[ML] Use EIS v2 authorization endpoint for inference API #138249

jonathan-buttner · 2025-11-18T16:44:17Z

This PR switches the inference API to point to the EIS v2 authorization endpoint and handles the new response format.

The EIS v2 authorization endpoint provides all the necessary fields for the inference API to create an inference endpoint. The inference API will only create endpoints that have a task type that is supported (the ES integration defines a set of supported task types).

Note: The AuthorizationResponseEntity is not registered even though it is a named writeable. It is a bit of a hack that it needs to be a named writeable just to get it to work through the Sender framework.

EIS v2 auth endpoint response format

{
  "inference_endpoints": [
    {
      "id": ".rainbow-sprinkles-elastic",
      "model_name": "rainbow-sprinkles",
      "task_types": {
        "eis": "chat",
        "elasticsearch": "chat_completion"
      }
      "status": "ga",
      "properties": [
        "multilingual"
      ],
      "release_date": "2024-05-01",
      "end_of_life_date": "2025-12-31"
    },
    {
      "id": ".elastic-elser-v2",
      "model_name": "elser_model_2",
      "task_types": {
        "eis": "embed/text/sparse",
        "elasticsearch": "sparse_embedding"
      }
      "status": "preview",
      "properties": [
        "english"
      ],
      "release_date": "2024-05-01",
      "configuration": {
        "chunking_settings": {
          "strategy": "sentence",
          "max_chunk_size": 250,
          "sentence_overlap": 1
        }
      }
    },
    {
      "id": ".jina-embeddings-v3",
      "model_name": "jina-embeddings-v3",
      "task_types": {
        "eis": "embed/text/dense",
        "elasticsearch": "text_embedding"
      }
      "status": "beta",
      "properties": [
        "multilingual",
        "open-weights"
      ],
      "release_date": "2024-05-01",
      "configuration": {
        "similarity": "cosine",
        "dimensions": 1024,
        "element_type": "float",
        "chunking_settings": {
          "strategy": "sentence",
          "max_chunk_size": 250,
          "sentence_overlap": 1
        }
      }
    }
  ]
}

Testing

eis-gateway

make TLS_CLIENT_AUTH=NoClientCert run

Start elasticsearch

./gradlew :run -Drun.license_type=trial -Dtests.es.xpack.inference.elastic.url=https://localhost:8443 -Dtests.es.xpack.inference.elastic.http.ssl.verification_mode=none -Dtests.es.xpack.inference.elastic.authorization_request_interval="5s" -Dtests.es.xpack.inference.elastic.max_authorization_request_jitter="1s" -Dtests.es.xpack.inference.elastic.ccm_supported_environment=false

You should see preconfigured EIS endpoints:

GET _inference/_all

Response

{
    "endpoints": [
        {
            "inference_id": ".elser-2-elastic",
            "task_type": "sparse_embedding",
            "service": "elastic",
            "service_settings": {
                "model_id": "elser_model_2"
            },
            "chunking_settings": {
                "strategy": "sentence",
                "max_chunk_size": 250,
                "sentence_overlap": 1
            }
        },
...

…uth-v2

…earch into ml-eis-auth-v2

…uth-v2

jonathan-buttner · 2025-11-18T22:01:30Z

...c/main/java/org/elasticsearch/xpack/core/inference/action/StoreInferenceEndpointsAction.java


    public static class Request extends AcknowledgedRequest<Request> {
-        private final List<Model> models;
+        private final List<? extends Model> models;


These changes are so the authorization logic can return a list of a child class of Model.

You can avoid needing to use <? extends Model> in this PR by making a small change to ElasticInferenceServiceAuthorizationModel.getEndpoints():

public List<Model> getEndpoints(Set<String> endpointIds) { return endpointIds.stream().<Model>map(authorizedEndpoints::get).filter(Objects::nonNull).toList(); }

By letting the stream know that is should be a Stream<Model> after the map() call instead of it inferring the Stream<ElasticInferenceServiceModel> type, the return type can use List<Model>

jonathan-buttner · 2025-11-20T13:31:47Z

...t/java/org/elasticsearch/xpack/inference/MockElasticInferenceServiceAuthorizationServer.java

