X-Git-Url: https://git.verplant.org/?a=blobdiff_plain;f=src%2Fwrite_graphite.c;h=6baace14410e48f72761d21d6d2db1c9fdb854ad;hb=43ef00c0eb99991902d3c9a5fbe582cde049b055;hp=99c6f4d80a91a669ee0ac86f0bfe0620de6bb13f;hpb=96e0f2341bace029acefe0a88bab96ae326c0ff5;p=collectd.git
diff --git a/src/write_graphite.c b/src/write_graphite.c
index 99c6f4d8..6baace14 100644
--- a/src/write_graphite.c
+++ b/src/write_graphite.c
@@ -35,6 +35,8 @@
*
* Host "localhost"
* Port "2003"
+ * Protocol "udp"
+ * LogSendErrors true
* Prefix "collectd"
*
*
@@ -47,31 +49,23 @@
#include "utils_cache.h"
#include "utils_complain.h"
-#include "utils_parse_option.h"
#include "utils_format_graphite.h"
/* Folks without pthread will need to disable this plugin. */
#include
-#include
#include
-#ifndef WG_DEFAULT_NODE
-# define WG_DEFAULT_NODE "localhost"
-#endif
-
-#ifndef WG_DEFAULT_SERVICE
-# define WG_DEFAULT_SERVICE "2003"
-#endif
-
-#ifndef WG_DEFAULT_ESCAPE
-# define WG_DEFAULT_ESCAPE '_'
-#endif
+#define WG_DEFAULT_NODE "localhost"
+#define WG_DEFAULT_SERVICE "2003"
+#define WG_DEFAULT_PROTOCOL "tcp"
+#define WG_DEFAULT_LOG_SEND_ERRORS 1
+#define WG_DEFAULT_ESCAPE '_'
/* Ethernet - (IPv6 + TCP) = 1500 - (40 + 32) = 1428 */
-#ifndef WG_SEND_BUF_SIZE
-# define WG_SEND_BUF_SIZE 1428
-#endif
+#define WG_SEND_BUF_SIZE 1428
+
+#define WG_MIN_RECONNECT_INTERVAL TIME_T_TO_CDTIME_T (1)
/*
* Private variables
@@ -84,6 +78,8 @@ struct wg_callback
char *node;
char *service;
+ char *protocol;
+ _Bool log_send_errors;
char *prefix;
char *postfix;
char escape_char;
@@ -97,8 +93,37 @@ struct wg_callback
pthread_mutex_t send_lock;
c_complain_t init_complaint;
+ cdtime_t last_connect_time;
+
+ /* Force reconnect useful for load balanced environments */
+ cdtime_t last_reconnect_time;
+ cdtime_t reconnect_interval;
+ _Bool reconnect_interval_reached;
};
+/* wg_force_reconnect_check closes cb->sock_fd when it was open for longer
+ * than cb->reconnect_interval. Must hold cb->send_lock when calling. */
+static void wg_force_reconnect_check (struct wg_callback *cb)
+{
+ cdtime_t now;
+
+ if (cb->reconnect_interval == 0)
+ return;
+
+ /* check if address changes if addr_timeout */
+ now = cdtime ();
+ if ((now - cb->last_reconnect_time) < cb->reconnect_interval)
+ return;
+
+ /* here we should close connection on next */
+ close (cb->sock_fd);
+ cb->sock_fd = -1;
+ cb->last_reconnect_time = now;
+ cb->reconnect_interval_reached = 1;
+
+ INFO ("write_graphite plugin: Connection closed after %.3f seconds.",
+ CDTIME_T_TO_DOUBLE (now - cb->last_reconnect_time));
+}
/*
* Functions
@@ -116,12 +141,15 @@ static int wg_send_buffer (struct wg_callback *cb)
ssize_t status = 0;
status = swrite (cb->sock_fd, cb->send_buf, strlen (cb->send_buf));
- if (status < 0)
+ if (status != 0)
{
- char errbuf[1024];
- ERROR ("write_graphite plugin: send failed with status %zi (%s)",
- status, sstrerror (errno, errbuf, sizeof (errbuf)));
-
+ if (cb->log_send_errors)
+ {
+ char errbuf[1024];
+ ERROR ("write_graphite plugin: send to %s:%s (%s) failed with status %zi (%s)",
+ cb->node, cb->service, cb->protocol,
+ status, sstrerror (errno, errbuf, sizeof (errbuf)));
+ }
close (cb->sock_fd);
cb->sock_fd = -1;
@@ -169,28 +197,39 @@ static int wg_callback_init (struct wg_callback *cb)
struct addrinfo ai_hints;
struct addrinfo *ai_list;
