Merge branch 'main' into docs/add-weighted-rrf-documentation

Author: mridula-s109

server/src/main/java/org/elasticsearch/inference/telemetry/InferenceStats.java

@@ -49,19 +49,11 @@ public static InferenceStats create(MeterRegistry meterRegistry) {
  );
  }
 
- public static Map<String, Object> modelAttributes(Model model) {
- var modelAttributesMap = new HashMap<String, Object>();
- modelAttributesMap.put("service", model.getConfigurations().getService());
- modelAttributesMap.put("task_type", model.getTaskType().toString());
-
- if (Objects.nonNull(model.getServiceSettings().modelId())) {
- modelAttributesMap.put("model_id", model.getServiceSettings().modelId());
- }
-
- return modelAttributesMap;
+ public static Map<String, Object> serviceAttributes(Model model) {
+ return Map.of("service", model.getConfigurations().getService(), "task_type", model.getTaskType().toString());
  }
 
- public static Map<String, Object> modelAttributes(UnparsedModel model) {
+ public static Map<String, Object> serviceAttributes(UnparsedModel model) {
  return Map.of("service", model.service(), "task_type", model.taskType().toString());
  }
 
@@ -77,9 +69,9 @@ public static Map<String, Object> responseAttributes(@Nullable Throwable throwab
  return Map.of("error.type", throwable.getClass().getSimpleName());
  }
 
- public static Map<String, Object> modelAndResponseAttributes(Model model, @Nullable Throwable throwable) {
+ public static Map<String, Object> serviceAndResponseAttributes(Model model, @Nullable Throwable throwable) {
  var metricAttributes = new HashMap<String, Object>();
- metricAttributes.putAll(modelAttributes(model));
+ metricAttributes.putAll(serviceAttributes(model));
  metricAttributes.putAll(responseAttributes(throwable));
  return metricAttributes;
  }

server/src/test/java/org/elasticsearch/inference/telemetry/InferenceStatsTests.java

@@ -25,8 +25,8 @@
 import java.util.Map;
 
 import static org.elasticsearch.inference.telemetry.InferenceStats.create;
-import static org.elasticsearch.inference.telemetry.InferenceStats.modelAttributes;
 import static org.elasticsearch.inference.telemetry.InferenceStats.responseAttributes;
+import static org.elasticsearch.inference.telemetry.InferenceStats.serviceAttributes;
 import static org.hamcrest.Matchers.is;
 import static org.hamcrest.Matchers.nullValue;
 import static org.mockito.ArgumentMatchers.assertArg;
@@ -41,23 +41,20 @@ public static InferenceStats mockInferenceStats() {
  return new InferenceStats(mock(), mock(), mock());
  }
 
- public void testRecordWithModel() {
+ public void testRecordWithService() {
  var longCounter = mock(LongCounter.class);
  var stats = new InferenceStats(longCounter, mock(), mock());
 
- stats.requestCount().incrementBy(1, modelAttributes(model("service", TaskType.ANY, "modelId")));
+ stats.requestCount().incrementBy(1, serviceAttributes(model("service", TaskType.ANY, "modelId")));
 
- verify(longCounter).incrementBy(
- eq(1L),
- eq(Map.of("service", "service", "task_type", TaskType.ANY.toString(), "model_id", "modelId"))
- );
+ verify(longCounter).incrementBy(eq(1L), eq(Map.of("service", "service", "task_type", TaskType.ANY.toString())));
  }
 
  public void testRecordWithoutModel() {
  var longCounter = mock(LongCounter.class);
  var stats = new InferenceStats(longCounter, mock(), mock());
 
- stats.requestCount().incrementBy(1, modelAttributes(model("service", TaskType.ANY, null)));
+ stats.requestCount().incrementBy(1, serviceAttributes(model("service", TaskType.ANY, null)));
 
  verify(longCounter).incrementBy(eq(1L), eq(Map.of("service", "service", "task_type", TaskType.ANY.toString())));
  }
@@ -72,15 +69,14 @@ public void testRecordDurationWithoutError() {
  var stats = new InferenceStats(mock(), histogramCounter, mock());
 
  Map<String, Object> metricAttributes = new HashMap<>();
- metricAttributes.putAll(modelAttributes(model("service", TaskType.ANY, "modelId")));
+ metricAttributes.putAll(serviceAttributes(model("service", TaskType.ANY, "modelId")));
  metricAttributes.putAll(responseAttributes(null));
 
  stats.inferenceDuration().record(expectedLong, metricAttributes);
 
  verify(histogramCounter).record(eq(expectedLong), assertArg(attributes -> {
  assertThat(attributes.get("service"), is("service"));
  assertThat(attributes.get("task_type"), is(TaskType.ANY.toString()));
- assertThat(attributes.get("model_id"), is("modelId"));
  assertThat(attributes.get("status_code"), is(200));
  assertThat(attributes.get("error.type"), nullValue());
  }));
@@ -100,15 +96,14 @@ public void testRecordDurationWithElasticsearchStatusException() {
  var expectedError = String.valueOf(statusCode.getStatus());
 
