ESQL: Skip unused STATS groups by adding a Top N BlockHash implementation #127148

ivancea · 2025-04-22T12:58:12Z

Current behaviour

In queries like: FROM x | STATS SUM(y) BY group | SORT group | LIMIT 100
We are currently aggregating values for all the values of group, and then ignoring them at the SORT+LIMIT part.

New behaviour

We can safely ignore groups that aren't part of that "Top N", fully skipping the aggregation part, and avoiding sending the values to the coordinator.
For that, we have some steps to do:

Add a new BlockHash implementation that keeps such "Top" values and ignores everything else. The BlockHashes are the ones actually calling the aggs through a callback, so this step covers most of the compute engine work.
At the language side, we have to detect such Stats+TopN, and inject that information into the HashAggregationOperator (Or something like that). The operator receives a BlockHash supplier as a constructor parameter, so most work may be in the rules/mapping side
Tests and benchmarks
Repeat most of the points for every ElementTypes we want to cover in the BlockHash

In this PR

This PR will cover a first usecase (a single Long group) in the compute side. It will:

Add a new LongTopNBlockHash implementation taking care of skipping unused values
- Add a TopNUniqueSet to take care of storing the top N values (without nulls)
- Add a TopNMultivalueDedupeLong class helping with it (An adaptation of the existing MultivalueDedupeLong)
Add some tests to HashAggregationOperator. It wasn't changed much, but helps a bit with the E2E
Add MicroBenchmarks for TopN groupings, to ensure we're actually improving things with this

Benchmarks

Some benchmarks, to understand the affectation:

[1.000.000 rows, 100.000 groups, top 100 (0.1%)]
Benchmark                 (grouping)            (op)    Score  Error  Units
AggregatorBenchmark.run        longs             sum   31.968  0.646  ns/op
AggregatorBenchmark.run        longs  count_distinct  101.675 15.180  ns/op
AggregatorBenchmark.run  top_n_longs             sum   12.335  0.572  ns/op
AggregatorBenchmark.run  top_n_longs  count_distinct   12.664  0.130  ns/op


[1.000.000 rows, 100.000 groups, top 10.000 (10%)]
Benchmark                 (grouping)            (op)    Score  Error  Units
AggregatorBenchmark.run        longs             sum   93.090  6.776  ns/op
AggregatorBenchmark.run        longs  count_distinct  102.230 15.631  ns/op
AggregatorBenchmark.run  top_n_longs             sum   25.084  4.752  ns/op
AggregatorBenchmark.run  top_n_longs  count_distinct   32.591  1.103  ns/op


[1.000.000 rows, 100.000 groups, top 100.000 (100%)]
Benchmark                 (grouping)            (op)    Score   Error  Units
AggregatorBenchmark.run        longs             sum   93.898   6.845  ns/op
AggregatorBenchmark.run        longs  count_distinct  129.644 375.625  ns/op
AggregatorBenchmark.run  top_n_longs             sum  147.875   1.621  ns/op
AggregatorBenchmark.run  top_n_longs  count_distinct  209.728   3.630  ns/op

… in hash

elasticsearchmachine · 2025-04-22T12:59:00Z

Hi @ivancea, I've created a changelog YAML for you.

ivancea · 2025-04-22T13:26:28Z

...compute/src/test/java/org/elasticsearch/compute/aggregation/blockhash/BlockHashTestCase.java

Just moved some code from BlockHashTests to here, as it's being shared with the new implementation

ivancea · 2025-04-22T13:27:25Z

...compute/src/test/java/org/elasticsearch/compute/aggregation/blockhash/BlockHashTestCase.java

+            for (int c = 0; c < actualKeys.length; c++) {
+                if (expectedKeys[r][c] == null) {
+                    assertThat("expected null key", actualKeys[c].isNull(r), equalTo(true));
+                    continue;


This was a return before, which was skipping the checks on all further values. I fixed it, and a BlockHashTest case that failed because of it

ivancea · 2025-04-22T13:28:11Z

...ql/compute/src/test/java/org/elasticsearch/compute/aggregation/blockhash/BlockHashTests.java

                        new Object[][] {
                            new Object[] { 1L, "cat" },
                            new Object[] { null, null },
                            new Object[] { 0L, "dog" },
-                            new Object[] { 1L, null },
+                            new Object[] { 0L, null },


