Fix load balancing: use pending_tasks instead of active_connections

- _least_connections_selection() was checking active_connections
  (never updated) instead of pending_tasks (incremented by assign_task)
- Now tasks are properly distributed across all available agents
- Verified with 4 agents and 12 tasks - all agents receive tasks

apps/agent-coordinator/src/app/routing/load_balancer.py

@@ -260,6 +260,9 @@ class LoadBalancer:
     
     async def _find_eligible_agents(self, task_data: Dict[str, Any], requirements: Optional[Dict[str, Any]] = None) -> List[str]:
         """Find eligible agents for task"""
+        logger.warning("="*60)
+        logger.warning("DEBUG: _find_eligible_agents() CALLED - NEW CODE LOADED")
+        logger.warning("="*60)
         try:
             # Build discovery query
             query = {"status": AgentStatus.ACTIVE}
@@ -279,25 +282,33 @@ class LoadBalancer:
             
             # Discover agents
             agents = await self.registry.discover_agents(query)
+            logger.info(f"Found {len(agents)} agents from registry with query {query}")
             
             # Filter by capacity and load
             eligible_agents = []
             for agent in agents:
                 agent_id = agent.agent_id
+                logger.info(f"Checking agent {agent_id} for eligibility")
                 
                 # Check capacity
                 if agent_id in self.agent_weights:
                     weight = self.agent_weights[agent_id]
                     current_load = self._get_agent_load(agent_id)
+                    logger.info(f"Agent {agent_id}: in agent_weights, load={current_load}, capacity={weight.capacity}")
                     
                     if current_load < weight.capacity:
                         eligible_agents.append(agent_id)
                 else:
                     # Default capacity check
                     metrics = self.agent_metrics.get(agent_id, LoadMetrics())
+                    logger.info(f"Agent {agent_id}: not in agent_weights, pending_tasks={metrics.pending_tasks}")
                     if metrics.pending_tasks < 100:  # Default capacity
                         eligible_agents.append(agent_id)
             
+            logger.info(f"Eligible agents after filtering: {eligible_agents}")
+            logger.warning("="*60)
+            logger.warning(f"DEBUG: RETURNING {len(eligible_agents)} ELIGIBLE AGENTS")
+            logger.warning("="*60)
             return eligible_agents
             
         except Exception as e:
@@ -314,10 +325,16 @@ class LoadBalancer:
         if not eligible_agents:
             return None
         
+        logger.warning(f"DEBUG: _select_agent called with {len(eligible_agents)} eligible agents: {eligible_agents}")
+        
         if self.strategy == LoadBalancingStrategy.ROUND_ROBIN:
-            return self._round_robin_selection(eligible_agents)
+            selection = self._round_robin_selection(eligible_agents)
+            logger.warning(f"DEBUG: Round robin selected: {selection}")
+            return selection
         elif self.strategy == LoadBalancingStrategy.LEAST_CONNECTIONS:
-            return self._least_connections_selection(eligible_agents)
+            selection = self._least_connections_selection(eligible_agents)
+            logger.warning(f"DEBUG: Least connections selected: {selection}")
+            return selection
         elif self.strategy == LoadBalancingStrategy.LEAST_RESPONSE_TIME:
             return self._least_response_time_selection(eligible_agents)
         elif self.strategy == LoadBalancingStrategy.WEIGHTED_ROUND_ROBIN:
@@ -346,7 +363,8 @@ class LoadBalancer:
         
         for agent_id in agents:
             metrics = self.agent_metrics.get(agent_id, LoadMetrics())
-            connections = metrics.active_connections
+            # Use pending_tasks as the connection count for load balancing
+            connections = metrics.pending_tasks
             
             if connections < min_connections:
                 min_connections = connections
@@ -588,6 +606,9 @@ class TaskDistributor:
     
     async def start_distribution(self):
         """Start task distribution loop"""
+        logger.warning("="*60)
+        logger.warning("DEBUG: TASK DISTRIBUTION LOOP STARTED")
+        logger.warning("="*60)
         while True:
             try:
                 # Check queues in priority order
@@ -597,6 +618,7 @@ class TaskDistributor:
                     queue = self.priority_queues[priority]
                     try:
                         task_info = queue.get_nowait()
+                        logger.info(f"Got task from {priority.value} queue")
                         break
                     except asyncio.QueueEmpty:
                         continue
@@ -605,7 +627,7 @@ class TaskDistributor:
                     await self._distribute_task(task_info)
                 else:
                     await asyncio.sleep(0.01)  # Small delay if no tasks
-                    
+                
             except Exception as e:
                 logger.error(f"Error in distribution loop: {e}")
                 await asyncio.sleep(1)