puda-comms 0.0.2__py3-none-any.whl → 0.0.4__py3-none-any.whl

puda_comms/command_service.py

@@ -10,14 +10,13 @@ This service handles:
 import asyncio
 import json
 import logging
-import os
 import signal
 from datetime import datetime, timezone
-from typing import Dict, Any, Optional, Tuple
+from typing import Dict, Any, Optional
 import nats
 from nats.js.client import JetStreamContext
 from nats.aio.msg import Msg
-from puda_comms.models import CommandRequest, CommandResponse, CommandResponseStatus, NATSMessage, MessageHeader, MessageType
+from puda_comms.models import CommandRequest, CommandResponseStatus, NATSMessage, MessageHeader, MessageType
 
 logger = logging.getLogger(__name__)
 
@@ -38,7 +37,7 @@ class ResponseHandler:
     def __init__(self, js: JetStreamContext, machine_id: str):
         self.js = js
         self.machine_id = machine_id
-        self._pending_responses: Dict[str, Tuple[asyncio.Event, CommandResponse]] = {}
+        self._pending_responses: Dict[str, Dict[str, Any]] = {}  # {'event': asyncio.Event, 'response': Optional[NATSMessage]}
         self._queue_consumer = None
         self._immediate_consumer = None
         self._initialized = False
@@ -103,8 +102,8 @@ class ResponseHandler:
                 
                 # Get the pending response
                 pending = self._pending_responses[key]
-                # Store the full NATSMessage JSON structure
-                pending['response'] = message.model_dump()
+                # Store the NATSMessage directly
+                pending['response'] = message
                 # Signal that response was received
                 # Don't delete here - let get_response() delete it after retrieval
                 pending['event'].set()
@@ -153,7 +152,7 @@ class ResponseHandler:
         }
         return event
     
-    def get_response(self, run_id: str, step_number: int) -> Optional[Dict[str, Any]]:
+    def get_response(self, run_id: str, step_number: int) -> Optional[NATSMessage]:
         """
         Get the response for a pending command.
         
@@ -162,7 +161,7 @@ class ResponseHandler:
             step_number: Step number for the command
         
         Returns:
-            The NATSMessage dict structure if available, None otherwise
+            The NATSMessage if available, None otherwise
         """
         key = f"{run_id}:{str(step_number)}"
         if key in self._pending_responses:
@@ -178,8 +177,16 @@ class ResponseHandler:
         if key in self._pending_responses:
             del self._pending_responses[key]
     
+    def cancel_all_pending(self):
+        """Cancel all pending responses by setting their events. This wakes up any waiting tasks immediately."""
+        for pending in self._pending_responses.values():
+            pending['event'].set()
+    
     async def cleanup(self):
         """Clean up subscriptions."""
+        # Cancel all pending responses first to wake up waiting tasks
+        self.cancel_all_pending()
+        
         if self._queue_consumer:
             try:
                 await self._queue_consumer.unsubscribe()
@@ -200,11 +207,6 @@ class CommandService:
     Handles connection management, command parsing, and response handling.
     Can send commands to multiple machines.
     
-    Supports async context manager usage for automatic cleanup:
-        async with CommandService() as service:
-            await service.send_queue_command(...)
-        # Automatically disconnects on exit
-    
     Automatically registers signal handlers (SIGTERM, SIGINT) for graceful shutdown.
     """
     
@@ -212,20 +214,19 @@ class CommandService:
     
     def __init__(
         self,
-        servers: Optional[list[str]] = None
+        servers: list[str]
     ):
         """
         Initialize NATS service.
         
         Args:
-            servers: List of NATS server URLs. If None, reads from NATS_SERVERS env var.
+            servers: List of NATS server URLs. Must be a non-empty list.
+        
+        Raises:
+            ValueError: If servers is None or empty.
         """
-        if servers is None:
-            nats_servers_env = os.getenv(
-                "NATS_SERVERS",
-                "nats://192.168.50.201:4222,nats://192.168.50.201:4223,nats://192.168.50.201:4224"
-            )
-            servers = [s.strip() for s in nats_servers_env.split(",")]
+        if servers is None or len(servers) == 0:
+            raise ValueError("Please provide a non-empty list of NATS server URLs")
         
         self.servers = servers
         self.nc: Optional[nats.NATS] = None
@@ -254,24 +255,50 @@ class CommandService:
         """
         Connect to NATS servers.
         