This was the failing test. I checked it, and it this typo

ivancea · 2025-04-22T13:30:14Z

...l/compute/src/test/java/org/elasticsearch/compute/operator/HashAggregationOperatorTests.java

@@ -96,4 +101,154 @@ protected void assertSimpleOutput(List<Page> input, List<Page> results) {
            max.assertSimpleGroup(input, maxs, i, group);
        }
    }
+
+    public void testTopNNullsLast() {


These tests are quite shallow, but as the BlockHash is also tested, and no Operator changes were made, I think this is fine for now (?).
When we link the new TopNBlockHash with the language side, I'll review this again

👍

I could see us doing a lot more randomized fun to this, out of paranoia.

nik9000 · 2025-04-22T13:50:19Z

...compute/src/main/java/org/elasticsearch/compute/aggregation/blockhash/LongTopNBlockHash.java

+        }
+
+        // TODO: Make the custom dedupe *less custom*
+        MultivalueDedupe.HashResult result = new TopNMultivalueDedupeLong(block, hasNull, this::isAcceptable).hashAdd(blockFactory, hash);


MultivalueDedupe might be what you need.

Adding here: This TopNMultivalueDedupeLong is currently a plain copy of that class, with the methods slightly changed to work with the TopN logic.
I'll see if I can merge it into the original one. Otherwise, I may trim the unused parts, and leave it as a new class. Or a nested class

nik9000 · 2025-04-22T13:50:28Z

...compute/src/main/java/org/elasticsearch/compute/aggregation/blockhash/LongTopNBlockHash.java

+        for (int i = 0; i < positions; i++) {
+            long v = vector.getLong(i);
+            acceptValue(v);
+        }


…cs, and added it to benchmark

…ests

ivancea · 2025-05-28T15:14:45Z

...in/esql/compute/src/main/java/org/elasticsearch/compute/aggregation/blockhash/BlockHash.java

    /**
     * @param isCategorize Whether this group is a CATEGORIZE() or not.
     *                     May be changed in the future when more stateful grouping functions are added.
     */
-    public record GroupSpec(int channel, ElementType elementType, boolean isCategorize) {
+    public record GroupSpec(int channel, ElementType elementType, boolean isCategorize, @Nullable TopNDef topNDef) {


This is the connection between the QL and compute code. It's always null now; it should be filled with info from the nearest affecting TopN

ivancea · 2025-05-29T11:24:43Z

...pute/src/main/java/org/elasticsearch/compute/operator/mvdedupe/TopNMultivalueDedupeLong.java

This class is a clone of the original one, with the extra filtering to ignore values out of the top.
I just removed all the unused methods.

Depending on your thoughts on it, the next steps would be:
(In this PR)

Clone and adapt the tests too
(In next PRs)

Convert to a stringtemplate to make it work with other types

We may need to restore some methods to work with packed values. But only if we implement packed values for the TopN optimization

ivancea · 2025-05-29T12:31:24Z

...k/plugin/esql/compute/src/test/java/org/elasticsearch/compute/data/sort/TopNSetTestCase.java

+
+import static org.hamcrest.Matchers.equalTo;
+
+public abstract class TopNSetTestCase<T extends Releasable, V extends Comparable<V>> extends ESTestCase {


Similar structure to BucketedSortTestCase

elasticsearchmachine · 2025-05-29T13:12:17Z

Pinging @elastic/es-analytical-engine (Team:Analytics)

nik9000

LGTM.

It's a step towards greatness! Hopefully.

nik9000 · 2025-06-10T16:52:41Z

...in/esql/compute/src/main/java/org/elasticsearch/compute/aggregation/blockhash/BlockHash.java

    /**
     * @param isCategorize Whether this group is a CATEGORIZE() or not.
     *                     May be changed in the future when more stateful grouping functions are added.
     */
-    public record GroupSpec(int channel, ElementType elementType, boolean isCategorize) {
+    public record GroupSpec(int channel, ElementType elementType, boolean isCategorize, @Nullable TopNDef topNDef) {


nik9000 · 2025-06-10T16:56:03Z

...l/compute/src/test/java/org/elasticsearch/compute/operator/HashAggregationOperatorTests.java

@@ -96,4 +101,154 @@ protected void assertSimpleOutput(List<Page> input, List<Page> results) {
            max.assertSimpleGroup(input, maxs, i, group);
        }
    }
+
+    public void testTopNNullsLast() {


👍

I could see us doing a lot more randomized fun to this, out of paranoia.

