report last version

[ue-pp-terminationdetectionalgorithm.git] / control.c

diff --git a/control.c b/control.c
index e58ba34936332059d4dd2effe9f3f8e1df94d187..67a4df8ed42efda5d0917535895e46707fa40ea6 100644 (file)
--- a/control.c
+++ b/control.c
@@ -1,6 +1,5 @@
 /*
   Parallel Programming - 2026/2027
 /*
   Parallel Programming - 2026/2027

-

   Implementation of a termination detection algorithm (Dijkstra-Scholten)
 */
 
   Implementation of a termination detection algorithm (Dijkstra-Scholten)
 */
 


@@ -14,12 +13,18 @@
 #include "control.h"
 #include "util.h"
 
 #include "control.h"
 #include "util.h"
 

-// Tag for control messages. Distinct from basic algorithm's tags.
+// Tag for control messages. Distinct from basic algorithm's tags (1).

 #define CONTROL_SIGNAL 99
 
 #define CONTROL_SIGNAL 99
 

-// Type for control messages (empty payload, just a signal)
+// Message types for control routing
+typedef enum {
+    MSG_ACK,
+    MSG_KILL
+} msg_type_t;
+
+// Type for control messages

 typedef struct {
 typedef struct {

-    int dummy;
+    msg_type_t type;

 } control_message_t;
 
 // Node operational states
 } control_message_t;
 
 // Node operational states


@@ -28,29 +33,26 @@ typedef enum {
     PASSIVE_STATE
 } node_state_t;
 
     PASSIVE_STATE
 } node_state_t;
 

-// --- SHARED LOCAL STATE ---
-// Must be protected by mutex because basic_thread and control_thread access it concurrently
+
+// SHARED LOCAL STATE - Must be protected by mutex because basic_thread and control_thread access it concurrently

 static pthread_mutex_t state_mutex = PTHREAD_MUTEX_INITIALIZER;
 static node_state_t    state       = PASSIVE_STATE;
 static int             parent      = -1; // -1 indicates null/no parent
 static int             deficit     = 0;  // C_i counter (unacknowledged messages)
 static pthread_mutex_t state_mutex = PTHREAD_MUTEX_INITIALIZER;
 static node_state_t    state       = PASSIVE_STATE;
 static int             parent      = -1; // -1 indicates null/no parent
 static int             deficit     = 0;  // C_i counter (unacknowledged messages)

-static bool            initiator = false;
+static bool            is_initiator = false;
+static bool            started      = false; // RACE CONDITION FIX

 
 

-// --- HELPER FUNCTION ---
-// MUST BE DECLARED ABOVE THE OTHER FUNCTIONS!
-// Evaluates if the node can detach from the tree. Caller must hold state_mutex.
-static void try_resolve_tree(int my_id)
-{
-    if (state == PASSIVE_STATE && deficit == 0)
-    {
-        if (initiator) {
+
+// HELPER FUNCTION - Evaluates if the node can detach from the tree. Caller must hold state_mutex.
+static void try_resolve_tree(int my_id){
+    if (started && state == PASSIVE_STATE && deficit == 0){
+        if(is_initiator){

             // Root is passive and deficit is 0 -> Global Termination!
             return; 
         }
 
             // Root is passive and deficit is 0 -> Global Termination!
             return; 
         }
 

-        if (parent != -1)
-        {
-            control_message_t sig = {0};
+        if(parent != -1){
+            control_message_t sig = { MSG_ACK };

             // Send acknowledgment signal up the tree to our parent
             MPI_Send(&sig, sizeof(control_message_t), MPI_BYTE, parent, 
                      CONTROL_SIGNAL, MPI_COMM_WORLD);
             // Send acknowledgment signal up the tree to our parent
             MPI_Send(&sig, sizeof(control_message_t), MPI_BYTE, parent, 
                      CONTROL_SIGNAL, MPI_COMM_WORLD);


@@ -61,52 +63,74 @@ static void try_resolve_tree(int my_id)
     }
 }
 
     }
 }
 

-/*
-  Main loop of the termination detection algorithm.
-*/
-void *detect_termination(void *_args)
-{
+
+//  Main loop of the termination detection algorithm.
+void *detect_termination(void *_args){

     thread_args_t *args = _args;
     int id = args->id;
     thread_args_t *args = _args;
     int id = args->id;

-    initiator = args->initiator;
-    
-    // (Removed unused 'processes' variable to fix compiler warning)
+    int processes = args->processes; //  MPI interface - needed for the KILL broadcast
+    is_initiator = args->initiator;

 
 

-    while (true)
-    {
+    while(true){

         pthread_mutex_lock(&state_mutex);
         
         pthread_mutex_lock(&state_mutex);
         

-        // CHECK GLOBAL TERMINATION CONDITION
-        if (initiator && state == PASSIVE_STATE && deficit == 0)
-        {
+        // CHECK GLOBAL TERMINATION CONDITION - Only evaluate if the basic algorithm has officially started
+        if(started && is_initiator && state == PASSIVE_STATE && deficit == 0){

             trace("%d: [CONTROL] GLOBAL TERMINATION DETECTED!\n", id);
             trace("%d: [CONTROL] GLOBAL TERMINATION DETECTED!\n", id);