+        Limits connection attempts to 3. After 3 failed attempts, gives up and logs error.
+        
         Returns:
             True if connected successfully, False otherwise
         """
         if self._connected:
             return True
         
-        try:
-            self.nc = await nats.connect(servers=self.servers)
-            self.js = self.nc.jetstream()
-            
-            self._connected = True
-            logger.info("Connected to NATS servers: %s", self.servers)
-            return True
-            
-        except Exception as e:
-            logger.error("Failed to connect to NATS: %s", e)
-            self._connected = False
-            return False
+        max_attempts = 3
+        connect_timeout = 3  # 3 seconds timeout per connection attempt
+        
+        for attempt in range(1, max_attempts + 1):
+            try:
+                logger.info("Connection attempt %d/%d to NATS servers: %s", attempt, max_attempts, self.servers)
+                self.nc = await asyncio.wait_for(
+                    nats.connect(
+                        servers=self.servers,
+                        connect_timeout=connect_timeout,
+                        reconnect_time_wait=2,
+                        max_reconnect_attempts=0  # No reconnection during initial connection
+                    ),
+                    timeout=connect_timeout + 1  # Slightly longer timeout for the wait_for
+                )
+                self.js = self.nc.jetstream()
+                
+                self._connected = True
+                logger.info("Connected to NATS servers")
+                return True
+                
+            except asyncio.TimeoutError:
+                logger.warning("Connection attempt %d/%d timed out after %d seconds", attempt, max_attempts, connect_timeout)
+                if attempt < max_attempts:
+                    logger.info("Retrying connection...")
+                else:
+                    logger.error("Failed to connect after %d attempts. Giving up.", max_attempts)
+            except Exception as e:
+                logger.warning("Connection attempt %d/%d failed: %s", attempt, max_attempts, e)
+                if attempt < max_attempts:
+                    logger.info("Retrying connection...")
+                else:
+                    logger.error("Failed to connect after %d attempts. Giving up.", max_attempts)
+        
+        self._connected = False
+        return False
     
     async def _get_response_handler(self, machine_id: str) -> ResponseHandler:
         """
@@ -316,6 +343,8 @@ class CommandService:
         request: CommandRequest,
         machine_id: str,
         run_id: str,
+        user_id: str,
+        username: str,
         timeout: int = 120
     ) -> Optional[NATSMessage]:
         """
@@ -325,6 +354,8 @@ class CommandService:
             request: CommandRequest model containing command details
             machine_id: Machine ID to send the command to
             run_id: Run ID for the command
+            user_id: User ID who initiated the command
+            username: Username who initiated the command
             timeout: Maximum time to wait for response in seconds
         
         Returns:
@@ -337,8 +368,8 @@ class CommandService:
         subject = f"{NAMESPACE}.{machine_id}.cmd.queue"
         
         logger.info(
-            "Sending queue command: machine_id=%s, command=%s, run_id=%s, step_number=%s",
-            machine_id, request.name, run_id, request.step_number
+            "Sending queue command: subject=%s, command=%s, run_id=%s, step_number=%s",
+            subject, request.name, run_id, request.step_number
         )
         
         # Get or create response handler for this machine
@@ -347,7 +378,7 @@ class CommandService:
         response_event = response_handler.register_pending(run_id, request.step_number)
         
         # Build payload
-        payload = self._build_command_payload(request, machine_id, run_id)
+        payload = self._build_command_payload(request, machine_id, run_id, user_id, username)
 
         try:
             # Publish to JetStream
@@ -370,11 +401,7 @@ class CommandService:
             await asyncio.sleep(0.1)
             