…tion (elastic#127148) - Add a new `LongTopNBlockHash` implementation taking care of skipping unused values. - Add a `TopNUniqueSet` to take care of storing the top N values (without nulls). - Add a `TopNMultivalueDedupeLong` class helping with it (An adaptation of the existing `MultivalueDedupeLong`). - Add some tests to `HashAggregationOperator`. It wasn't changed much, but helps a bit with the E2E. - Add MicroBenchmarks for TopN groupings, to ensure we're actually improving things with this.

* propgating retrievers to inner retrievers * test feature taken care of * Small changes in concurrent multipart upload interfaces (#128977) Small changes in BlobContainer interface and wrapper. Relates ES-11815 * Unmute FollowingEngineTests#testProcessOnceOnPrimary() test (#129054) The reason the test fails is that operations contained _seq_no field with different doc value types (with no skippers and with skippers) and this isn't allowed, since field types need to be consistent in a Lucene index. The initial operations were generated not knowing about the fact the index mode was set to logsdb or time_series. Causing the operations to not have doc value skippers. However when replaying the operations via following engine, the operations did have doc value skippers. The fix is to set `index.seq_no.index_options` to `points_and_doc_values`, so that the initial operations are indexed without doc value skippers. This test doesn't gain anything from storing seqno with doc value skippers, so there is no loss of testing coverage. Closes #128541 * [Build] Add support for publishing to maven central (#128659) This ensures we package an aggregation zip with all artifacts we want to publish to maven central as part of a release. Running zipAggregation will produce a zip file in the build/nmcp/zip folder. The content of this zip is meant to match the maven artifacts we have currently declared as dra maven artifacts. * ESQL: Check for errors while loading blocks (#129016) Runs a sanity check after loading a block of values. Previously we were doing a quick check if assertions were enabled. Now we do two quick checks all the time. Better - we attach information about how a block was loaded when there's a problem. Relates to #128959 * Make `PhaseCacheManagementTests` project-aware (#129047) The functionality in `PhaseCacheManagement` was already project-aware, but these tests were still using deprecated methods. * Vector test tools (#128934) This adds some testing tools for verifying vector recall and latency directly without having to spin up an entire ES node and running a rally track. Its pretty barebones and takes inspiration from lucene-util, but I wanted access to our own formats and tooling to make our lives easier. Here is an example config file. This will build the initial index, run queries at num_candidates: 50, then again at num_candidates 100 (without reindexing, and re-using the cached nearest neighbors). ``` [{ "doc_vectors" : "path", "query_vectors" : "path", "num_docs" : 10000, "num_queries" : 10, "index_type" : "hnsw", "num_candidates" : 50, "k" : 10, "hnsw_m" : 16, "hnsw_ef_construction" : 200, "index_threads" : 4, "reindex" : true, "force_merge" : false, "vector_space" : "maximum_inner_product", "dimensions" : 768 }, { "doc_vectors" : "path", "query_vectors" : "path", "num_docs" : 10000, "num_queries" : 10, "index_type" : "hnsw", "num_candidates" : 100, "k" : 10, "hnsw_m" : 16, "hnsw_ef_construction" : 200, "vector_space" : "maximum_inner_product", "dimensions" : 768 } ] ``` To execute: ``` ./gradlew :qa:vector:checkVec --args="/Path/to/knn_tester_config.json" ``` Calling `./gradlew :qa:vector:checkVecHelp` gives some guidance on how to use it, additionally providing a way to run it via java directly (useful to bypass gradlew guff). * ES|QL: refactor generative tests (#129028) * Add a test of LOOKUP JOIN against a time series index (#129007) Add a spec test of `LOOKUP JOIN` against a time series index. * Make ILM `ClusterStateWaitStep` project-aware (#129042) This is part of an iterative process to make ILM project-aware. * Mute org.elasticsearch.xpack.esql.qa.mixed.MixedClusterEsqlSpecIT test {lookup-join.LookupJoinOnTimeSeriesIndex ASYNC} #129078 * Remove `ClusterState` param from ILM `AsyncBranchingStep` (#129076) The `ClusterState` parameter of the `asyncPredicate` is not used anywhere. * Mute org.elasticsearch.xpack.esql.qa.mixed.MixedClusterEsqlSpecIT test {lookup-join.LookupJoinOnTimeSeriesIndex SYNC} #129082 * Mute org.elasticsearch.upgrades.UpgradeClusterClientYamlTestSuiteIT