "%s%s.%s.%s", prefix, vl->plugin, vl->type,
-                  ds_name);
+        snprintf(ret, ret_len, "%s%s.%s.%s", prefix, vl->plugin, vl->type,
+                 ds_name);
       } else {
-        ssnprintf(ret, ret_len, "%s%s.%s.%s.%s", prefix, vl->plugin, vl->type,
-                  vl->type_instance, ds_name);
+        snprintf(ret, ret_len, "%s%s.%s.%s.%s", prefix, vl->plugin, vl->type,
+                 vl->type_instance, ds_name);
       }
     } else { /* vl->plugin_instance != "" */
       if (vl->type_instance[0] == '\0') {
-        ssnprintf(ret, ret_len, "%s%s.%s.%s.%s", prefix, vl->plugin,
-                  vl->plugin_instance, vl->type, ds_name);
+        snprintf(ret, ret_len, "%s%s.%s.%s.%s", prefix, vl->plugin,
+                 vl->plugin_instance, vl->type, ds_name);
       } else {
-        ssnprintf(ret, ret_len, "%s%s.%s.%s.%s.%s", prefix, vl->plugin,
-                  vl->plugin_instance, vl->type, vl->type_instance, ds_name);
+        snprintf(ret, ret_len, "%s%s.%s.%s.%s.%s", prefix, vl->plugin,
+                 vl->plugin_instance, vl->type, vl->type_instance, ds_name);
       }
     }
   } else { /* ds_name == NULL */
     if (vl->plugin_instance[0] == '\0') {
       if (vl->type_instance[0] == '\0') {
-        ssnprintf(ret, ret_len, "%s%s.%s", prefix, vl->plugin, vl->type);
+        snprintf(ret, ret_len, "%s%s.%s", prefix, vl->plugin, vl->type);
       } else {
-        ssnprintf(ret, ret_len, "%s%s.%s.%s", prefix, vl->plugin,
-                  vl->type_instance, vl->type);
+        snprintf(ret, ret_len, "%s%s.%s.%s", prefix, vl->plugin,
+                 vl->type_instance, vl->type);
       }
     } else { /* vl->plugin_instance != "" */
       if (vl->type_instance[0] == '\0') {
-        ssnprintf(ret, ret_len, "%s%s.%s.%s", prefix, vl->plugin,
-                  vl->plugin_instance, vl->type);
+        snprintf(ret, ret_len, "%s%s.%s.%s", prefix, vl->plugin,
+                 vl->plugin_instance, vl->type);
       } else {
-        ssnprintf(ret, ret_len, "%s%s.%s.%s.%s", prefix, vl->plugin,
-                  vl->plugin_instance, vl->type, vl->type_instance);
+        snprintf(ret, ret_len, "%s%s.%s.%s.%s", prefix, vl->plugin,
+                 vl->plugin_instance, vl->type, vl->type_instance);
       }
     }
   }
@@ -397,8 +455,8 @@ static int wt_send_message(const char *key, const char *value, cdtime_t time,
   }
 
   status =
-      ssnprintf(message, sizeof(message), "put %s %.0f %s fqdn=%s %s %s\r\n",
-                key, CDTIME_T_TO_DOUBLE(time), value, host, tags, host_tags);
+      snprintf(message, sizeof(message), "put %s %.0f %s fqdn=%s %s %s\r\n",
+               key, CDTIME_T_TO_DOUBLE(time), value, host, tags, host_tags);
   sfree(temp);
   if (status < 0)
     return -1;
@@ -406,7 +464,7 @@ static int wt_send_message(const char *key, const char *value, cdtime_t time,
 
   if (message_len >= sizeof(message)) {
     ERROR("write_tsdb plugin: message buffer too small: "
-          "Need %zu bytes.",
+          "Need %" PRIsz " bytes.",
           message_len + 1);
     return -1;
   }
@@ -439,8 +497,8 @@ static int wt_send_message(const char *key, const char *value, cdtime_t time,
   cb->send_buf_fill += message_len;
   cb->send_buf_free -= message_len;
 
-  DEBUG("write_tsdb plugin: [%s]:%s buf %zu/%zu (%.1f %%) \"%s\"", cb->node,
-        cb->service, cb->send_buf_fill, sizeof(cb->send_buf),
+  DEBUG("write_tsdb plugin: [%s]:%s buf %" PRIsz "/%" PRIsz " (%.1f %%) \"%s\"",
+        cb->node, cb->service, cb->send_buf_fill, sizeof(cb->send_buf),
         100.0 * ((double)cb->send_buf_fill) / ((double)sizeof(cb->send_buf)),
         message);
 
@@ -523,10 +581,8 @@ static int wt_config_tsd(oconfig_item_t *ci) {
     return -1;
   }
   cb->sock_fd = -1;
-  cb->node = NULL;
-  cb->service = NULL;
-  cb->host_tags = NULL;
-  cb->store_rates = 0;
+  cb->connect_failed_log_enabled = 1;
+  cb->next_random_ttl = new_random_ttl();
 
   pthread_mutex_init(&cb->send_lock, NULL);
 
@@ -550,9 +606,9 @@ static int wt_config_tsd(oconfig_item_t *ci) {
     }
   }
 
-  ssnprintf(callback_name, sizeof(callback_name), "write_tsdb/%s/%s",
-            cb->node != NULL ? cb->node : WT_DEFAULT_NODE,
-            cb->service != NULL ? cb->service : WT_DEFAULT_SERVICE);
+  snprintf(callback_name, sizeof(callback_name), "write_tsdb/%s/%s",
+           cb->node != NULL ? cb->node : WT_DEFAULT_NODE,
+           cb->service != NULL ? cb->service : WT_DEFAULT_SERVICE);
 
   user_data_t user_data = {.data = cb, .free_func = wt_callback_free};
 
@@ -565,11 +621,18 @@ static int wt_config_tsd(oconfig_item_t *ci) {
 }
 
 static int wt_config(oconfig_item_t *ci) {
+  if ((resolve_interval == 0) && (resolve_jitter == 0))
+    resolve_interval = resolve_jitter = plugin_get_interval();
+
   for (int i = 0; i < ci->children_num; i++) {
     oconfig_item_t *child = ci->children + i;
 
     if (strcasecmp("Node", child->key) == 0)
       wt_config_tsd(child);
+    else if (strcasecmp("ResolveInterval", child->key) == 0)
+      cf_util_get_cdtime(child, &resolve_interval);
+    else if (strcasecmp("ResolveJitter", child->key) == 0)
+      cf_util_get_cdtime(child, &resolve_jitter);
     else {
       ERROR("write_tsdb plugin: Invalid configuration "
             "option: %s.",
@@ -583,5 +646,3 @@ static int wt_config(oconfig_item_t *ci) {
 void module_register(void) {
   plugin_register_complex_config("write_tsdb", wt_config);
 }
-
-/* vim: set sw=4 ts=4 sts=4 tw=78 et : */