-            """;
-
-        webServer.enqueue(new MockResponse().setResponseCode(200).setBody(responseJson));
+        var authResponse = getEisAuthorizationResponseWithMultipleEndpoints("ignored");


The URL is typically passed in to this method. We don't have access to it yet because the webserver may not have been started yet. These tests doesn't actually need the parts of the getEisAuthorizationResponseWithMultipleEndpoints response that leverage the passed in url here anyway.

jonathan-buttner · 2025-11-20T13:32:18Z

x-pack/plugin/inference/qa/inference-service-tests/build.gradle

  clusterPlugins project(':x-pack:plugin:inference:qa:test-service-plugin')
+
+  // Allow javaRestTest to see unit-test classes from x-pack:plugin:inference so we can import some variables
+  javaRestTestImplementation(testArtifact(project(xpackModule('inference'))))


This is so the qa tests can imported from the unit tests in the inference plugin.

jonathan-buttner · 2025-11-20T13:33:48Z

...pack/inference/services/elastic/authorization/ElasticInferenceServiceAuthorizationModel.java

+    }
+
+    private static Map<String, Object> getChunkingSettingsMap(AuthorizationResponseEntity.Configuration configuration) {
+        return Objects.requireNonNullElse(configuration.chunkingSettings(), new HashMap<>());


Default to an empty map so the chunking settings use the "newer" default logic (default to the sentence strategy rather than word).

Could we add a comment here that this happens? Or would it be possible to return a chunking strategy object instead of a generic map and fallback to the actual default chunking strategy?

Yeah I'll add a comment. I think it's probably best to keep things consistent and have ChunkingSettingsBuilder.fromMap() handle what to do if the settings aren't provided.

jonathan-buttner · 2025-11-20T13:38:20Z

...pack/inference/services/elastic/authorization/ElasticInferenceServiceAuthorizationModel.java

+            new ElasticInferenceServiceDenseTextEmbeddingsServiceSettings(
+                authorizedEndpoint.modelName(),
+                getSimilarityMeasure(config),
+                config.dimensions(),


Right now the element type isn't used because we hard code to float. I have an issue to fix that after this PR is merged though.

jonathan-buttner · 2025-11-20T14:06:06Z

.../inference/services/elastic/response/ElasticInferenceServiceAuthorizationResponseEntity.java

@@ -1,194 +0,0 @@
-/*


This represented the v1 authorization endpoint response. We no longer interact with that so we don't need this.

jonathan-buttner · 2025-11-20T14:06:40Z

.../java/org/elasticsearch/xpack/inference/services/elastic/InternalPreconfiguredEndpoints.java

-
    // elser-2
-    public static final String DEFAULT_ELSER_2_MODEL_ID = "elser_model_2";
    public static final String DEFAULT_ELSER_ENDPOINT_ID_V2 = ".elser-2-elastic";


I left this because semantic text references it. Once we implement the heuristics logic we can remove this as well.

jonathan-buttner · 2025-11-20T14:11:38Z

...st/java/org/elasticsearch/xpack/inference/services/elastic/ElasticInferenceServiceTests.java

    public void testHideFromConfigurationApi_ThrowsUnsupported_WithAvailableModels() throws Exception {
-        try (
-            var service = createServiceWithMockSender(
-                ElasticInferenceServiceAuthorizationModel.of(


This was unused by createServiceWithMockSender so removing.

jonathan-buttner · 2025-11-20T14:12:44Z

...rence/services/elastic/response/ElasticInferenceServiceAuthorizationResponseEntityTests.java

+        {
+          "inference_endpoints": [
+            {
+              "id": ".rainbow-sprinkles-elastic",


I figured it'd be easier to read if we didn't do a Strings.format() here to specify the id, name, and other field. I'm open to changing it though. I'm also open to other ideas to avoid the duplication.

jonathan-buttner · 2025-11-20T14:13:30Z

...rence/services/elastic/response/ElasticInferenceServiceAuthorizationResponseEntityTests.java

+    public static final String RERANK_V1_MODEL_NAME = "elastic-rerank-v1";
+    public static final String EIS_RERANK_PATH = "rerank/text/text-similarity";
+
+    public record EisAuthorizationResponse(


This is to encapsulate a testing using a specific eis response and what the expected entities should be created from that json response.

