#include #include #include #include #include #include #include #include #include "shared.h" #define NODES 4 void *load_balance_thread(void *arg) { while (1) { pthread_mutex_lock(&status_array->mutex); float total = 0; float max = -1, min = 1e9; int max_id = -1, min_id = -1; for (int i = 0; i < NODES; i++) { float load = status_array->statuses[i].load; total += load; if (load > max) { max = load; max_id = i; } if (load < min) { min = load; min_id = i; } } float avg = total / NODES; if ((max - min) > (avg / 2)) { printf("[daemon] Triggering migration: max load node %d -> min load node %d\n", max_id, min_id); // Call external script/function to checkpoint, scp, and restart char cmd[512]; snprintf(cmd, sizeof(cmd), "mpirun -np 1 -host node%d checkpointandbalance.sh %d &", min_id, max_id); printf("[daemon] Running on node%d: %s\n", node_id, cmd); system(cmd); char cmd[512]; snprintf(cmd, sizeof(cmd), "ssh node%d \"./migrate_heavy_job.sh %d %d\"", max_id, max_id, min_id); system(cmd); } pthread_mutex_unlock(&status_array->mutex); sleep(60); } return NULL; } void run_job_on_node(const char *binary_path, int node_id) { char cmd[512]; snprintf(cmd, sizeof(cmd), "mpirun -np 1 -host node%d %s &", node_id, binary_path); printf("[daemon] Running on node%d: %s\n", node_id, cmd); system(cmd); } int main() { int shm_fd = shm_open(SHM_NAME, O_CREAT | O_RDWR, 0666); if (shm_fd == -1) { perror("shm_open"); return 1; } ftruncate(shm_fd, sizeof(JobQueue)); JobQueue *queue = mmap(NULL, sizeof(JobQueue), PROT_READ | PROT_WRITE, MAP_SHARED, shm_fd, 0); if (queue == MAP_FAILED) { perror("mmap"); return 1; } int stat_fd = shm_open(NODE_STATUS_SHM, O_RDWR, 0666); if (stat_fd == -1) { perror("shm_open NODE_STATUS_SHM"); return 1; } NodeStatusArray *status_array = mmap(NULL, sizeof(NodeStatusArray), PROT_READ | PROT_WRITE, MAP_SHARED, stat_fd, 0); if (status_array == MAP_FAILED) { perror("mmap NODE_STATUS_SHM"); return 1; } static int initialized = 0; if (!initialized) { pthread_mutexattr_t attr; pthread_mutexattr_init(&attr); pthread_mutexattr_setpshared(&attr, PTHREAD_PROCESS_SHARED); pthread_mutex_init(&queue->mutex, &attr); pthread_mutexattr_destroy(&attr); queue->head = queue->tail = 0; // Initialize job_node array for (int i = 0; i < MAX_JOBS; i++) { queue->job_node[i] = -1; } initialized = 1; } int next_node_id = 0; while (1) { pthread_mutex_lock(&queue->mutex); while (queue->head != queue->tail) { int idx = queue->head; char job[MAX_PATH]; strncpy(job, queue->jobs[idx], MAX_PATH); // Assign node to job queue->job_node[idx] = next_node_id; // Move head forward in the queue queue->head = (queue->head + 1) % MAX_JOBS; pthread_mutex_unlock(&queue->mutex); printf("[daemon] Dispatching job at index %d to node %d: %s\n", idx, next_node_id, job); run_job_on_node(job, next_node_id); next_node_id = (next_node_id + 1) % NODES; pthread_mutex_lock(&queue->mutex); } pthread_mutex_unlock(&queue->mutex); sleep(1); } return 0; }