[ML] Return a Conflict status code if the model deployment is stopped by a user (#125204) (#125486)

Author: davidkyle

docs/changelog/125204.yaml

@@ -0,0 +1,6 @@
+pr: 125204
+summary: Return a Conflict status code if the model deployment is stopped by a user
+area: Machine Learning
+type: bug
+issues:
+ - 123745

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/deployment/AbstractControlMessagePyTorchAction.java

@@ -87,7 +87,7 @@ final BytesReference buildControlMessage(String requestId) throws IOException {
 
     private void processResponse(PyTorchResult result) {
         if (result.isError()) {
-            onFailure(result.errorResult().error());
+            onFailure(result.errorResult());
             return;
         }
         onSuccess(getResult(result));

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/deployment/AbstractPyTorchAction.java

@@ -16,6 +16,7 @@
 import org.elasticsearch.threadpool.ThreadPool;
 import org.elasticsearch.xpack.core.ml.utils.ExceptionsHelper;
 import org.elasticsearch.xpack.ml.MachineLearning;
+import org.elasticsearch.xpack.ml.inference.pytorch.results.ErrorResult;
 import org.elasticsearch.xpack.ml.job.process.AbstractInitializableRunnable;
 
 import java.util.concurrent.atomic.AtomicBoolean;
@@ -116,8 +117,9 @@ public void onFailure(Exception e) {
         getLogger().debug(() -> format("[%s] request [%s] received failure but listener already notified", deploymentId, requestId), e);
     }
 
-    protected void onFailure(String errorMessage) {
-        onFailure(new ElasticsearchStatusException("Error in inference process: [" + errorMessage + "]", RestStatus.INTERNAL_SERVER_ERROR));
+    protected void onFailure(ErrorResult errorResult) {
+        var restStatus = errorResult.isStopping() ? RestStatus.CONFLICT : RestStatus.INTERNAL_SERVER_ERROR;
+        onFailure(new ElasticsearchStatusException("Error in inference process: [" + errorResult.error() + "]", restStatus));
     }
 
     boolean isNotified() {

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/deployment/InferencePyTorchAction.java

@@ -178,7 +178,7 @@ private void processResult(
         NlpTask.ResultProcessor inferenceResultsProcessor
     ) {
         if (pyTorchResult.isError()) {
-            onFailure(pyTorchResult.errorResult().error());
+            onFailure(pyTorchResult.errorResult());
             return;
         }
 

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/pytorch/process/PyTorchResultProcessor.java

@@ -130,11 +130,12 @@ public void process(PyTorchProcess process) {
             var errorResult = new ErrorResult(
                 isStopping
                     ? "inference canceled as process is stopping"
-                    : "inference native process died unexpectedly with failure [" + e.getMessage() + "]"
+                    : "inference native process died unexpectedly with failure [" + e.getMessage() + "]",
+                isStopping
             );
             notifyAndClearPendingResults(errorResult);
         } finally {
-            notifyAndClearPendingResults(new ErrorResult("inference canceled as process is stopping"));
+            notifyAndClearPendingResults(new ErrorResult("inference canceled as process is stopping", true));
             processorCompletionLatch.countDown();
         }
         logger.debug(() -> "[" + modelId + "] Results processing finished");

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/pytorch/results/ErrorResult.java

@@ -14,7 +14,7 @@
 
 import java.io.IOException;
 
-public record ErrorResult(String error) implements ToXContentObject {
+public record ErrorResult(String error, boolean isStopping) implements ToXContentObject {
 
     public static final ParseField ERROR = new ParseField("error");
 
@@ -23,6 +23,10 @@ public record ErrorResult(String error) implements ToXContentObject {
         a -> new ErrorResult((String) a[0])
     );
 
+    public ErrorResult(String error) {
+        this(error, false);
+    }
+
     static {
         PARSER.declareString(ConstructingObjectParser.constructorArg(), ERROR);
     }

x-pack/plugin/ml/src/test/java/org/elasticsearch/xpack/ml/inference/pytorch/process/PyTorchResultProcessorTests.java

@@ -134,11 +134,16 @@ public void testCancelPendingRequest() {
     public void testPendingRequestAreCalledAtShutdown() {
         var processor = new PyTorchResultProcessor("foo", s -> {});
 
+        Consumer<PyTorchResult> resultChecker = r -> {
+            assertTrue(r.errorResult().isStopping());
+            assertEquals(r.errorResult().error(), "inference canceled as process is stopping");
+        };
+
         var listeners = List.of(
-            new AssertingResultListener(r -> assertEquals(r.errorResult().error(), "inference canceled as process is stopping")),
-            new AssertingResultListener(r -> assertEquals(r.errorResult().error(), "inference canceled as process is stopping")),
-            new AssertingResultListener(r -> assertEquals(r.errorResult().error(), "inference canceled as process is stopping")),
-            new AssertingResultListener(r -> assertEquals(r.errorResult().error(), "inference canceled as process is stopping"))
+            new AssertingResultListener(resultChecker),
+            new AssertingResultListener(resultChecker),
+            new AssertingResultListener(resultChecker),
+            new AssertingResultListener(resultChecker)
         );
 
         int i = 0;
@@ -153,6 +158,33 @@ public void testPendingRequestAreCalledAtShutdown() {
         }
     }
 
+    public void testPendingRequestAreCalledOnException() {
+        var processor = new PyTorchResultProcessor("foo", s -> {});
+
+        Consumer<PyTorchResult> resultChecker = r -> {
+            assertFalse(r.errorResult().isStopping());
+            assertEquals(r.errorResult().error(), "inference native process died unexpectedly with failure [mocked exception]");
+        };
+
+        var listeners = List.of(
+            new AssertingResultListener(resultChecker),
+            new AssertingResultListener(resultChecker),
+            new AssertingResultListener(resultChecker),
+            new AssertingResultListener(resultChecker)
+        );
+
+        int i = 0;
+        for (var l : listeners) {
+            processor.registerRequest(Integer.toString(i++), l);
+        }
+
+        processor.process(throwingNativeProcess());
+
+        for (var l : listeners) {
+            assertTrue(l.hasResponse);
+        }
+    }
+
     public void testsHandleUnknownResult() {
         var processor = new PyTorchResultProcessor("deployment-foo", settings -> {});
         var listener = new AssertingResultListener(
@@ -379,4 +411,10 @@ private NativePyTorchProcess mockNativeProcess(Iterator<PyTorchResult> results)
         when(process.readResults()).thenReturn(results);
         return process;
     }
+
+    private NativePyTorchProcess throwingNativeProcess() {
+        var process = mock(NativePyTorchProcess.class);
+        when(process.readResults()).thenThrow(new RuntimeException("mocked exception"));
+        return process;
+    }
 }