…uth-v2

elasticsearchmachine · 2025-11-21T15:40:13Z

Pinging @elastic/ml-core (Team:ML)

Safety measure to make sure we've the correct default rerank endpoint in place in case #138249 doesn't make it. Endpoint name: `.elastic-rerank-v1` -> `.jina-reranker-v2` Model name: `elastic-rerank-v1` -> `jina-reranker-v2`

…uth-v2

timgrein

Nice work! 👏 First round of reviews, will give it another go as it's pretty large

timgrein · 2025-11-25T14:54:38Z

.../main/java/org/elasticsearch/xpack/inference/action/TransportGetInferenceServicesAction.java


-            var config = ElasticInferenceService.createConfiguration(authorizationModel.getAuthorizedTaskTypes());
-            if (requestedTaskType != null && authorizationModel.getAuthorizedTaskTypes().contains(requestedTaskType) == false) {
+            var config = ElasticInferenceService.createConfiguration(authorizationModel.getTaskTypes());


Could we add a comment why we do an early return here? Didn't understand it a first glance. I assume we return here, because the auth model of EIS doesn't support the requested task type and therefore we simply return the ones we already have?

Yep exactly, if the user is looking for text_embedding, but they aren't authorized by EIS for any inference endpoints for text embedding, then we don't include EIS as a provider in that situation.

I'll add a comment.

timgrein · 2025-11-25T14:55:28Z

.../main/java/org/elasticsearch/xpack/inference/action/TransportGetInferenceServicesAction.java

                e
            );
-            delegate.onResponse(ElasticInferenceServiceAuthorizationModel.newDisabledService());
+            delegate.onResponse(AuthorizationModel.empty());


Does empty() imply "forbid all"? Maybe we could rename the method then to reflect the result of an empty auth model?

timgrein · 2025-11-25T14:56:12Z

...ava/org/elasticsearch/xpack/inference/services/elastic/authorization/AuthorizationModel.java

+import java.util.stream.Collectors;
+
+/**
+ * Transforms the response from {@link ElasticInferenceServiceAuthorizationRequestHandler} into a format for consumption by the service.


Suggested change

* Transforms the response from {@link ElasticInferenceServiceAuthorizationRequestHandler} into a format for consumption by the service.

* Transforms the response from {@link ElasticInferenceServiceAuthorizationRequestHandler} into a format for consumption by the {@link ElasticInferenceService}.

The service is in this case the ElasticInferenceService, right?

timgrein · 2025-11-25T14:57:40Z

...ava/org/elasticsearch/xpack/inference/services/elastic/authorization/AuthorizationModel.java

+/**
+ * Transforms the response from {@link ElasticInferenceServiceAuthorizationRequestHandler} into a format for consumption by the service.
+ */
+public class AuthorizationModel {


Suggested change

public class AuthorizationModel {

public class ElasticInferenceServiceAuthorizationModel {

nit: Wondering why we're using the name AuthorizationModel instead of ElasticInferenceServiceAuthorizationModel here as we usually prefix every EIS-related class with ElasticInferenceService. It's anyway 100% specific to EIS, right?

I'll rename this 👍 after we discuss as a team we can do a broader rename as needed for the other files too.

timgrein · 2025-11-25T14:59:28Z

...ava/org/elasticsearch/xpack/inference/services/elastic/authorization/AuthorizationModel.java

+            return switch (taskType) {
+                case CHAT_COMPLETION -> createCompletionModel(authorizedEndpoint, TaskType.CHAT_COMPLETION, components);
+                case COMPLETION -> createCompletionModel(authorizedEndpoint, TaskType.COMPLETION, components);
+                case SPARSE_EMBEDDING -> createSparseEmbeddingsModel(authorizedEndpoint, components);


Suggested change

case SPARSE_EMBEDDING -> createSparseEmbeddingsModel(authorizedEndpoint, components);

case SPARSE_EMBEDDING -> createSparseTextEmbeddingsModel(authorizedEndpoint, components);

nit: for consistency, I think sparse always implies a text embedding model on the other hand - feel free to ignore

timgrein · 2025-11-25T15:00:36Z

...pack/inference/services/elastic/authorization/ElasticInferenceServiceAuthorizationModel.java