             # Get the response
-            response_data = response_handler.get_response(run_id, request.step_number)
-            if response_data is None:
-                return None
-            
-            return NATSMessage.model_validate(response_data)
+            return response_handler.get_response(run_id, request.step_number)
 
         except Exception as e:
             logger.error("Error sending queue command: %s", e)
@@ -387,6 +414,8 @@ class CommandService:
         requests: list[CommandRequest],
         machine_id: str,
         run_id: str,
+        user_id: str,
+        username: str,
         timeout: int = 120
     ) -> Optional[NATSMessage]:
         """
@@ -400,6 +429,8 @@ class CommandService:
             requests: List of CommandRequest models to send sequentially
             machine_id: Machine ID to send the commands to
             run_id: Run ID for all commands
+            user_id: User ID who initiated the commands
+            username: Username who initiated the commands
             timeout: Maximum time to wait for each response in seconds
         
         Returns:
@@ -435,6 +466,8 @@ class CommandService:
                 request=request,
                 machine_id=machine_id,
                 run_id=run_id,
+                user_id=user_id,
+                username=username,
                 timeout=timeout
             )
             
@@ -495,6 +528,8 @@ class CommandService:
         request: CommandRequest,
         machine_id: str,
         run_id: str,
+        user_id: str,
+        username: str,
         timeout: int = 120
     ) -> Optional[NATSMessage]:
         """
@@ -504,6 +539,8 @@ class CommandService:
             request: CommandRequest model containing command details
             machine_id: Machine ID to send the command to
             run_id: Run ID for the command
+            user_id: User ID who initiated the command
+            username: Username who initiated the command
             timeout: Maximum time to wait for response in seconds
         
         Returns:
@@ -528,7 +565,7 @@ class CommandService:
         response_received = response_handler.register_pending(run_id, request.step_number)
         
         # Build payload
-        payload = self._build_command_payload(request, machine_id, run_id)
+        payload = self._build_command_payload(request, machine_id, run_id, user_id, username)
 
         try:
             # Publish to JetStream
@@ -551,11 +588,7 @@ class CommandService:
             await asyncio.sleep(0.1)
             
             # Get the response
-            response_data = response_handler.get_response(run_id, request.step_number)
-            if response_data is None:
-                return None
-            
-            return NATSMessage.model_validate(response_data)
+            return response_handler.get_response(run_id, request.step_number)
             
         except Exception as e:
             logger.error("Error sending immediate command: %s", e)
@@ -608,7 +641,9 @@ class CommandService:
         self,
         command_request: CommandRequest,
         machine_id: str,
-        run_id: str
+        run_id: str,
+        user_id: str,
+        username: str
     ) -> NATSMessage:
         """
         Build a command payload in the expected format.
@@ -617,6 +652,8 @@ class CommandService:
             command_request: CommandRequest model containing command details
             machine_id: Machine ID for the command
             run_id: Run ID for the command
+            user_id: User ID who initiated the command
+            username: Username who initiated the command
         
         Returns:
             NATSMessage object ready for NATS transmission
@@ -625,6 +662,8 @@ class CommandService:
             message_type=MessageType.COMMAND,
             version="1.0",
             timestamp=datetime.now(timezone.utc).strftime('%Y-%m-%dT%H:%M:%SZ'),
+            user_id=user_id,
+            username=username,
             machine_id=machine_id,
             run_id=run_id
         )

puda_comms/machine_client.py

@@ -20,7 +20,7 @@ from puda_comms.models import (
     ImmediateCommand,
 )
 from nats.js.client import JetStreamContext
-from nats.js.api import StreamConfig
+from nats.js.api import StreamConfig, ConsumerConfig
 from nats.js.errors import NotFoundError
 from nats.aio.msg import Msg
 
@@ -69,11 +69,13 @@ class MachineClient:
         
         # Default subscriptions
         self._cmd_queue_sub = None
+        self._cmd_queue_task = None  # Background task for pull consumer
         self._cmd_immediate_sub = None
         
         # Connection state
         self._is_connected = False
-        self._reconnect_handlers = []
+        self._queue_handler = None
+        self._immediate_handler = None
         
         # Queue control state
         self._pause_lock = asyncio.Lock()
@@ -184,30 +186,22 @@ class MachineClient:
             logger.error("Error ensuring %s stream: %s", stream_name, e, exc_info=True)
             raise
     
