res/res_pjsip_outbound_registration: Add registration statistics for StatsD

This patch adds outbound registration statistics for StatsD. This includes
the following:
 * A GUAGE metric for the overall count of outbound registrations.
 * A GUAGE metric for each state an outbound registration can be in. As the
   outbound registrations change state, the overall count of how many
   outbound registrations are in the particular state is changed.

These statistics are particularly useful for systems with a large number of
SIP trunks, and where measuring the change in state of the trunks is useful
for monitoring.

ASTERISK-25571

Change-Id: Iba6ff248f5d1c1e01acbb63e9f0da1901692eb37

diff --git a/CHANGES b/CHANGES
index 4a55a06ff2c3c5b17d307bce95824e08a345bb28..348927b1d89c77edddfa40e8c662e1bbad7cb729 100644 (file)
--- a/CHANGES
+++ b/CHANGES
@@ -36,10 +36,19 @@ Dialplan Functions
 
 res_pjsip_outbound_registration
 -------------------------------
-* A new 'fatal_retry_interval' option has been added to outbound registration.
-  When set (default is zero), and upon receiving a failure response to an
-  outbound registration, registration is retried at the given interval up to
-  'max_retries'.
+ * A new 'fatal_retry_interval' option has been added to outbound registration.
+   When set (default is zero), and upon receiving a failure response to an
+   outbound registration, registration is retried at the given interval up to
+   'max_retries'.
+
+ * If res_statsd is loaded and a StatsD server is configured, basic statistics
+   regarding the state of outbound registrations will now be emitted. This
+   includes:
+   - A GUAGE statistic for the overall number of outbound registrations, i.e.:
+       PJSIP.registrations.count
+   - A GUAGE statistic for the overall number of outbound registrations in a
+     particular state, e.g.:
+       PJSIP.registrations.state.Registered
 
 res_pjsip
 ------------------

diff --git a/res/res_pjsip_outbound_registration.c b/res/res_pjsip_outbound_registration.c
index 04ff1df607200478846398dbda13e55bd6896720..22657ecd6ed3e87e2f80295aecd4deb03589475a 100644 (file)
--- a/res/res_pjsip_outbound_registration.c
+++ b/res/res_pjsip_outbound_registration.c
@@ -35,6 +35,7 @@
 #include "asterisk/stasis_system.h"
 #include "asterisk/threadstorage.h"
 #include "asterisk/threadpool.h"
+#include "asterisk/statsd.h"
 #include "res_pjsip/include/res_pjsip_private.h"
 
 /*** DOCUMENTATION
@@ -610,6 +611,19 @@ static void schedule_registration(struct sip_outbound_registration_client_state
        }
 }
 
+static void update_client_state_status(struct sip_outbound_registration_client_state *client_state, enum sip_outbound_registration_status status)
+{
+       if (client_state->status == status) {
+               return;
+       }
+
+       ast_statsd_log_string_va("PJSIP.registrations.state.%s", AST_STATSD_GUAGE, "-1", 1.0,
+               sip_outbound_registration_status_str(client_state->status));
+       ast_statsd_log_string_va("PJSIP.registrations.state.%s", AST_STATSD_GUAGE, "+1", 1.0,
+               sip_outbound_registration_status_str(status));
+       client_state->status = status;
+}
+
 /*! \brief Callback function for unregistering (potentially) and destroying state */
 static int handle_client_state_destruction(void *data)
 {
@@ -643,7 +657,7 @@ static int handle_client_state_destruction(void *data)
                                (int) info.server_uri.slen, info.server_uri.ptr,
                                (int) info.client_uri.slen, info.client_uri.ptr);
 
-                       client_state->status = SIP_REGISTRATION_STOPPING;
+                       update_client_state_status(client_state, SIP_REGISTRATION_STOPPING);
                        client_state->destroy = 1;
                        if (pjsip_regc_unregister(client_state->client, &tdata) == PJ_SUCCESS
                                && registration_client_send(client_state, tdata) == PJ_SUCCESS) {
@@ -662,7 +676,7 @@ static int handle_client_state_destruction(void *data)
                client_state->client = NULL;
        }
 
-       client_state->status = SIP_REGISTRATION_STOPPED;
+       update_client_state_status(client_state, SIP_REGISTRATION_STOPPED);
        ast_sip_auth_vector_destroy(&client_state->outbound_auths);
        ao2_ref(client_state, -1);
 
@@ -724,7 +738,7 @@ static int sip_outbound_registration_is_temporal(unsigned int code,
 static void schedule_retry(struct registration_response *response, unsigned int interval,
                           const char *server_uri, const char *client_uri)
 {
-       response->client_state->status = SIP_REGISTRATION_REJECTED_TEMPORARY;
+       update_client_state_status(response->client_state, SIP_REGISTRATION_REJECTED_TEMPORARY);
        schedule_registration(response->client_state, interval);
 