+    }
+
+    private static Map<String, Object> getChunkingSettingsMap(AuthorizationResponseEntity.Configuration configuration) {
+        return Objects.requireNonNullElse(configuration.chunkingSettings(), new HashMap<>());


Could we add a comment here that this happens? Or would it be possible to return a chunking strategy object instead of a generic map and fallback to the actual default chunking strategy?

timgrein · 2025-11-25T15:04:45Z

.../inference/services/elastic/response/ElasticInferenceServiceAuthorizationResponseEntity.java

+
+    public record TaskTypeObject(String eisTaskType, String elasticsearchTaskType) implements Writeable, ToXContentObject {
+
+        private static final String EIS_TASK_TYPE_FIELD = "eis";


Suggested change

private static final String EIS_TASK_TYPE_FIELD = "eis";

private static final String ELASTIC_INFERENCE_SERVICE_TASK_TYPE_FIELD = "eis";

nit: I think at some point in the past we've agreed that we shouldn't use "EIS", but always the written out version "Elastic Inference Service" in our code

I'm going to hold off on this one, we can rename based on the team's discussion. The field that we're returning from the EIS auth v2 endpoint is the string eis 😄

…uth-v2

Copilot

Pull request overview

This PR migrates the Elastic Inference Service (EIS) integration from v1 to v2 authorization endpoint, introducing a new response format that provides comprehensive endpoint configuration including task types, model names, and optional settings like chunking and embedding dimensions. The v2 format enables dynamic endpoint creation directly from the authorization response, eliminating the need for hardcoded preconfigured endpoint mappings.

Key Changes:

Switched authorization endpoint from /api/v1/authorizations to /api/v2/authorizations with enhanced response parsing
Replaced static preconfigured endpoint mappings with dynamic model creation from authorization response
Refactored authorization model to store complete endpoint objects instead of just model names and task types

Reviewed changes

Copilot reviewed 29 out of 29 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
ElasticInferenceServiceAuthorizationResponseEntity.java	Complete restructure to parse v2 response format with endpoint IDs, task types, and configuration objects
ElasticInferenceServiceAuthorizationRequest.java	Updated to use URIBuilder for constructing v2 endpoint URL
ElasticInferenceServiceAuthorizationModel.java	Major refactor to create full model objects from authorization response instead of just tracking IDs
PreconfiguredEndpointModelAdapter.java	Deleted - no longer needed with dynamic model creation
InternalPreconfiguredEndpoints.java	Gutted to only retain minimal constants, removing hardcoded endpoint mappings
AuthorizationPoller.java	Updated to work with new model objects instead of inference IDs
StoreInferenceEndpointsAction.java	Made generic to accept `List<? extends Model>` instead of concrete `List<Model>`
ElasticInferenceServiceCompletionModel.java	Removed @nullable annotations from constructor parameters
Multiple test files	Updated to use new test helpers and response formats from v2
build.gradle	Added test artifact dependency to share test constants

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review for a chance to win a $100 gift card. Take the survey.

Copilot · 2025-11-25T22:09:13Z

...pack/inference/services/elastic/authorization/ElasticInferenceServiceAuthorizationModel.java

+        this.authorizedEndpoints = authorizedEndpoints.stream()
+            .collect(
+                Collectors.toMap(ElasticInferenceServiceModel::getInferenceEntityId, Function.identity(), (firstModel, secondModel) -> {
+                    logger.warn("Found inference id collision for id [{}], ignoring second model", firstModel.inferenceEntityId());


[nitpick] The duplicate ID handling on line 257 uses a lambda that logs a warning and returns the first model. However, the warning message references firstModel.inferenceEntityId() which may not be the most informative - consider including information about which model is being kept and which is being discarded (including their task types) to help with debugging.

Suggested change

logger.warn("Found inference id collision for id [{}], ignoring second model", firstModel.inferenceEntityId());

logger.warn(

"Found inference id collision for id [{}]. Keeping model with id [{}] (taskType={}), discarding model with id [{}] (taskType={})",

firstModel.getInferenceEntityId(),

firstModel.getInferenceEntityId(),

firstModel.getTaskType(),

secondModel.getInferenceEntityId(),

secondModel.getTaskType()

);

Copilot · 2025-11-25T22:09:14Z

...ch/xpack/inference/services/elastic/request/ElasticInferenceServiceAuthorizationRequest.java

+    private static URI createUri(String url) throws ElasticsearchStatusException {
        try {
-            // TODO, consider transforming the base URL into a URI for better error handling.
-            return new URI(url + "/api/v1/authorizations");
+            return new URIBuilder(url).setPath(AUTHORIZATION_PATH).build();


Using URIBuilder is better than simple string concatenation, but there's a potential issue: if the url already contains a path, setPath() will replace it entirely instead of appending to it. Consider using appendPath() or combining the existing path with the new one to avoid unexpected behavior.

url only has the base path so I think we're ok here.