-    async def _ensure_command_queue_stream(self):
-        """Ensure COMMAND_QUEUE stream exists with WorkQueue retention policy."""
+    async def _ensure_all_streams(self):
+        """Ensure all required streams exist with correct retention policies."""
         await self._ensure_stream(
             self.STREAM_COMMAND_QUEUE,
-            f"{self.NAMESPACE}.*.cmd.queue"
+            f"{self.NAMESPACE}.*.cmd.queue",
+            retention='workqueue'
         )
-    
-    async def _ensure_command_immediate_stream(self):
-        """Ensure COMMAND_IMMEDIATE stream exists with WorkQueue retention policy."""
         await self._ensure_stream(
             self.STREAM_COMMAND_IMMEDIATE,
             f"{self.NAMESPACE}.*.cmd.immediate"
         )
-    
-    async def _ensure_response_queue_stream(self):
-        """Ensure RESPONSE_QUEUE stream exists with Interest retention policy."""
         await self._ensure_stream(
             self.STREAM_RESPONSE_QUEUE,
             f"{self.NAMESPACE}.*.cmd.response.queue",
             retention='interest'
         )
-    
-    async def _ensure_response_immediate_stream(self):
-        """Ensure RESPONSE_IMMEDIATE stream exists with Interest retention policy."""
         await self._ensure_stream(
             self.STREAM_RESPONSE_IMMEDIATE,
             f"{self.NAMESPACE}.*.cmd.response.immediate",
@@ -230,7 +224,17 @@ class MachineClient:
     
     async def _cleanup_subscriptions(self):
         """Unsubscribe from all subscriptions."""
-        # Clean up subscriptions
+        # Clean up queue subscription (pull consumer)
+        if self._cmd_queue_task:
+            try:
+                self._cmd_queue_task.cancel()
+                await self._cmd_queue_task
+            except asyncio.CancelledError:
+                pass
+            except Exception:
+                pass
+            self._cmd_queue_task = None
+        
         if self._cmd_queue_sub:
             try:
                 await self._cmd_queue_sub.unsubscribe()
@@ -252,6 +256,7 @@ class MachineClient:
         self.kv = None
         # Subscriptions will be recreated on reconnection
         self._cmd_queue_sub = None
+        self._cmd_queue_task = None
         self._cmd_immediate_sub = None
     
     # ==================== CONNECTION MANAGEMENT ====================
@@ -261,6 +266,7 @@ class MachineClient:
         try:
             self.nc = await nats.connect(
                 servers=self.servers,
+                connect_timeout=10,  # 10 seconds timeout for initial connection
                 reconnect_time_wait=2,
                 max_reconnect_attempts=-1,
                 error_cb=self._error_callback,
@@ -269,10 +275,7 @@ class MachineClient:
                 closed_cb=self._closed_callback
             )
             self.js = self.nc.jetstream()
-            await self._ensure_command_queue_stream()
-            await self._ensure_command_immediate_stream()
-            await self._ensure_response_queue_stream()
-            await self._ensure_response_immediate_stream()
+            await self._ensure_all_streams()
             self.kv = await self._get_or_create_kv_bucket()
             self._is_connected = True
             logger.info("Connected to NATS servers: %s", self.servers)
@@ -298,32 +301,16 @@ class MachineClient:
         
         if self.nc:
             self.js = self.nc.jetstream()
-            await self._ensure_command_queue_stream()
-            await self._ensure_command_immediate_stream()
-            await self._ensure_response_queue_stream()
-            await self._ensure_response_immediate_stream()
+            await self._ensure_all_streams()
             self.kv = await self._get_or_create_kv_bucket()
             await self._resubscribe_handlers()
     
     async def _resubscribe_handlers(self):
         """Re-subscribe to all handlers after reconnection."""
-        subscribe_methods = {
-            'queue': self.subscribe_queue,
-            'immediate': self.subscribe_immediate,
-        }
-        
-        for handler_info in self._reconnect_handlers:
-            try:
-                handler_type = handler_info['type']
-                handler = handler_info['handler']
-                subscribe_method = subscribe_methods.get(handler_type)
-                
-                if subscribe_method:
-                    await subscribe_method(handler)
-                else:
-                    logger.warning("Unknown handler type: %s", handler_type)
-            except Exception as e:
-                logger.error("Failed to re-subscribe %s: %s", handler_type, e)
+        if self._queue_handler:
+            await self.subscribe_queue(self._queue_handler)
+        if self._immediate_handler:
+            await self.subscribe_immediate(self._immediate_handler)
     
