Merge pull request #294 from bill1600/cnheal

Add logic to save parodus connection status in health file

src/config.c

@@ -365,6 +365,7 @@ int parseCommandLine(int argc,char **argv,ParodusCfg * cfg)
 	{"client-cert-path",        required_argument, 0, 'P'},
 	{"token-server-url",        required_argument, 0, 'U'},
 	{"crud-config-file",        required_argument, 0, 'C'},
+	{"connection-health-file",  required_argument, 0, 'S'},
 	{"mtls-client-key-path",    required_argument, 0, 'K'},
 	{"mtls-client-cert-path",    required_argument, 0,'M'},
         {0, 0, 0, 0}
@@ -382,6 +383,7 @@ int parseCommandLine(int argc,char **argv,ParodusCfg * cfg)
 	cfg->jwt_algo = 0;
 	parStrncpy (cfg->jwt_key, "", sizeof(cfg->jwt_key));
 	cfg->crud_config_file = NULL;
+	cfg->connection_health_file = NULL;
 	cfg->client_cert_path = NULL;
 	cfg->token_server_url = NULL;
 	cfg->cloud_status = NULL;
@@ -392,7 +394,7 @@ int parseCommandLine(int argc,char **argv,ParodusCfg * cfg)
 
       /* getopt_long stores the option index here. */
       int option_index = 0;
-      c = getopt_long (argc, argv, "m:s:f:d:r:n:b:u:t:o:i:l:p:e:D:j:a:k:c:T:w:J:46:C:K:M",
+      c = getopt_long (argc, argv, "m:s:f:d:r:n:b:u:t:o:i:l:p:e:D:j:a:k:c:T:w:J:46:C:S:K:M",
 				long_options, &option_index);
 
       /* Detect the end of the options. */
@@ -530,10 +532,15 @@ int parseCommandLine(int argc,char **argv,ParodusCfg * cfg)
           ParodusInfo("boot_retry_wait is %d\n",cfg->boot_retry_wait);
           break;
 
-		case 'C':
-		  cfg->crud_config_file = strdup(optarg);
-		  ParodusInfo("crud_config_file is %s\n", cfg->crud_config_file);
-		  break;
+	case 'S':
+	  cfg->connection_health_file = strdup(optarg);
+	  ParodusInfo("connection_health_file is %s\n", cfg->connection_health_file);
+	  break;
+
+	case 'C':
+	  cfg->crud_config_file = strdup(optarg);
+	  ParodusInfo("crud_config_file is %s\n", cfg->crud_config_file);
+	  break;
 
 		case 'P':
 		cfg->client_cert_path = strdup(optarg);
@@ -628,6 +635,7 @@ void setDefaultValuesToCfg(ParodusCfg *cfg)
     parStrncpy(cfg->webpa_uuid, "1234567-345456546",sizeof(cfg->webpa_uuid));
     ParodusPrint("cfg->webpa_uuid is :%s\n", cfg->webpa_uuid);
     cfg->crud_config_file = NULL;
+    cfg->connection_health_file = NULL;
     cfg->client_cert_path = NULL;
     cfg->token_server_url = NULL;
 
@@ -779,6 +787,15 @@ void loadParodusCfg(ParodusCfg * config,ParodusCfg *cfg)
     parStrncpy(cfg->webpa_uuid, "1234567-345456546",sizeof(cfg->webpa_uuid));
     ParodusPrint("cfg->webpa_uuid is :%s\n", cfg->webpa_uuid);
 
+    if(config->connection_health_file != NULL)
+    {
+        cfg->connection_health_file = strdup(config->connection_health_file);
+    }
+    else
+    {
+        ParodusPrint("connection_health_file is NULL. set to empty\n");
+    }
+
     if(config->crud_config_file != NULL)
     {
         cfg->crud_config_file = strdup(config->crud_config_file);

src/config.h

@@ -101,6 +101,7 @@ typedef struct
     char token_read_script[64];
     char *client_cert_path;
     char *token_server_url;
+    char *connection_health_file;
     char *mtls_client_key_path;
     char *mtls_client_cert_path;
 	char *crud_config_file;

src/conn_interface.c

@@ -74,6 +74,7 @@ void createSocketConnection(void (* initKeypress)())
     //ParodusCfg *tmpCfg = (ParodusCfg*)config_in;
     noPollCtx *ctx;
     bool seshat_registered = false;
+    int create_conn_rtn = 0;
     unsigned int webpa_ping_timeout_ms = 1000 * get_parodus_cfg()->webpa_ping_timeout;
     unsigned int heartBeatTimer = 0;
     struct timespec start_svc_alive_timer;
@@ -94,7 +95,10 @@ void createSocketConnection(void (* initKeypress)())
     nopoll_log_set_handler (ctx, __report_log, NULL);
     #endif
 
-    if(!createNopollConnection(ctx))
+    start_conn_in_progress ();
+    create_conn_rtn = createNopollConnection(ctx);
+    stop_conn_in_progress ();
+    if(!create_conn_rtn)
     {
 		ParodusError("Unrecovered error, terminating the process\n");
 		abort();
@@ -177,7 +181,9 @@ void createSocketConnection(void (* initKeypress)())
 		free(get_parodus_cfg()->cloud_disconnect);
 		reset_cloud_disconnect_reason(get_parodus_cfg());
             }
+            start_conn_in_progress ();
             createNopollConnection(ctx);
+            stop_conn_in_progress ();
         }
        } while(!get_close_retry() && !g_shutdown);
 

src/connection.c

@@ -662,3 +662,38 @@ void close_and_unref_connection(noPollConn *conn)
     }
 }
 
+void write_conn_in_prog_file (const char *msg)
+{
+  int fd;
+  FILE *fp;
+  unsigned long timestamp;
+  ParodusCfg *cfg = get_parodus_cfg();
+
+  if (NULL == cfg->connection_health_file)
+    return;
+  fd = open (cfg->connection_health_file, O_CREAT | O_WRONLY | O_SYNC, 0666);
+  if (fd < 0) {
+    ParodusError ("Error(1) %d opening file %s\n", errno, cfg->connection_health_file);
+    return;
+  }
+  ftruncate (fd, 0);
+  fp = fdopen (fd, "w");
+  if (fp == NULL) {
+    ParodusError ("Error(2) %d opening file %s\n", errno, cfg->connection_health_file);
+    return;
+  }
+  timestamp = (unsigned long) time(NULL);
+  fprintf (fp, "{%s=%lu}\n", msg, timestamp);
+  fclose (fp);
+}
+
+void start_conn_in_progress (void)
+{
+  write_conn_in_prog_file ("START");
+}   
+
+void stop_conn_in_progress (void)
+{
+  write_conn_in_prog_file ("STOP");
+}   
+

src/connection.h

@@ -57,6 +57,13 @@ void set_global_reconnect_status(bool status);
 
 int get_cloud_disconnect_time();
 void set_cloud_disconnect_time(int disconnTime);
+
+/**
+ * @brief Interface to self heal connection in progress getting stuck
+ */
+void start_conn_in_progress (void);
+void stop_conn_in_progress (void);
+
 #ifdef __cplusplus
 }
 #endif

tests/test_conn_interface.c

@@ -74,6 +74,15 @@ noPollMutexUnlock 	mutex_unlock
     UNUSED(mutex_create); UNUSED(mutex_destroy); UNUSED(mutex_lock); UNUSED(mutex_unlock);
     function_called();
 }
+
+void start_conn_in_progress (void)
+{
+}   
+
+void stop_conn_in_progress (void)
+{
+}   
+
 void packMetaData()
 {
     function_called();