[workflow] Allows model being shared between workflows

deepjavalibrary · Apr 25, 2023 · 9125dbb · 9125dbb
1 parent 7edc759
commit 9125dbb
Show file tree

Hide file tree

Showing 7 changed files with 53 additions and 9 deletions.
diff --git a/serving/src/main/java/ai/djl/serving/models/ModelManager.java b/serving/src/main/java/ai/djl/serving/models/ModelManager.java
@@ -83,7 +83,10 @@ public CompletableFuture<Void> registerWorkflow(Workflow workflow) {
 
         return CompletableFuture.supplyAsync(
                 () -> {
-                    for (ModelInfo<Input, Output> model : workflow.getModels()) {
+                    Map<String, ModelInfo<Input, Output>> models = workflow.getModelMap();
+                    for (Map.Entry<String, ModelInfo<Input, Output>> entry : models.entrySet()) {
+                        String key = entry.getKey();
+                        ModelInfo<Input, Output> model = entry.getValue();
                         try {
                             // download model and configure per model settings
                             model.initialize();
@@ -92,14 +95,21 @@ public CompletableFuture<Void> registerWorkflow(Workflow workflow) {
                             String engine = model.getEngineName();
                             DependencyManager dm = DependencyManager.getInstance();
                             dm.installEngine(engine);
-                            wlm.registerModel(model);
+                            WorkerPool<Input, Output> wp = wlm.getWorkerPool(model);
+                            if (wp != null) {
+                                models.put(key, wp.getModel());
+                                wp.increaseRef();
+                                logger.info("Model {} is registered by other workflow", model);
+                                continue;
+                            }
 
+                            wlm.registerModel(model);
                             String[] devices = model.getLoadOnDevices();
                             logger.info("Loading model on {}:{}", engine, Arrays.toString(devices));
                             for (String deviceName : devices) {
                                 int minWorkers = model.getMinWorkers();
                                 int maxWorkers = model.getMaxWorkers();
-                                modelManager.initWorkers(model, deviceName, minWorkers, maxWorkers);
+                                initWorkers(model, deviceName, minWorkers, maxWorkers);
                             }
                         } catch (IOException | ModelException e) {
                             throw new CompletionException(e);

diff --git a/serving/src/main/java/ai/djl/serving/workflow/Workflow.java b/serving/src/main/java/ai/djl/serving/workflow/Workflow.java
@@ -115,6 +115,15 @@ public Collection<ModelInfo<Input, Output>> getModels() {
         return models.values();
     }
 
+    /**
+     * Returns the model map in the workflow.
+     *
+     * @return the model map in the workflow
+     */
+    public Map<String, ModelInfo<Input, Output>> getModelMap() {
+        return models;
+    }
+
     /**
      * Executes a workflow with an input.
      *

diff --git a/serving/src/test/java/ai/djl/serving/ModelServerTest.java b/serving/src/test/java/ai/djl/serving/ModelServerTest.java
@@ -264,7 +264,7 @@ public void testWorkflows()
             assertTrue(server.isRunning());
             Channel channel = initTestChannel();
 
-            testPredictionsModels(channel);
+            testPredictions(channel, new String[] {"/predictions/m"});
             testPredictionsWorkflows(channel);
 
             channel.close().sync();

diff --git a/serving/src/test/java/ai/djl/serving/WorkflowTest.java b/serving/src/test/java/ai/djl/serving/WorkflowTest.java
@@ -78,7 +78,7 @@ public void testFunctions() throws IOException, BadWorkflowException {
     public void testLocalPerf() throws IOException, BadWorkflowException {
         Path workflowFile = Paths.get("src/test/resources/workflows/localPerf.json");
         Workflow workflow = WorkflowDefinition.parse(workflowFile).toWorkflow();
-        ModelInfo<Input, Output> m = workflow.getModels().stream().findFirst().get();
+        ModelInfo<Input, Output> m = workflow.getModels().iterator().next();
 
         Assert.assertEquals(m.getQueueSize(), 102);
         Assert.assertEquals(m.getMaxIdleSeconds(), 62);

diff --git a/serving/src/test/resources/workflow.config.properties b/serving/src/test/resources/workflow.config.properties
@@ -2,7 +2,7 @@
 inference_address=https://127.0.0.1:8443
 management_address=https://127.0.0.1:8443
 model_store=build/models
-load_models=https://resources.djl.ai/test-models/mlp.tar.gz,[mlp:v1:MXNet:*]=https://resources.djl.ai/test-models/mlp.tar.gz,https://resources.djl.ai/test-models/basic-serving-workflow.json
+load_models=m=https://resources.djl.ai/test-models/mlp.tar.gz,https://resources.djl.ai/test-models/basic-serving-workflow.json
 private_key_file=src/test/resources/key.pem
 certificate_file=src/test/resources/certs.pem
 max_request_size=10485760
diff --git a/wlm/src/main/java/ai/djl/serving/wlm/WorkLoadManager.java b/wlm/src/main/java/ai/djl/serving/wlm/WorkLoadManager.java
@@ -71,8 +71,11 @@ public <I, O> WorkerPool<I, O> registerModel(ModelInfo<I, O> modelInfo) {
      */
     public void unregisterModel(ModelInfo<?, ?> model) {
         WorkerPool<?, ?> pool = getWorkerPool(model);
-        pool.shutdownWorkers();
-        workerPools.remove(model);
+        if (pool.decreaseRef() <= 0) {
+            logger.info("Unloading model: {}", model);
+            pool.shutdownWorkers();
+            workerPools.remove(model);
+        }
     }
 
     /**

diff --git a/wlm/src/main/java/ai/djl/serving/wlm/WorkerPool.java b/wlm/src/main/java/ai/djl/serving/wlm/WorkerPool.java
@@ -28,6 +28,7 @@
 import java.util.concurrent.ConcurrentHashMap;
 import java.util.concurrent.ExecutorService;
 import java.util.concurrent.LinkedBlockingDeque;
+import java.util.concurrent.atomic.AtomicInteger;
 import java.util.stream.Collectors;
 
 /**
@@ -43,6 +44,7 @@ public class WorkerPool<I, O> {
     private ExecutorService threadPool;
     private Map<Device, WorkerGroup<I, O>> workerGroups;
     private LinkedBlockingDeque<WorkerJob<I, O>> jobQueue;
+    private AtomicInteger refCnt;
 
     /**
      * Construct and initial data structure.
@@ -54,9 +56,29 @@ public class WorkerPool<I, O> {
         this.model = model;
         this.threadPool = threadPool;
         workerGroups = new ConcurrentHashMap<>();
+        refCnt = new AtomicInteger(1);
     }
 
-    ModelInfo<I, O> getModel() {
+    /** Increases the reference count. */
+    public void increaseRef() {
+        refCnt.incrementAndGet();
+    }
+
+    /**
+     * Decrease the reference count and return the current count.
+     *
+     * @return the current count
+     */
+    public int decreaseRef() {
+        return refCnt.decrementAndGet();
+    }
+
+    /**
+     * Returns the model of the worker pool.
+     *
+     * @return the model of the worker pool
+     */
+    public ModelInfo<I, O> getModel() {
         return model;
     }