     async def _closed_callback(self):
         """Callback when connection is closed."""
@@ -438,7 +425,7 @@ class MachineClient:
     async def process_queue_cmd(
         self, 
         msg: Msg,
-        handler: Callable[[CommandRequest], Awaitable[CommandResponse]]
+        handler: Callable[[NATSMessage], Awaitable[CommandResponse]]
     ) -> None:
         """
         Handle the lifecycle of a single message: Parse -> Handle -> Ack/Nak/Term.
@@ -660,9 +647,54 @@ class MachineClient:
             )
             await self.publish_state({'state': 'error', 'run_id': None})
     
+    async def _verify_or_recreate_consumer(self, durable_name: str):
+        """
+        Check if consumer exists and verify/update its configuration.
+        Deletes and recreates the consumer if configuration doesn't match.
+        
+        Args:
+            durable_name: Name of the durable consumer to verify
+        """
+        # Check if consumer exists and verify/update its configuration
+        try:
+            consumer_info = await self.js.consumer_info(self.STREAM_COMMAND_QUEUE, durable_name)
+            logger.debug("Durable consumer %s already exists", durable_name)
+            
+            # Check if consumer config matches what we need
+            config = consumer_info.config
+            needs_recreate = False
+            if getattr(config, 'filter_subject', None) != self.cmd_queue:
+                logger.warning("Consumer filter_subject mismatch: expected %s, got %s", 
+                             self.cmd_queue, getattr(config, 'filter_subject', None))
+                needs_recreate = True
+            if getattr(config, 'ack_policy', None) != 'explicit':
+                logger.warning("Consumer ack_policy mismatch: expected explicit, got %s", 
+                             getattr(config, 'ack_policy', None))
+                needs_recreate = True
+            if getattr(config, 'deliver_policy', None) != 'all':
+                logger.warning("Consumer deliver_policy mismatch: expected all, got %s", 
+                             getattr(config, 'deliver_policy', None))
+                needs_recreate = True
+            
+            if needs_recreate:
+                # Consumer exists but config doesn't match - delete and recreate
+                logger.info("Consumer config mismatch, deleting and recreating: %s", durable_name)
+                try:
+                    await self.js.delete_consumer(self.STREAM_COMMAND_QUEUE, durable_name)
+                except Exception as e:
+                    logger.warning("Error deleting consumer: %s", e)
+            else:
+                # Log consumer state for diagnostics
+                logger.info("Consumer exists with correct config - pending: %d, delivered: %d, ack_pending: %d",
+                           consumer_info.num_pending, consumer_info.delivered.consumer_seq,
+                           consumer_info.num_ack_pending)
+        except NotFoundError:
+            # Consumer doesn't exist, will be created by pull_subscribe
+            logger.debug("Durable consumer %s does not exist, will be created", durable_name)
+    
     async def subscribe_queue(self, handler: Callable[[NATSMessage], Awaitable[CommandResponse]]):
         """
-        Subscribe to queue commands with default consumer.
+        Subscribe to queue commands with pull consumer.
         
         Args:
             handler: Async function that processes command payloads and returns CommandResponse
@@ -672,19 +704,65 @@ class MachineClient:
             return
         
         # Ensure stream exists before attempting to subscribe
-        await self._ensure_command_queue_stream()
+        await self._ensure_all_streams()
         