struct addrinfo *ai_ptr;
+ cdtime_t now;
int status;
- const char *node = cb->node ? cb->node : WG_DEFAULT_NODE;
- const char *service = cb->service ? cb->service : WG_DEFAULT_SERVICE;
+ char connerr[1024] = "";
if (cb->sock_fd > 0)
return (0);
+ /* Don't try to reconnect too often. By default, one reconnection attempt
+ * is made per second. */
+ now = cdtime ();
+ if ((now - cb->last_connect_time) < WG_MIN_RECONNECT_INTERVAL)
+ return (EAGAIN);
+ cb->last_connect_time = now;
+
memset (&ai_hints, 0, sizeof (ai_hints));
#ifdef AI_ADDRCONFIG
ai_hints.ai_flags |= AI_ADDRCONFIG;
#endif
ai_hints.ai_family = AF_UNSPEC;
- ai_hints.ai_socktype = SOCK_STREAM;
+
+ if (0 == strcasecmp ("tcp", cb->protocol))
+ ai_hints.ai_socktype = SOCK_STREAM;
+ else
+ ai_hints.ai_socktype = SOCK_DGRAM;
ai_list = NULL;
- status = getaddrinfo (node, service, &ai_hints, &ai_list);
+ status = getaddrinfo (cb->node, cb->service, &ai_hints, &ai_list);
if (status != 0)
{
- ERROR ("write_graphite plugin: getaddrinfo (%s, %s) failed: %s",
- node, service, gai_strerror (status));
+ ERROR ("write_graphite plugin: getaddrinfo (%s, %s, %s) failed: %s",
+ cb->node, cb->service, cb->protocol, gai_strerror (status));
return (-1);
}
@@ -199,12 +238,19 @@ static int wg_callback_init (struct wg_callback *cb)
{
cb->sock_fd = socket (ai_ptr->ai_family, ai_ptr->ai_socktype,
ai_ptr->ai_protocol);
- if (cb->sock_fd < 0)
+ if (cb->sock_fd < 0) {
+ char errbuf[1024];
+ snprintf (connerr, sizeof (connerr), "failed to open socket: %s",
+ sstrerror (errno, errbuf, sizeof (errbuf)));
continue;
+ }
status = connect (cb->sock_fd, ai_ptr->ai_addr, ai_ptr->ai_addrlen);
if (status != 0)
{
+ char errbuf[1024];
+ snprintf (connerr, sizeof (connerr), "failed to connect to remote "
+ "host: %s", sstrerror (errno, errbuf, sizeof (errbuf)));
close (cb->sock_fd);
cb->sock_fd = -1;
continue;
@@ -217,21 +263,28 @@ static int wg_callback_init (struct wg_callback *cb)
if (cb->sock_fd < 0)
{
- char errbuf[1024];
+ if (connerr[0] == '\0')
+ /* this should not happen but try to get a message anyway */
+ sstrerror (errno, connerr, sizeof (connerr));
c_complain (LOG_ERR, &cb->init_complaint,
- "write_graphite plugin: Connecting to %s:%s failed. "
- "The last error was: %s", node, service,
- sstrerror (errno, errbuf, sizeof (errbuf)));
+ "write_graphite plugin: Connecting to %s:%s via %s failed. "
+ "The last error was: %s", cb->node, cb->service, cb->protocol, connerr);
return (-1);
}
else
{
c_release (LOG_INFO, &cb->init_complaint,
- "write_graphite plugin: Successfully connected to %s:%s.",
- node, service);
+ "write_graphite plugin: Successfully connected to %s:%s via %s.",
+ cb->node, cb->service, cb->protocol);
}
- wg_reset_buffer (cb);
+ /* wg_force_reconnect_check does not flush the buffer before closing a
+ * sending socket, so only call wg_reset_buffer() if the socket was closed
+ * for a different reason (tracked in cb->reconnect_interval_reached). */
+ if (!cb->reconnect_interval_reached || (cb->send_buf_free == 0))
+ wg_reset_buffer (cb);
+ else
+ cb->reconnect_interval_reached = 0;
return (0);
}
@@ -257,6 +310,7 @@ static void wg_callback_free (void *data)
sfree(cb->name);
sfree(cb->node);
+ sfree(cb->protocol);
sfree(cb->service);
sfree(cb->prefix);
sfree(cb->postfix);
