Merge remote-tracking branch 'origin/master' into docs

README.md

@@ -63,12 +63,14 @@ Everything else is included.
 
 ## Quickstart
 
+There's no need to clone this repo, unless you want to build from source. TrustGraph is a fully containerized app that deploys as a set of Docker containers. To configure TrustGraph on the command line:
+
 ```
 npx @trustgraph/config
 ```
 
-TrustGraph downloads as Docker containers and can be run locally with Docker, Podman, or Minikube. The config tool will generate:
+The config process will generate an app config that can be run locally with Docker, Podman, or Minikube. The process will output:
-- `deploy.zip` with either a `docker-compose.yaml` file for a Docker/Podman deploy or `resources.yaml` for Kubernetes
+- `deploy.zip` with either a `docker-compose.yaml` file for a Docker/Podman or `resources.yaml` for Kubernetes
 - Deployment instructions as `INSTALLATION.md`
 
 <p align="center">
@@ -76,23 +78,7 @@ TrustGraph downloads as Docker containers and can be run locally with Docker, Po
 width="80%" controls></video>
 </p>
 
-For a browser based quickstart, try the [Configuration Terminal](https://config-ui.demo.trustgraph.ai/). 
+For a browser based configuration, try the [Configuration Terminal](https://config-ui.demo.trustgraph.ai/). 
-
-<details>
-<summary>Table of Contents</summary>
-<br>
-
-- [**What is a Context Graph?**](#watch-what-is-a-context-graph)<br>
-- [**Context Graphs in Action**](#watch-context-graphs-in-action)<br>
-- [**Getting Started**](#getting-started-with-trustgraph)<br>
-- [**Context Cores**](#context-cores)<br>
-- [**Tech Stack**](#tech-stack)<br>
-- [**Observability & Telemetry**](#observability--telemetry)<br>
-- [**Contributing**](#contributing)<br>
-- [**License**](#license)<br>
-- [**Support & Community**](#support--community)<br>
-
-</details>
 
 ## Watch What is a Context Graph?
 

tests/unit/test_gateway/test_dispatch_manager.py

@@ -49,7 +49,8 @@ class TestDispatcherManager:
         assert manager.prefix == "api-gateway"  # default prefix
         assert manager.flows == {}
         assert manager.dispatchers == {}
-        
+        assert isinstance(manager.dispatcher_lock, asyncio.Lock)
+
         # Verify manager was added as handler to config receiver
         mock_config_receiver.add_handler.assert_called_once_with(manager)
 
@@ -543,18 +544,99 @@ class TestDispatcherManager:
         mock_backend = Mock()
         mock_config_receiver = Mock()
         manager = DispatcherManager(mock_backend, mock_config_receiver)
-        
+
         # Setup test flow with interface but unsupported kind
         manager.flows["test_flow"] = {
             "interfaces": {
                 "invalid-kind": {"request": "req", "response": "resp"}
             }
         }
-        
+
         with patch('trustgraph.gateway.dispatch.manager.request_response_dispatchers') as mock_rr_dispatchers, \
              patch('trustgraph.gateway.dispatch.manager.sender_dispatchers') as mock_sender_dispatchers:
             mock_rr_dispatchers.__contains__.return_value = False
             mock_sender_dispatchers.__contains__.return_value = False
-            
+
             with pytest.raises(RuntimeError, match="Invalid kind"):
-                await manager.invoke_flow_service("data", "responder", "test_flow", "invalid-kind")
+                await manager.invoke_flow_service("data", "responder", "test_flow", "invalid-kind")
+
+    @pytest.mark.asyncio
+    async def test_invoke_global_service_concurrent_calls_create_single_dispatcher(self):
+        """Concurrent calls for the same service must create exactly one dispatcher.
+
+        Before the fix, await dispatcher.start() yielded to the event loop and
+        multiple coroutines could all pass the 'key not in self.dispatchers' check
+        before any of them wrote the result back, creating duplicate Pulsar consumers.
+        """
+        mock_backend = Mock()
+        mock_config_receiver = Mock()
+        manager = DispatcherManager(mock_backend, mock_config_receiver)
+
+        async def slow_start():
+            # Yield to the event loop so other coroutines get a chance to run,
+            # reproducing the window that caused the original race condition.
+            await asyncio.sleep(0)
+
+        with patch('trustgraph.gateway.dispatch.manager.global_dispatchers') as mock_dispatchers:
+            mock_dispatcher_class = Mock()
+            mock_dispatcher = Mock()
+            mock_dispatcher.start = AsyncMock(side_effect=slow_start)
+            mock_dispatcher.process = AsyncMock(return_value="result")
+            mock_dispatcher_class.return_value = mock_dispatcher
+            mock_dispatchers.__getitem__.return_value = mock_dispatcher_class
+
+            results = await asyncio.gather(*[
+                manager.invoke_global_service("data", "responder", "config")
+                for _ in range(5)
+            ])
+
+        assert mock_dispatcher_class.call_count == 1, (
+            "Dispatcher class instantiated more than once — duplicate consumer bug"
+        )
+        assert mock_dispatcher.start.call_count == 1
+        assert manager.dispatchers[(None, "config")] is mock_dispatcher
+        assert all(r == "result" for r in results)
+
+    @pytest.mark.asyncio
+    async def test_invoke_flow_service_concurrent_calls_create_single_dispatcher(self):
+        """Concurrent calls for the same flow+kind must create exactly one dispatcher.
+
+        invoke_flow_service has the same check-then-create pattern as
+        invoke_global_service and is protected by the same dispatcher_lock.
+        """
+        mock_backend = Mock()
+        mock_config_receiver = Mock()
+        manager = DispatcherManager(mock_backend, mock_config_receiver)
+
+        manager.flows["test_flow"] = {
+            "interfaces": {
+                "agent": {
+                    "request": "agent_request_queue",
+                    "response": "agent_response_queue",
+                }
+            }
+        }
+
+        async def slow_start():
+            await asyncio.sleep(0)
+
+        with patch('trustgraph.gateway.dispatch.manager.request_response_dispatchers') as mock_rr_dispatchers:
+            mock_dispatcher_class = Mock()
+            mock_dispatcher = Mock()
+            mock_dispatcher.start = AsyncMock(side_effect=slow_start)
+            mock_dispatcher.process = AsyncMock(return_value="result")
+            mock_dispatcher_class.return_value = mock_dispatcher
+            mock_rr_dispatchers.__getitem__.return_value = mock_dispatcher_class
+            mock_rr_dispatchers.__contains__.return_value = True
+
+            results = await asyncio.gather(*[
+                manager.invoke_flow_service("data", "responder", "test_flow", "agent")
+                for _ in range(5)
+            ])
+
+        assert mock_dispatcher_class.call_count == 1, (
+            "Dispatcher class instantiated more than once — duplicate consumer bug"
+        )
+        assert mock_dispatcher.start.call_count == 1
+        assert manager.dispatchers[("test_flow", "agent")] is mock_dispatcher
+        assert all(r == "result" for r in results)