Copilot · 2025-11-25T22:09:14Z

.../inference/services/elastic/response/ElasticInferenceServiceAuthorizationResponseEntity.java

+        }
+
+        public Configuration(StreamInput in) throws IOException {
+            this(in.readOptionalString(), in.readOptionalVInt(), in.readOptionalString(), in.readGenericMap());


The readGenericMap() and writeGenericMap() methods can handle null values correctly, but in the constructor you're calling in.readGenericMap() which will never return null (it returns an empty map for null). This means chunkingSettings will never actually be null after deserialization, which is inconsistent with the field being marked as @Nullable. Consider updating the logic to use in.readOptionalMap() or document that null becomes an empty map.

Suggested change

this(in.readOptionalString(), in.readOptionalVInt(), in.readOptionalString(), in.readGenericMap());

this(in.readOptionalString(), in.readOptionalVInt(), in.readOptionalString(), in.readOptionalMap());

Copilot · 2025-11-25T22:09:14Z

.../inference/services/elastic/response/ElasticInferenceServiceAuthorizationResponseEntity.java

+            out.writeOptionalString(similarity);
+            out.writeOptionalVInt(dimensions);
+            out.writeOptionalString(elementType);
+            out.writeGenericMap(chunkingSettings);


In line 264, writeGenericMap() is called with chunkingSettings, but according to the Javadoc, this method does not support null values and will throw a NullPointerException if the map is null. Since chunkingSettings is marked as @Nullable, this could cause serialization to fail. Use a null check before writing or use a method that handles null values.

Suggested change

out.writeGenericMap(chunkingSettings);

out.writeOptionalGenericMap(chunkingSettings);

writeGenericMap does support null values:

public void writeGenericMap(@Nullable Map<String, Object> map) throws IOException { writeGenericValue(map); }

Safety measure to make sure we've the correct default rerank endpoint in place in case elastic#138249 doesn't make it. Endpoint name: `.elastic-rerank-v1` -> `.jina-reranker-v2` Model name: `elastic-rerank-v1` -> `jina-reranker-v2`

...sterTest/java/org/elasticsearch/xpack/inference/integration/AuthorizationTaskExecutorIT.java

.../main/java/org/elasticsearch/xpack/inference/action/TransportGetInferenceServicesAction.java

DonalEvans · 2025-12-02T00:17:03Z

...c/main/java/org/elasticsearch/xpack/core/inference/action/StoreInferenceEndpointsAction.java


    public static class Request extends AcknowledgedRequest<Request> {
-        private final List<Model> models;
+        private final List<? extends Model> models;


You can avoid needing to use <? extends Model> in this PR by making a small change to ElasticInferenceServiceAuthorizationModel.getEndpoints():

public List<Model> getEndpoints(Set<String> endpointIds) { return endpointIds.stream().<Model>map(authorizedEndpoints::get).filter(Objects::nonNull).toList(); }

By letting the stream know that is should be a Stream<Model> after the map() call instead of it inferring the Stream<ElasticInferenceServiceModel> type, the return type can use List<Model>

...pack/inference/services/elastic/authorization/ElasticInferenceServiceAuthorizationModel.java

...g/elasticsearch/xpack/inference/services/elastic/authorization/AuthorizationPollerTests.java

...inference/services/elastic/authorization/ElasticInferenceServiceAuthorizationModelTests.java

DonalEvans · 2025-12-02T02:25:18Z

...rence/services/elastic/response/ElasticInferenceServiceAuthorizationResponseEntityTests.java

+    public static ElasticInferenceServiceAuthorizationResponseEntity.TaskTypeObject createTaskTypeObject(
+        String eisTaskType,
+        String elasticsearchTaskType
+    ) {
+        return new ElasticInferenceServiceAuthorizationResponseEntity.TaskTypeObject(eisTaskType, elasticsearchTaskType);
+    }


Does this method add much value? It might be simpler to just call the constructor directly in places that are currently calling this.