         try:
-            async def message_handler(msg: Msg):
-                """Wrapper to process queue messages."""
-                await self.process_queue_cmd(msg, handler)
-
-            self._cmd_queue_sub = await self.js.subscribe(
+            durable_name = f"cmd_queue_{self.machine_id}"
+            
+            await self._verify_or_recreate_consumer(durable_name)
+            
+            # Create pull subscription - this will create the consumer if it doesn't exist
+            # Pass config directly to ensure correct consumer configuration
+            consumer_config = ConsumerConfig(
+                durable_name=durable_name,
+                filter_subject=self.cmd_queue,
+                ack_policy="explicit",
+                deliver_policy="all",  # Required for WorkQueue: deliver all messages from the beginning
+            )
+            
+            self._cmd_queue_sub = await self.js.pull_subscribe(
                 subject=self.cmd_queue,
+                durable=durable_name,
                 stream=self.STREAM_COMMAND_QUEUE,
-                durable=f"cmd_queue_{self.machine_id}",
-                cb=message_handler
+                config=consumer_config
             )
+            
+            # Log final consumer info for diagnostics
+            try:
+                consumer_info = await self.js.consumer_info(self.STREAM_COMMAND_QUEUE, durable_name)
+                logger.info("Pull subscription created - subject: %s, durable: %s, stream: %s, pending: %d, ack_pending: %d",
+                           self.cmd_queue, durable_name, self.STREAM_COMMAND_QUEUE,
+                           consumer_info.num_pending, consumer_info.num_ack_pending)
+            except Exception as e:
+                logger.warning("Could not get consumer info after subscription: %s", e)
+                logger.info("Pull subscription created - subject: %s, durable: %s, stream: %s",
+                           self.cmd_queue, durable_name, self.STREAM_COMMAND_QUEUE)
+            
+            # Start background task to pull and process messages
+            async def pull_messages():
+                """Continuously pull messages from the queue."""
+                try:
+                    while True:
+                        try:
+                            # Fetch messages (batch of 1, timeout 1 second)
+                            msgs = await self._cmd_queue_sub.fetch(batch=1, timeout=1.0)
+                            if msgs:
+                                logger.debug("Pulled %d message(s) from queue", len(msgs))
+                            for msg in msgs:
+                                await self.process_queue_cmd(msg, handler)
+                        except asyncio.TimeoutError:
+                            # Timeout is expected when no messages are available
+                            continue
+                        except Exception as e:
+                            logger.error("Error pulling queue messages: %s", e, exc_info=True)
+                            await asyncio.sleep(1)  # Wait before retrying
+                except asyncio.CancelledError:
+                    logger.debug("Queue pull task cancelled")
+                    raise
+            
+            self._cmd_queue_task = asyncio.create_task(pull_messages())
+            logger.info("Started background task for pulling queue messages")
+            
         except NotFoundError:
             # Stream still not found after ensuring it exists - this shouldn't happen
             # but handle it gracefully with detailed diagnostics
@@ -702,10 +780,9 @@ class MachineClient:
                 logger.error("  Stream verification failed: %s", stream_check_error)
             raise
         
-        # Register handler for reconnection
-        if not any(h['type'] == 'queue' for h in self._reconnect_handlers):
-            self._reconnect_handlers.append({'type': 'queue', 'handler': handler})
-        logger.info("Subscribed to queue commands: %s (durable: cmd_queue_%s, stream: %s)", 
+        # Store handler for reconnection
+        self._queue_handler = handler
+        logger.info("Subscribed to queue commands: %s (durable: cmd_queue_%s, stream: %s, pull consumer)", 
                    self.cmd_queue, self.machine_id, self.STREAM_COMMAND_QUEUE)
     
     async def subscribe_immediate(self, handler: Callable[[NATSMessage], Awaitable[CommandResponse]]):
@@ -719,19 +796,26 @@ class MachineClient:
             logger.error("JetStream not available for immediate subscription")
             return
         
+        # Store handler for use in callback and reconnection
+        self._immediate_handler = handler
+        
         async def message_handler(msg: Msg):
-            """Wrapper to process immediate messages."""
-            await self.process_immediate_cmd(msg, handler)
+            """Process immediate messages using stored handler."""
+            await self.process_immediate_cmd(msg, self._immediate_handler)
         
         # Ensure stream exists before attempting to subscribe
-        await self._ensure_command_immediate_stream()
+        await self._ensure_stream(
+            self.STREAM_COMMAND_IMMEDIATE,
+            f"{self.NAMESPACE}.*.cmd.immediate",
+            retention='workqueue'
+        )
         
         try:
             self._cmd_immediate_sub = await self.js.subscribe(
                 subject=self.cmd_immediate,
                 stream=self.STREAM_COMMAND_IMMEDIATE,
                 durable=f"cmd_immed_{self.machine_id}",
-                cb=message_handler
+                cb=message_handler  # required for push consumer to handle messages
             )
         except NotFoundError:
             # Stream still not found after ensuring it exists - this shouldn't happen
@@ -740,9 +824,6 @@ class MachineClient:
                        self.STREAM_COMMAND_IMMEDIATE)
             raise
         
