Use random port for federate servers

Author: edwardalee

core/federated/RTI/rti_remote.c

@@ -1007,7 +1007,7 @@ void handle_federate_resign(federate_info_t *my_fed) {
     // an orderly shutdown.
     // close(my_fed->socket); //  from unistd.h
 
-    lf_print("Federate %d has resigned.", my_fed->enclave.id);
+    lf_print("RTI: Federate %d has resigned.", my_fed->enclave.id);
 
     // Check downstream federates to see whether they should now be granted a TAG.
     // To handle cycles, need to create a boolean array to keep

core/federated/federate.c

@@ -116,22 +116,9 @@ federation_metadata_t federation_metadata = {
 };
 
 void create_server(int specified_port, int id) {
-    if (specified_port > UINT16_MAX ||
-        specified_port < 0) {
-        lf_print_error(
-            "create_server(): The specified port (%d) is out of range."
-            " Starting with %d instead.",
-            specified_port,
-            DEFAULT_PORT
-        );
-        specified_port = 0;
-    }
+    assert(specified_port <= UINT16_MAX && specified_port >= 0);
     uint16_t port = (uint16_t)specified_port;
-    if (specified_port == 0) {
-        // Use the default starting port + 1 + id (to try to get unique ports for each federate)
-        port = DEFAULT_PORT + 1 + id;
-    }
-    lf_print("Creating a socket server on port %d.", port);
+    LF_PRINT_LOG("Creating a socket server on port %d.", port);
     // Create an IPv4 socket for TCP (not UDP) communication over IP (0).
     int socket_descriptor = create_real_time_tcp_socket_errexit();
 
@@ -149,40 +136,37 @@ void create_server(int specified_port, int id) {
             socket_descriptor,
             (struct sockaddr *) &server_fd,
             sizeof(server_fd));
-    // If the binding fails with this port and no particular port was specified
-    // in the LF program, then try the next few ports in sequence.
     int count = 0;
-    while (result != 0 && count++ < PORT_BIND_RETRY_LIMIT) {
-        if (specified_port != 0) {
-            port++;
-            server_fd.sin_port = htons(port);
-            LF_PRINT_DEBUG("Failed to get port. Trying %d.", port);
-        } else {
-            LF_PRINT_DEBUG("Failed to get port %d. Will try again after waiting.", port);
-            lf_sleep(PORT_BIND_RETRY_INTERVAL);
-        }
-        server_fd.sin_port = htons(port);
+    while (result < 0 && count++ < PORT_BIND_RETRY_LIMIT) {
+        lf_sleep(PORT_BIND_RETRY_INTERVAL);
         result = bind(
                 socket_descriptor,
                 (struct sockaddr *) &server_fd,
                 sizeof(server_fd));
     }
     if (result < 0) {
-        if (specified_port == 0) {
-            lf_print_error_and_exit("Failed to bind socket. Cannot find a usable port.");
-        } else {
-            lf_print_error_and_exit("Failed to bind socket. Specified port is not available.");
+        lf_print_error_and_exit("Failed to bind socket on port %d.", port);
+    }
+
+    // Set the global server port.
+    if (specified_port == 0) {
+        // Need to retrieve the port number assigned by the OS.
+        struct sockaddr_in assigned;
+        socklen_t addr_len = sizeof(assigned);
+        if (getsockname(socket_descriptor, (struct sockaddr *) &assigned, &addr_len) < 0) {
+            lf_print_error_and_exit("Failed to retrieve assigned port number.");
         }
+        _fed.server_port = ntohs(assigned.sin_port);
+    } else {
+        _fed.server_port = port;
     }
-    LF_PRINT_LOG("Server for communicating with other federates started using port %d.", port);
 
     // Enable listening for socket connections.
     // The second argument is the maximum number of queued socket requests,
     // which according to the Mac man page is limited to 128.
     listen(socket_descriptor, 128);
 
-    // Set the global server port
-    _fed.server_port = port;
+    lf_print("Server for communicating with other federates started using port %d.", _fed.server_port);
 
     // Send the server port number to the RTI
     // on an MSG_TYPE_ADDRESS_ADVERTISEMENT message (@see net_common.h).
@@ -2288,6 +2272,7 @@ void terminate_execution(environment_t* env) {
     // function should NEVER be called while holding any mutex lock.
     lf_mutex_lock(&outbound_socket_mutex);
     for (int i=0; i < NUMBER_OF_FEDERATES; i++) {
+
         // Close outbound connections, in case they have not closed themselves.
         // This will result in EOF being sent to the remote federate, I think.
         _lf_close_outbound_socket(i);
@@ -2319,10 +2304,11 @@ void terminate_execution(environment_t* env) {
 
