improve gateway resilience

git-svn-id: http://svn.freeswitch.org/svn/freeswitch/trunk@10528 d0543943-73ff-0310-b7d9-9358b9ac24b2

diff --git a/src/mod/endpoints/mod_sofia/mod_sofia.c b/src/mod/endpoints/mod_sofia/mod_sofia.c
index bb922bd8a4b9e0520789553052eb82ac6f315ade..ffab3d2a62d9ae45994115bc980999dd72cec08a 100644 (file)
--- a/src/mod/endpoints/mod_sofia/mod_sofia.c
+++ b/src/mod/endpoints/mod_sofia/mod_sofia.c
@@ -1368,6 +1368,7 @@ static const char *sofia_state_names[] = {
        "REGED",
        "UNREGISTER",
        "FAILED",
+       "FAIL_WAIT",
        "EXPIRED",
        "NOREG",
        NULL

diff --git a/src/mod/endpoints/mod_sofia/mod_sofia.h b/src/mod/endpoints/mod_sofia/mod_sofia.h
index d29d86abf8b8b10e97fa3e1c9f4d16a09505f66e..1e2be121f968cfee929e23034129e2e1db58dd3b 100644 (file)
--- a/src/mod/endpoints/mod_sofia/mod_sofia.h
+++ b/src/mod/endpoints/mod_sofia/mod_sofia.h
@@ -257,6 +257,7 @@ typedef enum {
        REG_STATE_REGED,
        REG_STATE_UNREGISTER,
        REG_STATE_FAILED,
+       REG_STATE_FAIL_WAIT,
        REG_STATE_EXPIRED,
        REG_STATE_NOREG,
        REG_STATE_LAST
@@ -333,6 +334,7 @@ struct sofia_gateway {
        int deleted;
        switch_event_t *vars;
        char uuid_str[SWITCH_UUID_FORMATTED_LENGTH + 1];
+       int failures;
        struct sofia_gateway *next;
        sofia_gateway_subscription_t *subscriptions;
 };

diff --git a/src/mod/endpoints/mod_sofia/sofia.c b/src/mod/endpoints/mod_sofia/sofia.c
index c08ea25a5901d27866a5945733af831cd64b10ab..e8032307b058c4f41072e5b522cfa56374fc6dcf 100644 (file)
--- a/src/mod/endpoints/mod_sofia/sofia.c
+++ b/src/mod/endpoints/mod_sofia/sofia.c
@@ -1001,7 +1001,7 @@ static void parse_gateways(sofia_profile_t *profile, switch_xml_t gateways_tag)
                        for (param = switch_xml_child(gateway_tag, "param"); param; param = param->next) {
                                char *var = (char *) switch_xml_attr_soft(param, "name");
                                char *val = (char *) switch_xml_attr_soft(param, "value");
-
+                               
                                if (!strcmp(var, "register")) {
                                        register_str = val;
                                } else if (!strcmp(var, "scheme")) {
@@ -1028,6 +1028,8 @@ static void parse_gateways(sofia_profile_t *profile, switch_xml_t gateways_tag)
                                        expire_seconds = val;
                                } else if (!strcmp(var, "retry-seconds")) {
                                        retry_seconds = val;
+                               } else if (!strcmp(var, "retry_seconds")) { // support typo for back compat
+                                       retry_seconds = val;
                                } else if (!strcmp(var, "from-user")) {
                                        from_user = val;
                                } else if (!strcmp(var, "from-domain")) {
@@ -1089,6 +1091,7 @@ static void parse_gateways(sofia_profile_t *profile, switch_xml_t gateways_tag)
 
                        if (!switch_true(register_str)) {
                                gateway->state = REG_STATE_NOREG;
+                               gateway->status = SOFIA_GATEWAY_UP;
                        }
 
                        if (switch_strlen_zero(from_domain)) {
@@ -1108,7 +1111,7 @@ static void parse_gateways(sofia_profile_t *profile, switch_xml_t gateways_tag)
                        }
 