-        # Register handler for reconnection
-        if not any(h['type'] == 'immediate' for h in self._reconnect_handlers):
-            self._reconnect_handlers.append({'type': 'immediate', 'handler': handler})
         logger.info("Subscribed to immediate commands: %s (durable: cmd_immed_%s, stream: %s)", 
                    self.cmd_immediate, self.machine_id, self.STREAM_COMMAND_IMMEDIATE)
     

puda_comms/models.py

@@ -68,18 +68,19 @@ class CommandResponse(BaseModel):
 
 class MessageHeader(BaseModel):
     """Header for NATS messages."""
-    message_type: MessageType = Field(description="Type of message")
     version: str = Field(default="1.0", description="Message version")
-    timestamp: str = Field(default_factory=_get_current_timestamp, description="ISO format timestamp (auto-set on creation)")
+    message_type: MessageType = Field(description="Type of message")
+    user_id: str = Field(description="User ID")
+    username: str = Field(description="User name")
     machine_id: str = Field(description="Machine ID")
     run_id: Optional[str] = Field(default=None, description="Unique identifier (uuid) for the run/workflow")
-
+    timestamp: str = Field(default_factory=_get_current_timestamp, description="ISO format timestamp (auto-set on creation)")
 class NATSMessage(BaseModel):
     """
     Complete NATS message structure.
     
     Structure:
-    - header: MessageHeader with message_type, version, timestamp, machine_id, run_id
+    - header: MessageHeader with message_type, version, timestamp, user_id, username, machine_id, run_id
     - command: Optional CommandRequest (for command messages)
     - response: Optional CommandResponse data (for response messages)
     """

{puda_comms-0.0.2.dist-info → puda_comms-0.0.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: puda-comms
-Version: 0.0.2
+Version: 0.0.4
 Summary: Communication library for the PUDA platform.
 Author: zhao
 Author-email: zhao <20024592+agentzhao@users.noreply.github.com>
@@ -121,6 +121,8 @@ Header metadata for NATS messages.
 - `message_type` (MessageType): Type of message (COMMAND, RESPONSE, LOG, etc.)
 - `version` (str): Message version (default: "1.0")
 - `timestamp` (str): ISO 8601 UTC timestamp (auto-generated)
+- `user_id` (str): User ID who initiated the command
+- `username` (str): Username who initiated the command
 - `machine_id` (str): Identifier for the target machine
 - `run_id` (Optional[str]): Unique identifier (UUID) for the run/workflow
 
@@ -130,6 +132,8 @@ header = MessageHeader(
     message_type=MessageType.RESPONSE,
     version="1.0",
     timestamp="2026-01-20T02:00:46Z",
+    user_id="user123",
+    username="John Doe",
     machine_id="first",
     run_id="092073e6-13d0-4756-8d99-eff1612a5a72"
 )
@@ -154,6 +158,8 @@ Complete NATS message structure combining header with optional command or respon
     "message_type": "response",
     "version": "1.0",
     "timestamp": "2026-01-20T02:00:46Z",
+    "user_id": "user123",
+    "username": "John Doe",
     "machine_id": "first",
     "run_id": "092073e6-13d0-4756-8d99-eff1612a5a72"
   },
@@ -229,6 +235,8 @@ reply = await service.send_queue_command(
     request=request,
     machine_id="first",
     run_id=run_id,
+    user_id="user123",
+    username="John Doe",
     timeout=60  # Wait up to 60 seconds
 )
 