Yeah I think the reason I added it was because we changed the format of the task settings object a few times from a string to an object. The other minor benefit is that it removes the need for a long line. If you want me to remove it I can though.

Nah, no need to change it

...rence/services/elastic/response/ElasticInferenceServiceAuthorizationResponseEntityTests.java

…uth-v2

DonalEvans · 2025-12-03T00:38:27Z

.../main/java/org/elasticsearch/xpack/inference/action/TransportGetInferenceServicesAction.java

            threadPool.executor(UTILITY_THREAD_POOL_NAME).execute(() -> getEisAuthorization(authModelListener, eisSender));
        }).<List<InferenceServiceConfiguration>>andThen((configurationListener, authorizationModel) -> {
            var serviceConfigs = getServiceConfigurationsForServices(availableServices);
+            serviceConfigs.sort(Comparator.comparing(InferenceServiceConfiguration::getService));


Minor performance consideration; if we have both non-EIS and EIS services, we now sort the list twice. Maybe it would be better to combine the two if statements and sort within them before returning? Something like:

// If there was a requested task type and the authorization response from EIS doesn't support it, we'll exclude EIS as a valid // service if (authorizationModel.isAuthorized() == false || requestedTaskType != null && authorizationModel.getTaskTypes().contains(requestedTaskType) == false) { serviceConfigs.sort(Comparator.comparing(InferenceServiceConfiguration::getService)); configurationListener.onResponse(serviceConfigs); return; }

…uth-v2

jonathan-buttner added 3 commits November 14, 2025 16:51

Starting new response class

db5a7b3

Writing tests

5060aab

Fixing tests

6a31be2

jonathan-buttner added >non-issue :ml Machine learning Team:ML Meta label for the ML team v9.3.0 labels Nov 18, 2025

elasticsearchmachine and others added 16 commits November 18, 2025 16:51

[CI] Auto commit changes from spotless

70ef1a5

Successful tests

20d7881

Merge branch 'main' of github.com:elastic/elasticsearch into ml-eis-a…

4a56082

…uth-v2

Merge branch 'ml-eis-auth-v2' of github.com:jonathan-buttner/elastics…

057f15e

…earch into ml-eis-auth-v2

Removing unused code

2546292

Renaming

839ebe0

[CI] Auto commit changes from spotless

e7b2371

Working integration tests

f6b2d74

Fixing forbidden calls

f2a3d1a

Merge branch 'ml-eis-auth-v2' of github.com:jonathan-buttner/elastics…

3b370db

…earch into ml-eis-auth-v2

Merge branch 'main' of github.com:elastic/elasticsearch into ml-eis-a…

58361ea

…uth-v2

Fixing tests

b2aecd2

Merge branch 'main' of github.com:elastic/elasticsearch into ml-eis-a…

3c0e1b3

…uth-v2

Fixing integration tests

5715af6

Refactoring tests

7834662

Merge branch 'main' of github.com:elastic/elasticsearch into ml-eis-a…

cd4388a

…uth-v2

jonathan-buttner commented Nov 20, 2025

View reviewed changes

jonathan-buttner added 2 commits November 20, 2025 09:13

Adding some comments

a540427

Merge branch 'main' of github.com:elastic/elasticsearch into ml-eis-a…

92335ef

…uth-v2

jonathan-buttner marked this pull request as ready for review November 21, 2025 15:39

timgrein mentioned this pull request Nov 24, 2025

[Inference API] Rename default rerank model for EIS #138498

Merged

jonathan-buttner added 4 commits November 24, 2025 20:58

comments

b920069

Merge branch 'main' of github.com:elastic/elasticsearch into ml-eis-a…

e4e9095

…uth-v2

Adding support for completion

6b6d21b

Fixing tests

578546f

timgrein reviewed Nov 25, 2025

View reviewed changes

jonathan-buttner added 2 commits November 25, 2025 13:49

Addressing feedback

6aa0ea5

Merge branch 'main' of github.com:elastic/elasticsearch into ml-eis-a…

2379066

…uth-v2

jonathan-buttner requested a review from Copilot November 25, 2025 21:59

Copilot started reviewing on behalf of jonathan-buttner November 25, 2025 21:59 View session