+            
+            // Broadcast KILL signal to unblock all other control threads - use 
+            control_message_t kill_sig = {MSG_KILL};
+            for(int p = 0; p < processes; p++){
+                if(p != id){
+                    MPI_Send(
+                        &kill_sig,                      // pointer to the control_message_t struct.
+                        sizeof(control_message_t),      // the size of the message.
+                        MPI_BYTE,                       // sending raw memory bytes.
+                        p,                              // destinnationn, parent node.
+                        CONTROL_SIGNAL,                 // label 99.
+                        MPI_COMM_WORLD                  // the overall group of processes this message is restricted to.
+                    );
+                }
+            }
+            

             pthread_mutex_unlock(&state_mutex);
             break; // Breaks loop, tells main() to shut down
         }
         
         pthread_mutex_unlock(&state_mutex);
 
             pthread_mutex_unlock(&state_mutex);
             break; // Breaks loop, tells main() to shut down
         }
         
         pthread_mutex_unlock(&state_mutex);
 

-        // Check network for incoming child signals (non-blocking)
+        // Check network for incoming child signals (non-blocking) - use MPI_Recv to read messages sent over the network by other processes

         int flag = 0;
         MPI_Status status;
         MPI_Iprobe(MPI_ANY_SOURCE, CONTROL_SIGNAL, MPI_COMM_WORLD, &flag, &status);
 
         int flag = 0;
         MPI_Status status;
         MPI_Iprobe(MPI_ANY_SOURCE, CONTROL_SIGNAL, MPI_COMM_WORLD, &flag, &status);
 

-        if (flag)
-        {
+        if(flag){

             control_message_t sig;
             control_message_t sig;

-            MPI_Recv(&sig, sizeof(control_message_t), MPI_BYTE, status.MPI_SOURCE, 
-                     CONTROL_SIGNAL, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
-
-            pthread_mutex_lock(&state_mutex);
-            deficit--; // Rule D: A child finished its work and reported back
-            trace("%d: [CONTROL] Got signal from %d (New Deficit: %d)\n", 
-                  id, status.MPI_SOURCE, deficit);
-            
-            try_resolve_tree(id);
-            pthread_mutex_unlock(&state_mutex);
+            MPI_Recv(
+                &sig,                                   // pointer to the memory space where the incoming message will be stored
+                sizeof(control_message_t),              // the maximum size (in bytes) of the data willing to receive.
+                MPI_BYTE,                               // raw stream of bytes
+                status.MPI_SOURCE,                      // the ID (rank) of the process you want to receive from MPI_Iprobe.
+                CONTROL_SIGNAL,                         // label of the message
+                MPI_COMM_WORLD,                         // the overall group of processes this message belongs to.
+                MPI_STATUS_IGNORE                       // just to save memory and processing time.
+            );
+
+            if(sig.type == MSG_KILL){
+                trace("%d: [CONTROL] Received KILL signal. Shutting down.\n", id);
+                break;
+            }
+            else if(sig.type == MSG_ACK){
+                pthread_mutex_lock(&state_mutex);
+                deficit--; // Rule D: A child finished its work and reported back
+                trace("%d: [CONTROL] Got signal from %d (New Deficit: %d)\n", 
+                      id, status.MPI_SOURCE, deficit);
+                
+                try_resolve_tree(id);
+                pthread_mutex_unlock(&state_mutex);
+            }

         }
         }

-        else
-        {
+        else{

             // Sleep briefly to prevent pinning the CPU core at 100% usage
             usleep(1000); 
         }
             // Sleep briefly to prevent pinning the CPU core at 100% usage
             usleep(1000); 
         }


@@ -121,9 +145,11 @@ void *detect_termination(void *_args)
 void control_basic_start_hook(int id)
 {
     pthread_mutex_lock(&state_mutex);
 void control_basic_start_hook(int id)
 {
     pthread_mutex_lock(&state_mutex);

-    state   = initiator ? ACTIVE_STATE : PASSIVE_STATE;
+    is_initiator = (id == 0);
+    state   = is_initiator ? ACTIVE_STATE : PASSIVE_STATE;

     parent  = -1;
     deficit = 0;
     parent  = -1;
     deficit = 0;

+    started = true; // Signal the control thread that it is safe to check for termination

     pthread_mutex_unlock(&state_mutex);
 }
 
     pthread_mutex_unlock(&state_mutex);
 }
 


@@ -167,21 +193,19 @@ void control_basic_receive_hook(int id, int peer)
 {
     pthread_mutex_lock(&state_mutex);
     
 {
     pthread_mutex_lock(&state_mutex);
     

-    if (parent == -1)
-    {
-        // Rule B (Case 1): Woken up from passive state
-        state  = ACTIVE_STATE;
-        parent = peer; // Adopt sender as parent in the tree
+    // A node is only "out of the tree" if parent == -1.
+    // It can be PASSIVE but still in the tree if it is waiting for children (deficit > 0).
+    if(parent == -1 && !is_initiator){
+        // Not in the tree -> Join it
+        parent = peer; 

         trace("%d: [CONTROL] Joined tree under parent %d\n", id, parent);
         trace("%d: [CONTROL] Joined tree under parent %d\n", id, parent);

-    }
-    else
-    {
-        // Rule B (Case 2): Already active! Reject parent change, immediately signal back
-        control_message_t sig = {0};
+    }else{
+        // Already in the tree (or is root) -> Reject parent change, immediately signal back
+        control_message_t sig = { MSG_ACK };

         MPI_Send(&sig, sizeof(control_message_t), MPI_BYTE, peer, 
                  CONTROL_SIGNAL, MPI_COMM_WORLD);
         MPI_Send(&sig, sizeof(control_message_t), MPI_BYTE, peer, 
                  CONTROL_SIGNAL, MPI_COMM_WORLD);

-        trace("%d: [CONTROL] Rejected parent %d (Already Active)\n", id, peer);
+        trace("%d: [CONTROL] Rejected parent %d (Already in tree or Root)\n", id, peer);

     }
     
     pthread_mutex_unlock(&state_mutex);
     }
     
     pthread_mutex_unlock(&state_mutex);

-}
+}
\ No newline at end of file