@@ -237,6 +245,8 @@ reply = await service.send_queue_commands(
     requests=commands,
     machine_id="first",
     run_id=run_id,
+    user_id="user123",
+    username="John Doe",
     timeout=60  # Wait up to 60 seconds per command
 )
 ```
@@ -274,7 +284,9 @@ Always check the response status and handle errors appropriately:
 reply: NATSMessage = await service.send_queue_command(
     request=request,
     machine_id="first",
-    run_id=run_id
+    run_id=run_id,
+    user_id="user123",
+    username="John Doe"
 )
 
 if reply is None:
@@ -292,14 +304,30 @@ else:
 
 ### Configuration
 
-The `CommandService` reads NATS server URLs from the `NATS_SERVERS` environment variable, or defaults to:
+#### NATS Server Configuration
+
+The `CommandService` requires NATS server URLs to be specified explicitly. There are no default values. You must provide servers in one of two ways:
+
+**Option 1: Via environment variable (comma-separated string)**
+
+Set the `NATS_SERVERS` environment variable with comma-separated server URLs:
+
+```bash
+export NATS_SERVERS="nats://192.168.50.201:4222,nats://192.168.50.201:4223,nats://192.168.50.201:4224"
 ```
-nats://192.168.50.201:4222,nats://192.168.50.201:4223,nats://192.168.50.201:4224
+
+Then parse it when creating a `CommandService`:
+```python
+import os
+nats_servers = [s.strip() for s in os.getenv("NATS_SERVERS", "").split(",") if s.strip()]
+service = CommandService(servers=nats_servers)
 ```
 
-You can also specify servers explicitly:
+**Option 2: Directly as a list**
+
+Specify servers directly when creating a `CommandService`:
 ```python
-service = CommandService(servers=["nats://localhost:4222"])
+service = CommandService(servers=["nats://192.168.50.201:4222", "nats://192.168.50.201:4223", "nats://192.168.50.201:4224"])
 ```
 ## Validation
 

puda_comms-0.0.4.dist-info/RECORD

@@ -0,0 +1,8 @@
+puda_comms/__init__.py,sha256=lntvVFJJez_rv5lZy5mYj4_43B9Y3NRNzxWfBuSAQ1M,194
+puda_comms/command_service.py,sha256=KFremcEGfsTeUVQMIhyk1knYmUCvRYQ12vS_jy_14wA,25193
+puda_comms/execution_state.py,sha256=aTaejCnJgg1y_FP-ymIC1GQzqC81FIWo0RZ18XzAQnA,2881
+puda_comms/machine_client.py,sha256=wj6t_QHGs7l1Oc8JQ6hq2hqBd5C14TCPA_dTU9qOLzw,37430
+puda_comms/models.py,sha256=9ZGX0PR7SgMBOL5zVLrPuSUhZqutQU96PubyjyQLhf8,3617
+puda_comms-0.0.4.dist-info/WHEEL,sha256=ZyFSCYkV2BrxH6-HRVRg3R9Fo7MALzer9KiPYqNxSbo,79
+puda_comms-0.0.4.dist-info/METADATA,sha256=0cMHDub_3NZt7Cj5U1jzrQXI8atQqpMM-i3vSMrT5lo,11512
+puda_comms-0.0.4.dist-info/RECORD,,

puda_comms-0.0.2.dist-info/RECORD

@@ -1,8 +0,0 @@
-puda_comms/__init__.py,sha256=lntvVFJJez_rv5lZy5mYj4_43B9Y3NRNzxWfBuSAQ1M,194
-puda_comms/command_service.py,sha256=B4fKiQNF0slvGS1fXVoh5UZax_-xk4IS-KT96teSRfg,23272
-puda_comms/execution_state.py,sha256=aTaejCnJgg1y_FP-ymIC1GQzqC81FIWo0RZ18XzAQnA,2881
-puda_comms/machine_client.py,sha256=F2i0BYBuOLjKAnfZAblNrb3Lzs0yhEO1d4XA-k_dkIU,33039
-puda_comms/models.py,sha256=cVH5uKzyLmjzPeBcm3RIJMTkoynmxqe_P26GtZwlIN8,3500
-puda_comms-0.0.2.dist-info/WHEEL,sha256=ZyFSCYkV2BrxH6-HRVRg3R9Fo7MALzer9KiPYqNxSbo,79
-puda_comms-0.0.2.dist-info/METADATA,sha256=jHHcSSmdWOykobTsieX2bqDeRtqSaqdUd-xZeeWxJZ8,10585
-puda_comms-0.0.2.dist-info/RECORD,,