Copilot finished reviewing on behalf of jonathan-buttner November 25, 2025 22:00

Copilot AI reviewed Nov 25, 2025

View reviewed changes

DonalEvans reviewed Dec 2, 2025

View reviewed changes

jonathan-buttner added 2 commits December 2, 2025 17:10

Merge branch 'main' of github.com:elastic/elasticsearch into ml-eis-a…

eb6def2

…uth-v2

Addressing feedback

4d3613b

jonathan-buttner requested review from DonalEvans and timgrein December 2, 2025 22:52

DonalEvans reviewed Dec 3, 2025

View reviewed changes

jonathan-buttner added 2 commits December 2, 2025 19:50

Refactoring into single if and removing listener

f8c2635

Merge branch 'main' of github.com:elastic/elasticsearch into ml-eis-a…

06f1f66

…uth-v2

jonathan-buttner added the cloud-deploy Publish cloud docker image for Cloud-First-Testing label Dec 3, 2025

Merge branch 'main' into ml-eis-auth-v2

7ed5393

prwhelan approved these changes Dec 4, 2025

View reviewed changes

Merge branch 'main' of github.com:elastic/elasticsearch into ml-eis-a…

9869cdd

…uth-v2

jonathan-buttner requested a review from DonalEvans December 4, 2025 20:03

DonalEvans approved these changes Dec 4, 2025

View reviewed changes

jonathan-buttner enabled auto-merge (squash) December 4, 2025 21:23

jonathan-buttner merged commit a51f7b7 into elastic:main Dec 4, 2025
35 of 36 checks passed

jonathan-buttner deleted the ml-eis-auth-v2 branch December 4, 2025 22:22

	* Transforms the response from {@link ElasticInferenceServiceAuthorizationRequestHandler} into a format for consumption by the service.
	* Transforms the response from {@link ElasticInferenceServiceAuthorizationRequestHandler} into a format for consumption by the {@link ElasticInferenceService}.

	public class AuthorizationModel {
	public class ElasticInferenceServiceAuthorizationModel {

	case SPARSE_EMBEDDING -> createSparseEmbeddingsModel(authorizedEndpoint, components);
	case SPARSE_EMBEDDING -> createSparseTextEmbeddingsModel(authorizedEndpoint, components);


		public record TaskTypeObject(String eisTaskType, String elasticsearchTaskType) implements Writeable, ToXContentObject {

		private static final String EIS_TASK_TYPE_FIELD = "eis";

	private static final String EIS_TASK_TYPE_FIELD = "eis";
	private static final String ELASTIC_INFERENCE_SERVICE_TASK_TYPE_FIELD = "eis";

-                    logger.warn("Found inference id collision for id [{}], ignoring second model", firstModel.inferenceEntityId());
+                    logger.warn(
+                        "Found inference id collision for id [{}]. Keeping model with id [{}] (taskType={}), discarding model with id [{}] (taskType={})",
+                        firstModel.getInferenceEntityId(),
+                        firstModel.getInferenceEntityId(),
+                        firstModel.getTaskType(),
+                        secondModel.getInferenceEntityId(),
+                        secondModel.getTaskType()
+                    );

	this(in.readOptionalString(), in.readOptionalVInt(), in.readOptionalString(), in.readGenericMap());
	this(in.readOptionalString(), in.readOptionalVInt(), in.readOptionalString(), in.readOptionalMap());

	out.writeGenericMap(chunkingSettings);
	out.writeOptionalGenericMap(chunkingSettings);

[ML] Use EIS v2 authorization endpoint for inference API #138249

[ML] Use EIS v2 authorization endpoint for inference API #138249

Uh oh!

Conversation

jonathan-buttner commented Nov 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Testing

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

elasticsearchmachine commented Nov 21, 2025

Uh oh!

timgrein left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jonathan-buttner Nov 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Nov 25, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 25, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 25, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 25, 2025

Choose a reason for hiding this comment

Uh oh!

jonathan-buttner commented Nov 18, 2025 •

edited

Loading

jonathan-buttner Nov 25, 2025 •

edited

Loading