        if (response->rdata) {
@@ -784,12 +798,12 @@ static int handle_registration_response(void *data)
                if (response->expiration) {
                        /* If the registration went fine simply reschedule registration for the future */
                        ast_debug(1, "Outbound registration to '%s' with client '%s' successful\n", server_uri, client_uri);
-                       response->client_state->status = SIP_REGISTRATION_REGISTERED;
+                       update_client_state_status(response->client_state, SIP_REGISTRATION_REGISTERED);
                        response->client_state->retries = 0;
                        schedule_registration(response->client_state, response->expiration - REREGISTER_BUFFER_TIME);
                } else {
                        ast_debug(1, "Outbound unregistration to '%s' with client '%s' successful\n", server_uri, client_uri);
-                       response->client_state->status = SIP_REGISTRATION_UNREGISTERED;
+                       update_client_state_status(response->client_state, SIP_REGISTRATION_UNREGISTERED);
                }
        } else if (response->client_state->destroy) {
                /* We need to deal with the pending destruction instead. */
@@ -800,7 +814,7 @@ static int handle_registration_response(void *data)
                && sip_outbound_registration_is_temporal(response->code, response->client_state)) {
                if (response->client_state->retries == response->client_state->max_retries) {
                        /* If we received enough temporal responses to exceed our maximum give up permanently */
-                       response->client_state->status = SIP_REGISTRATION_REJECTED_PERMANENT;
+                       update_client_state_status(response->client_state, SIP_REGISTRATION_REJECTED_PERMANENT);
                        ast_log(LOG_WARNING, "Maximum retries reached when attempting outbound registration to '%s' with client '%s', stopping registration attempt\n",
                                server_uri, client_uri);
                } else {
@@ -813,7 +827,7 @@ static int handle_registration_response(void *data)
                        && response->client_state->forbidden_retry_interval
                        && response->client_state->retries < response->client_state->max_retries) {
                        /* A forbidden response retry interval is configured and there are retries remaining */
-                       response->client_state->status = SIP_REGISTRATION_REJECTED_TEMPORARY;
+                       update_client_state_status(response->client_state, SIP_REGISTRATION_REJECTED_TEMPORARY);
                        response->client_state->retries++;
                        schedule_registration(response->client_state, response->client_state->forbidden_retry_interval);
                        ast_log(LOG_WARNING, "403 Forbidden fatal response received from '%s' on registration attempt to '%s', retrying in '%u' seconds\n",
@@ -821,14 +835,14 @@ static int handle_registration_response(void *data)
                } else if (response->client_state->fatal_retry_interval
                           && response->client_state->retries < response->client_state->max_retries) {
                        /* Some kind of fatal failure response received, so retry according to configured interval */
-                       response->client_state->status = SIP_REGISTRATION_REJECTED_TEMPORARY;
+                       update_client_state_status(response->client_state, SIP_REGISTRATION_REJECTED_TEMPORARY);
                        response->client_state->retries++;
                        schedule_registration(response->client_state, response->client_state->fatal_retry_interval);
                        ast_log(LOG_WARNING, "'%d' fatal response received from '%s' on registration attempt to '%s', retrying in '%u' seconds\n",
                                response->code, server_uri, client_uri, response->client_state->fatal_retry_interval);
                } else {
                        /* Finally if there's no hope of registering give up */
-                       response->client_state->status = SIP_REGISTRATION_REJECTED_PERMANENT;
+                       update_client_state_status(response->client_state, SIP_REGISTRATION_REJECTED_PERMANENT);
                        if (response->rdata) {
                                ast_log(LOG_WARNING, "Fatal response '%d' received from '%s' on registration attempt to '%s', stopping outbound registration\n",
                                        response->code, server_uri, client_uri);
@@ -913,7 +927,6 @@ static void sip_outbound_registration_state_destroy(void *obj)
 
        ast_debug(3, "Destroying registration state for registration to server '%s' from client '%s'\n",
                        state->registration->server_uri, state->registration->client_uri);
-
        ao2_cleanup(state->registration);
 
        if (!state->client_state) {
@@ -932,6 +945,10 @@ static void sip_outbound_registration_client_state_destroy(void *obj)
 {
        struct sip_outbound_registration_client_state *client_state = obj;
 
+       ast_statsd_log_string("PJSIP.registrations.count", AST_STATSD_GUAGE, "-1", 1.0);
+       ast_statsd_log_string_va("PJSIP.registrations.state.%s", AST_STATSD_GUAGE, "-1", 1.0,
+               sip_outbound_registration_status_str(client_state->status));
+
        ast_taskprocessor_unreference(client_state->serializer);
 }
 
@@ -960,6 +977,10 @@ static struct sip_outbound_registration_state *sip_outbound_registration_state_a
        state->client_state->timer.user_data = state->client_state;
        state->client_state->timer.cb = sip_outbound_registration_timer_cb;
 
+       ast_statsd_log_string("PJSIP.registrations.count", AST_STATSD_GUAGE, "+1", 1.0);
+       ast_statsd_log_string_va("PJSIP.registrations.state.%s", AST_STATSD_GUAGE, "+1", 1.0,
+               sip_outbound_registration_status_str(state->client_state->status));
+
        state->registration = ao2_bump(registration);
        return state;
 }
@@ -1985,6 +2006,12 @@ static int load_module(void)
        ast_manager_register_xml("PJSIPRegister", EVENT_FLAG_SYSTEM | EVENT_FLAG_REPORTING, ami_register);
        ast_manager_register_xml("PJSIPShowRegistrationsOutbound", EVENT_FLAG_SYSTEM | EVENT_FLAG_REPORTING, ami_show_outbound_registrations);
 
+       /* Clear any previous statsd gauges in case we weren't shutdown cleanly */
+       ast_statsd_log("PJSIP.registrations.count", AST_STATSD_GUAGE, 0);
+       ast_statsd_log("PJSIP.registrations.state.Registered", AST_STATSD_GUAGE, 0);
+       ast_statsd_log("PJSIP.registrations.state.Unregistered", AST_STATSD_GUAGE, 0);
+       ast_statsd_log("PJSIP.registrations.state.Rejected", AST_STATSD_GUAGE, 0);
+
        /* Load configuration objects */
        ast_sorcery_load_object(ast_sip_get_sorcery(), "registration");