                        gateway->retry_seconds = atoi(retry_seconds);
-                       if (gateway->retry_seconds < 10) {
+                       if (gateway->retry_seconds < 5) {
                                switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_WARNING, "Invalid retry-seconds of %d on gateway %s, using the value of 30 instead.\n",
                                                                  gateway->retry_seconds, name);
                                gateway->retry_seconds = 30;
@@ -2217,7 +2220,6 @@ static void sofia_handle_sip_r_options(switch_core_session_t *session, int statu
                if (status == 200 || status == 404 || status == 501) {
                        if (gateway->state == REG_STATE_FAILED) {
                                gateway->state = REG_STATE_UNREGED;
-                               gateway->retry = 0;
                        }
                        gateway->status = SOFIA_GATEWAY_UP;
                } else {

diff --git a/src/mod/endpoints/mod_sofia/sofia_reg.c b/src/mod/endpoints/mod_sofia/sofia_reg.c
index 10d3cd5cb735f1cf97dc1a9d781e09961585892f..4d7a770954f5a091f180fa46449ab90051417efc 100644 (file)
--- a/src/mod/endpoints/mod_sofia/sofia_reg.c
+++ b/src/mod/endpoints/mod_sofia/sofia_reg.c
@@ -228,13 +228,18 @@ void sofia_reg_check_gateway(sofia_profile_t *profile, time_t now)
 
                switch (ostate) {
                case REG_STATE_NOREG:
-                       gateway_ptr->status = SOFIA_GATEWAY_UP;
+                       if (!gateway_ptr->ping && !gateway_ptr->pinging) {
+                               gateway_ptr->status = SOFIA_GATEWAY_UP;
+                       }
                        break;
                case REG_STATE_REGISTER:
                        if (profile->debug) {
                                switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Registered %s\n", gateway_ptr->name);
                        }
-                       if (gateway_ptr->expires > 60) {
+
+                       gateway_ptr->failures = 0;
+
+                       if (gateway_ptr->freq > 60) {
                                gateway_ptr->expires = now + (gateway_ptr->freq - 15);
                        } else {
                                gateway_ptr->expires = now + (gateway_ptr->freq - 2);
@@ -251,7 +256,7 @@ void sofia_reg_check_gateway(sofia_profile_t *profile, time_t now)
                case REG_STATE_UNREGED:
                        gateway_ptr->status = SOFIA_GATEWAY_DOWN;
                        sofia_reg_kill_reg(gateway_ptr, 0);
-
+                       
                        if ((gateway_ptr->nh = nua_handle(gateway_ptr->profile->nua, NULL,
                                                                                          SIPTAG_CALL_ID_STR(gateway_ptr->uuid_str),
                                                                                          NUTAG_URL(gateway_ptr->register_proxy),
@@ -289,22 +294,31 @@ void sofia_reg_check_gateway(sofia_profile_t *profile, time_t now)
                                                                   NUTAG_REGISTRAR(gateway_ptr->register_proxy),
                                                                   NUTAG_OUTBOUND("no-options-keepalive"), NUTAG_OUTBOUND("no-validate"), NUTAG_KEEPALIVE(0), TAG_NULL());
                                }
-
+                               gateway_ptr->retry = now + gateway_ptr->retry_seconds;
                                gateway_ptr->state = REG_STATE_TRYING;
 
                        } else {
-                               switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error registering %s\n", gateway_ptr->name);
+                               switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error registering %s failure #%d\n", gateway_ptr->name, ++gateway_ptr->failures);
                                gateway_ptr->state = REG_STATE_FAILED;
                        }
                        break;
 
                case REG_STATE_FAILED:
+                       switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_WARNING, "%s Failed Registration, setting retry to %d seconds.\n", 
+                                                                 gateway_ptr->name, gateway_ptr->retry_seconds * (gateway_ptr->failures + 1));
+                       gateway_ptr->retry =  now + (gateway_ptr->retry_seconds * (gateway_ptr->failures + 1));
                        sofia_reg_kill_reg(gateway_ptr, 0);
                        gateway_ptr->status = SOFIA_GATEWAY_DOWN;
-               case REG_STATE_TRYING:
-                       if (gateway_ptr->retry && now >= gateway_ptr->retry) {
+                       gateway_ptr->state = REG_STATE_FAIL_WAIT;
+                       break;
+               case REG_STATE_FAIL_WAIT:
+                       if (!gateway_ptr->retry || now >= gateway_ptr->retry) {
                                gateway_ptr->state = REG_STATE_UNREGED;
-                               gateway_ptr->retry = 0;
+                       }
+                       break;
+               case REG_STATE_TRYING:
+                       if (!gateway_ptr->retry || now >= gateway_ptr->retry) {
+                               gateway_ptr->state = REG_STATE_FAILED;
                        }
                        break;
                default:
@@ -1196,7 +1210,8 @@ void sofia_reg_handle_sip_r_register(int status,
                        break;
                default:
                        sofia_private->gateway->state = REG_STATE_FAILED;
-                       switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_WARNING, "Registration Failed with status %d\n", status);
+                       switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "%s Registration Failed with status %d. failure #%d\n", 
+                                                         sofia_private->gateway->name, status, ++sofia_private->gateway->failures);
                        break;
                }
                if (ostate != sofia_private->gateway->state) {