     LF_PRINT_DEBUG("Waiting for inbound p2p socket listener threads.");
     // Wait for each inbound socket listener thread to close.
-    if (_fed.number_of_inbound_p2p_connections > 0) {
+    if (_fed.number_of_inbound_p2p_connections > 0 && _fed.inbound_socket_listeners != NULL) {
         LF_PRINT_LOG("Waiting for %zu threads listening for incoming messages to exit.",
                 _fed.number_of_inbound_p2p_connections);
         for (int i=0; i < _fed.number_of_inbound_p2p_connections; i++) {
+            if (_fed.inbound_socket_listeners[i] == NULL) continue;
             // Ignoring errors here.
             lf_thread_join(_fed.inbound_socket_listeners[i], NULL);
         }

core/reactor_common.c

@@ -265,6 +265,11 @@ void _lf_trigger_reaction(environment_t* env, reaction_t* reaction, int worker_n
  * counts between time steps and at the end of execution.
  */
 void _lf_start_time_step(environment_t *env) {
+    if (_lf_execution_started == false) {
+        // Execution hasn't started, so this is probably being invoked in termination
+        // due to an error.
+        return;
+    }
     assert(env != GLOBAL_ENVIRONMENT);
     LF_PRINT_LOG("--------- Start time step at tag " PRINTF_TAG ".", env->current_tag.time - start_time, env->current_tag.microstep);
     // Handle dynamically created tokens for mutable inputs.
@@ -1726,14 +1731,13 @@ void termination(void) {
     // It should only be called for the top-level environment, which, after convention, is the first environment.
     terminate_execution(env);
 
-
     // In order to free tokens, we perform the same actions we would have for a new time step.
     for (int i = 0; i<num_envs; i++) {
-        lf_print("---- Terminating environment %u", env->id);
-        if (!env->initialized) {
+        if (env == NULL || !env->initialized) {
             lf_print_warning("---- Environment %u was never initialized", env->id);
             continue;
         }
+        lf_print("---- Terminating environment %u", env->id);
         // Stop any tracing, if it is running.
         stop_trace_locked(env->trace);
 
@@ -1743,7 +1747,6 @@ void termination(void) {
         // Free events and tokens suspended by modal reactors.
         _lf_terminate_modal_reactors(env);
     #endif
-
         // If the event queue still has events on it, report that.
         if (env->event_q != NULL && pqueue_size(env->event_q) > 0) {
             lf_print_warning("---- There are %zu unprocessed future events on the event queue.", pqueue_size(env->event_q));

include/core/federated/federate.h

@@ -305,22 +305,14 @@ void* listen_to_federates(void*);
  * (@see net_common.h). This function expects no response
  * from the RTI.
  *
- * If a port is specified by the user, that will be used
- * as the only possibility for the server. This function
- * will fail if that port is not available. If a port is not
- * specified, the DEFAULT_PORT + 1 + id will be attempted,
- * The function will keep incrementing the port in this case
- * until the number of tries reaches PORT_BIND_RETRY_LIMIT (@see net_common.h).
+ * If a port is specified by the user, that will be used.
+ * Otherwise, a random port will be assigned.  If the bind fails,
+ * it will retry after PORT_BIND_RETRY_INTERVAL until it has tried
+ * PORT_BIND_RETRY_LIMIT times. Then it will fail.
  *
- * @note This function is similar to create_server(...) in rti.c.
- * However, it contains specific log messages for the peer to
- * peer connections between federates. It also additionally
- * sends an address advertisement (MSG_TYPE_ADDRESS_ADVERTISEMENT) message to the
- * RTI informing it of the port.  Finally, it tries multiple ports
- * because the port used by a federate is not as important as the port used
- * by the RTI.
+ * @note This function is different from create_server(...) in rti.c.
  *
- * @param specified_port The specified port by the user.
+ * @param specified_port The specified port by the user or 0 to use a random port.
  * @param id The id of the federate (to help find a unique port).
  */
 void create_server(int specified_port, int id);

include/core/federated/network/net_common.h

@@ -243,14 +243,10 @@ THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define PORT_BIND_RETRY_LIMIT 100
 
 /**
- * Default port number for the RTI and base number for federates' socket server.
+ * Default port number for the RTI.
  * Unless a specific port has been specified by the LF program in the "at"
  * for the RTI or on the command line, when the RTI starts up, it will attempt
- * to open a socket server on this port. The federates start attempting to bind
- * to ports one greater than this plus the federate ID, and if that port is not,
- * then they try incrementing the port number with up to PORT_BIND_RETRY_LIMIT
- * attempts.  The RTI will try repeatedly on the same port number with a delay
- * of PORT_BIND_RETRY_INTERVAL and a maximum number of attempts of PORT_BIND_RETRY_LIMIT.
+ * to open a socket server on this port.
  */
 #define DEFAULT_PORT 15045u