trustgraph-flow/trustgraph/gateway/dispatch/manager.py

@@ -116,6 +116,7 @@ class DispatcherManager:
 
         self.flows = {}
         self.dispatchers = {}
+        self.dispatcher_lock = asyncio.Lock()
 
     async def start_flow(self, id, flow):
         logger.info(f"Starting flow {id}")
@@ -163,30 +164,28 @@ class DispatcherManager:
 
         key = (None, kind)
 
-        if key in self.dispatchers:
+        if key not in self.dispatchers:
-            return await self.dispatchers[key].process(data, responder)
+            async with self.dispatcher_lock:
+                if key not in self.dispatchers:
+                    request_queue = None
+                    response_queue = None
+                    if kind in self.queue_overrides:
+                        request_queue = self.queue_overrides[kind].get("request")
+                        response_queue = self.queue_overrides[kind].get("response")
 
-        # Get queue overrides if specified for this service
+                    dispatcher = global_dispatchers[kind](
-        request_queue = None
+                        backend = self.backend,
-        response_queue = None
+                        timeout = 120,
-        if kind in self.queue_overrides:
+                        consumer = f"{self.prefix}-{kind}-request",
-            request_queue = self.queue_overrides[kind].get("request")
+                        subscriber = f"{self.prefix}-{kind}-request",
-            response_queue = self.queue_overrides[kind].get("response")
+                        request_queue = request_queue,
+                        response_queue = response_queue,
+                    )
 
-        dispatcher = global_dispatchers[kind](
+                    await dispatcher.start()
-            backend = self.backend,
+                    self.dispatchers[key] = dispatcher
-            timeout = 120,
-            consumer = f"{self.prefix}-{kind}-request",
-            subscriber = f"{self.prefix}-{kind}-request",
-            request_queue = request_queue,
-            response_queue = response_queue,
-        )
 
-        await dispatcher.start()
+        return await self.dispatchers[key].process(data, responder)
-
-        self.dispatchers[key] = dispatcher
-
-        return await dispatcher.process(data, responder)
 
     def dispatch_flow_import(self):
         return self.process_flow_import
@@ -297,36 +296,35 @@ class DispatcherManager:
 
         key = (flow, kind)
 
-        if key in self.dispatchers:
+        if key not in self.dispatchers:
-            return await self.dispatchers[key].process(data, responder)
+            async with self.dispatcher_lock:
+                if key not in self.dispatchers:
+                    intf_defs = self.flows[flow]["interfaces"]
 
-        intf_defs = self.flows[flow]["interfaces"]
+                    if kind not in intf_defs:
+                        raise RuntimeError("This kind not supported by flow")
 
-        if kind not in intf_defs:
+                    qconfig = intf_defs[kind]
-            raise RuntimeError("This kind not supported by flow")
 
-        qconfig = intf_defs[kind]
+                    if kind in request_response_dispatchers:
+                        dispatcher = request_response_dispatchers[kind](
+                            backend = self.backend,
+                            request_queue = qconfig["request"],
+                            response_queue = qconfig["response"],
+                            timeout = 120,
+                            consumer = f"{self.prefix}-{flow}-{kind}-request",
+                            subscriber = f"{self.prefix}-{flow}-{kind}-request",
+                        )
+                    elif kind in sender_dispatchers:
+                        dispatcher = sender_dispatchers[kind](
+                            backend = self.backend,
+                            queue = qconfig,
+                        )
+                    else:
+                        raise RuntimeError("Invalid kind")
 
-        if kind in request_response_dispatchers:
+                    await dispatcher.start()
-            dispatcher = request_response_dispatchers[kind](
+                    self.dispatchers[key] = dispatcher
-                backend = self.backend,
-                request_queue = qconfig["request"],
-                response_queue = qconfig["response"],
-                timeout = 120,
-                consumer = f"{self.prefix}-{flow}-{kind}-request",
-                subscriber = f"{self.prefix}-{flow}-{kind}-request",
-            )
-        elif kind in sender_dispatchers:
-            dispatcher = sender_dispatchers[kind](
-                backend = self.backend,
-                queue = qconfig,
-            )
-        else:
-            raise RuntimeError("Invalid kind")
-            
-        await dispatcher.start()
 
-        self.dispatchers[key] = dispatcher
+        return await self.dispatchers[key].process(data, responder)
-
-        return await dispatcher.process(data, responder)