  Map<String, Object> metricAttributes = new HashMap<>();
- metricAttributes.putAll(modelAttributes(model("service", TaskType.ANY, "modelId")));
+ metricAttributes.putAll(serviceAttributes(model("service", TaskType.ANY, "modelId")));
  metricAttributes.putAll(responseAttributes(exception));
 
  stats.inferenceDuration().record(expectedLong, metricAttributes);
 
  verify(histogramCounter).record(eq(expectedLong), assertArg(attributes -> {
  assertThat(attributes.get("service"), is("service"));
  assertThat(attributes.get("task_type"), is(TaskType.ANY.toString()));
- assertThat(attributes.get("model_id"), is("modelId"));
  assertThat(attributes.get("status_code"), is(statusCode.getStatus()));
  assertThat(attributes.get("error.type"), is(expectedError));
  }));
@@ -127,15 +122,14 @@ public void testRecordDurationWithOtherException() {
  var expectedError = exception.getClass().getSimpleName();
 
  Map<String, Object> metricAttributes = new HashMap<>();
- metricAttributes.putAll(modelAttributes(model("service", TaskType.ANY, "modelId")));
+ metricAttributes.putAll(serviceAttributes(model("service", TaskType.ANY, "modelId")));
  metricAttributes.putAll(responseAttributes(exception));
 
  stats.inferenceDuration().record(expectedLong, metricAttributes);
 
  verify(histogramCounter).record(eq(expectedLong), assertArg(attributes -> {
  assertThat(attributes.get("service"), is("service"));
  assertThat(attributes.get("task_type"), is(TaskType.ANY.toString()));
- assertThat(attributes.get("model_id"), is("modelId"));
  assertThat(attributes.get("status_code"), nullValue());
  assertThat(attributes.get("error.type"), is(expectedError));
  }));
@@ -152,7 +146,7 @@ public void testRecordDurationWithUnparsedModelAndElasticsearchStatusException()
  var unparsedModel = new UnparsedModel("inferenceEntityId", TaskType.ANY, "service", Map.of(), Map.of());
 
  Map<String, Object> metricAttributes = new HashMap<>();
- metricAttributes.putAll(modelAttributes(unparsedModel));
+ metricAttributes.putAll(serviceAttributes(unparsedModel));
  metricAttributes.putAll(responseAttributes(exception));
 
  stats.inferenceDuration().record(expectedLong, metricAttributes);
@@ -176,7 +170,7 @@ public void testRecordDurationWithUnparsedModelAndOtherException() {
  var unparsedModel = new UnparsedModel("inferenceEntityId", TaskType.ANY, "service", Map.of(), Map.of());
 
  Map<String, Object> metricAttributes = new HashMap<>();
- metricAttributes.putAll(modelAttributes(unparsedModel));
+ metricAttributes.putAll(serviceAttributes(unparsedModel));
  metricAttributes.putAll(responseAttributes(exception));
 
  stats.inferenceDuration().record(expectedLong, metricAttributes);

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/action/BaseTransportInferenceAction.java

@@ -49,9 +49,8 @@
 
 import static org.elasticsearch.ExceptionsHelper.unwrapCause;
 import static org.elasticsearch.core.Strings.format;
-import static org.elasticsearch.inference.telemetry.InferenceStats.modelAndResponseAttributes;
-import static org.elasticsearch.inference.telemetry.InferenceStats.modelAttributes;
 import static org.elasticsearch.inference.telemetry.InferenceStats.responseAttributes;
+import static org.elasticsearch.inference.telemetry.InferenceStats.serviceAndResponseAttributes;
 import static org.elasticsearch.xpack.inference.InferencePlugin.INFERENCE_API_FEATURE;
 
 /**
@@ -181,7 +180,7 @@ private static void validationHelper(Supplier<Boolean> validationFailure, Suppli
 
  private void recordRequestDurationMetrics(Model model, InferenceTimer timer, @Nullable Throwable t) {
  Map<String, Object> metricAttributes = new HashMap<>();
- metricAttributes.putAll(modelAttributes(model));
+ metricAttributes.putAll(InferenceStats.serviceAttributes(model));
  metricAttributes.putAll(responseAttributes(unwrapCause(t)));
 
  inferenceStats.inferenceDuration().record(timer.elapsedMillis(), metricAttributes);
@@ -270,7 +269,7 @@ protected <T> Flow.Publisher<T> streamErrorHandler(Flow.Publisher<T> upstream) {
 