@@ -306,6 +360,8 @@ static int wg_send_message (char const *message, struct wg_callback *cb)
pthread_mutex_lock (&cb->send_lock);
+ wg_force_reconnect_check (cb);
+
if (cb->sock_fd < 0)
{
status = wg_callback_init (cb);
@@ -337,9 +393,8 @@ static int wg_send_message (char const *message, struct wg_callback *cb)
cb->send_buf_fill += message_len;
cb->send_buf_free -= message_len;
- DEBUG ("write_graphite plugin: [%s]:%s buf %zu/%zu (%.1f %%) \"%s\"",
- cb->node,
- cb->service,
+ DEBUG ("write_graphite plugin: [%s]:%s (%s) buf %zu/%zu (%.1f %%) \"%s\"",
+ cb->node, cb->service, cb->protocol,
cb->send_buf_fill, sizeof (cb->send_buf),
100.0 * ((double) cb->send_buf_fill) / ((double) sizeof (cb->send_buf)),
message);
@@ -429,6 +484,7 @@ static int wg_config_node (oconfig_item_t *ci)
user_data_t user_data;
char callback_name[DATA_MAX_NAME_LEN];
int i;
+ int status = 0;
cb = malloc (sizeof (*cb));
if (cb == NULL)
@@ -439,8 +495,13 @@ static int wg_config_node (oconfig_item_t *ci)
memset (cb, 0, sizeof (*cb));
cb->sock_fd = -1;
cb->name = NULL;
- cb->node = NULL;
- cb->service = NULL;
+ cb->node = strdup (WG_DEFAULT_NODE);
+ cb->service = strdup (WG_DEFAULT_SERVICE);
+ cb->protocol = strdup (WG_DEFAULT_PROTOCOL);
+ cb->last_reconnect_time = cdtime();
+ cb->reconnect_interval = 0;
+ cb->reconnect_interval_reached = 0;
+ cb->log_send_errors = WG_DEFAULT_LOG_SEND_ERRORS;
cb->prefix = NULL;
cb->postfix = NULL;
cb->escape_char = WG_DEFAULT_ESCAPE;
@@ -449,7 +510,7 @@ static int wg_config_node (oconfig_item_t *ci)
/* FIXME: Legacy configuration syntax. */
if (strcasecmp ("Carbon", ci->key) != 0)
{
- int status = cf_util_get_string (ci, &cb->name);
+ status = cf_util_get_string (ci, &cb->name);
if (status != 0)
{
wg_callback_free (cb);
@@ -468,6 +529,22 @@ static int wg_config_node (oconfig_item_t *ci)
cf_util_get_string (child, &cb->node);
else if (strcasecmp ("Port", child->key) == 0)
cf_util_get_service (child, &cb->service);
+ else if (strcasecmp ("Protocol", child->key) == 0)
+ {
+ cf_util_get_string (child, &cb->protocol);
+
+ if (strcasecmp ("UDP", cb->protocol) != 0 &&
+ strcasecmp ("TCP", cb->protocol) != 0)
+ {
+ ERROR ("write_graphite plugin: Unknown protocol (%s)",
+ cb->protocol);
+ status = -1;
+ }
+ }
+ else if (strcasecmp ("ReconnectInterval", child->key) == 0)
+ cf_util_get_cdtime (child, &cb->reconnect_interval);
+ else if (strcasecmp ("LogSendErrors", child->key) == 0)
+ cf_util_get_boolean (child, &cb->log_send_errors);
else if (strcasecmp ("Prefix", child->key) == 0)
cf_util_get_string (child, &cb->prefix);
else if (strcasecmp ("Postfix", child->key) == 0)
@@ -487,14 +564,23 @@ static int wg_config_node (oconfig_item_t *ci)
{
ERROR ("write_graphite plugin: Invalid configuration "
"option: %s.", child->key);
+ status = -1;
}
+
+ if (status != 0)
+ break;
+ }
+
+ if (status != 0)
+ {
+ wg_callback_free (cb);
+ return (status);
}
/* FIXME: Legacy configuration syntax. */
if (cb->name == NULL)
- ssnprintf (callback_name, sizeof (callback_name), "write_graphite/%s/%s",
- cb->node != NULL ? cb->node : WG_DEFAULT_NODE,
- cb->service != NULL ? cb->service : WG_DEFAULT_SERVICE);
+ ssnprintf (callback_name, sizeof (callback_name), "write_graphite/%s/%s/%s",
+ cb->node, cb->service, cb->protocol);
else
ssnprintf (callback_name, sizeof (callback_name), "write_graphite/%s",
cb->name);