  private void recordRequestCountMetrics(Model model, Request request, String localNodeId) {
  Map<String, Object> requestCountAttributes = new HashMap<>();
- requestCountAttributes.putAll(modelAttributes(model));
+ requestCountAttributes.putAll(InferenceStats.serviceAttributes(model));
 
  inferenceStats.requestCount().incrementBy(1, requestCountAttributes);
  }
@@ -283,7 +282,7 @@ private void recordRequestDurationMetrics(
  @Nullable Throwable t
  ) {
  Map<String, Object> metricAttributes = new HashMap<>();
- metricAttributes.putAll(modelAndResponseAttributes(model, unwrapCause(t)));
+ metricAttributes.putAll(serviceAndResponseAttributes(model, unwrapCause(t)));
 
  inferenceStats.inferenceDuration().record(timer.elapsedMillis(), metricAttributes);
  }

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/action/filter/ShardBulkInferenceActionFilter.java

@@ -77,7 +77,7 @@
 import java.util.Map;
 import java.util.stream.Collectors;
 
-import static org.elasticsearch.inference.telemetry.InferenceStats.modelAndResponseAttributes;
+import static org.elasticsearch.inference.telemetry.InferenceStats.serviceAndResponseAttributes;
 import static org.elasticsearch.xpack.inference.InferencePlugin.INFERENCE_API_FEATURE;
 import static org.elasticsearch.xpack.inference.mapper.SemanticTextField.toSemanticTextFieldChunks;
 import static org.elasticsearch.xpack.inference.mapper.SemanticTextField.toSemanticTextFieldChunksLegacy;
@@ -461,7 +461,7 @@ public void onFailure(Exception exc) {
 
  private void recordRequestCountMetrics(Model model, int incrementBy, Throwable throwable) {
  Map<String, Object> requestCountAttributes = new HashMap<>();
- requestCountAttributes.putAll(modelAndResponseAttributes(model, throwable));
+ requestCountAttributes.putAll(serviceAndResponseAttributes(model, throwable));
  requestCountAttributes.put("inference_source", "semantic_text_bulk");
  inferenceStats.requestCount().incrementBy(incrementBy, requestCountAttributes);
  }

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elasticsearch/BaseElasticsearchInternalService.java

@@ -47,7 +47,7 @@
 
 import static org.elasticsearch.ExceptionsHelper.unwrapCause;
 import static org.elasticsearch.core.Strings.format;
-import static org.elasticsearch.inference.telemetry.InferenceStats.modelAndResponseAttributes;
+import static org.elasticsearch.inference.telemetry.InferenceStats.serviceAndResponseAttributes;
 import static org.elasticsearch.xpack.core.ClientHelper.INFERENCE_ORIGIN;
 import static org.elasticsearch.xpack.core.ClientHelper.executeAsyncWithOrigin;
 
@@ -126,7 +126,7 @@ public void start(Model model, TimeValue timeout, ActionListener<Boolean> finalL
  });
  subscribableListener.addTimeout(timeout, threadPool, inferenceExecutor);
  subscribableListener.addListener(ActionListener.wrap(started -> {
- inferenceStats.deploymentDuration().record(timer.elapsedMillis(), modelAndResponseAttributes(model, null));
+ inferenceStats.deploymentDuration().record(timer.elapsedMillis(), serviceAndResponseAttributes(model, null));
  finalListener.onResponse(started);
  }, e -> {
  if (e instanceof ElasticsearchTimeoutException) {
@@ -139,10 +139,11 @@ public void start(Model model, TimeValue timeout, ActionListener<Boolean> finalL
  model.getInferenceEntityId()
  )
  );
- inferenceStats.deploymentDuration().record(timer.elapsedMillis(), modelAndResponseAttributes(model, timeoutException));
+ inferenceStats.deploymentDuration()
+ .record(timer.elapsedMillis(), serviceAndResponseAttributes(model, timeoutException));
  finalListener.onFailure(timeoutException);
  } else {
- inferenceStats.deploymentDuration().record(timer.elapsedMillis(), modelAndResponseAttributes(model, unwrapCause(e)));
+ inferenceStats.deploymentDuration().record(timer.elapsedMillis(), serviceAndResponseAttributes(model, unwrapCause(e)));
  finalListener.onFailure(e);
  }
  }));

x-pack/plugin/inference/src/test/java/org/elasticsearch/xpack/inference/action/filter/ShardBulkInferenceActionFilterTests.java

@@ -356,7 +356,6 @@ public void testItemFailures() throws Exception {
  assertThat(statusCode, is(200));
  }
  assertThat(attributes.get("task_type"), is(model.getTaskType().toString()));
- assertThat(attributes.get("model_id"), is(model.getServiceSettings().modelId()));
  assertThat(attributes.get("service"), is(model.getConfigurations().getService()));
  assertThat(attributes.get("inference_source"), is("semantic_text_bulk"));
  }));