test {p0=upgraded_cluster/70_ilm/Test Lifecycle Still There And Indices Are Still Managed} #129097 * Mute org.elasticsearch.upgrades.UpgradeClusterClientYamlTestSuiteIT test {p0=upgraded_cluster/90_ml_data_frame_analytics_crud/Get mixed cluster outlier_detection job} #129098 * Mute org.elasticsearch.packaging.test.DockerTests test081SymlinksAreFollowedWithEnvironmentVariableFiles #128867 * Threadpool merge executor is aware of available disk space (#127613) This PR introduces 3 new settings: indices.merge.disk.check_interval, indices.merge.disk.watermark.high, and indices.merge.disk.watermark.high.max_headroom that control if the threadpool merge executor starts executing new merges when the disk space is getting low. The intent of this change is to avoid the situation where in-progress merges exhaust the available disk space on the node's local filesystem. To this end, the thread pool merge executor periodically monitors the available disk space, as well as the current disk space estimates required by all in-progress (currently running) merges on the node, and will NOT schedule any new merges if the disk space is getting low (by default below the 5% limit of the total disk space, or 100 GB, whichever is smaller (same as the disk allocation flood stage level)). * Add option to include or exclude vectors from _source retrieval (#128735) This PR introduces a new include_vectors option to the _source retrieval context. When set to false, vectors are excluded from the returned _source. This is especially efficient when used with synthetic source, as it avoids loading vector fields entirely. By default, vectors remain included unless explicitly excluded. * Remove direct minScore propagation to inner retrievers * cleaned up skip * Mute org.elasticsearch.index.engine.ThreadPoolMergeExecutorServiceDiskSpaceTests testAvailableDiskSpaceMonitorWhenFileSystemStatErrors #129149 * Add transport version for ML inference Mistral chat completion (#129033) * Add transport version for ML inference Mistral chat completion * Add changelog for Mistral Chat Completion version fix * Revert "Add changelog for Mistral Chat Completion version fix" This reverts commit 7a57416. * Correct index path validation (#129144) All we care about is if reindex is true or false. We shouldn't worry about force merge. Because if reindex is true, we will create the directory, if its false, we won't. * Mute org.elasticsearch.index.engine.ThreadPoolMergeExecutorServiceDiskSpaceTests testUnavailableBudgetBlocksNewMergeTasksFromStartingExecution #129148 * Implemented completion task for Google VertexAI (#128694) * Google Vertex AI completion model, response entity and tests * Fixed GoogleVertexAiServiceTest for Service configuration * Changelog * Removed downcasting and using `moveToFirstToken` * Create GoogleVertexAiChatCompletionResponseHandler for streaming and non streaming responses * Added unit tests * PR feedback * Removed googlevertexaicompletion model. Using just GoogleVertexAiChatCompletionModel for completion and chat completion * Renamed uri -> nonStreamingUri. Added streamingUri and getters in GoogleVertexAiChatCompletionModel * Moved rateLimitGroupHashing to subclasses of GoogleVertexAiModel * Fixed rate limit has of GoogleVertexAiRerankModel and refactored uri for GoogleVertexAiUnifiedChatCompletionRequest --------- Co-authored-by: lhoet-google <lhoet@google.com> Co-authored-by: Jonathan Buttner <56361221+jonathan-buttner@users.noreply.github.com> * ES|QL - kNN function initial support (#127322) * Remove optional seed from ES|QL SAMPLE (#128887) * Remove optional seed from ES|QL SAMPLE * make it clear that seed is for testing * [Inference API] Add "rerank" task type to "elastic" provider (#126022) * Rename target destination for microbenchmarks (#128878) * Include direct memory and non-heap memory in ML memory calculations (take #2) (#128742) * Include direct memory and non-heap memory in ML memory calculations. * Reduce ML_ONLY heap size, so that direct memory is accounted for. * [CI] Auto commit changes from spotless * changelog * improve docs * Reuse direct memory to heap factor * feature flag --------- Co-authored-by: elasticsearchmachine <infra-root+elasticsearchmachine@elastic.co> * Throw better exception for unsupported aggregations over shape fields (#129139) * Update Test Framework To Handle Query Rewrites That Rely on Non-Null Searchers (#129160) * Update ReproduceInfoPrinter to correctly print a reproduction line for Lucene & build candidate upgrade tests (#129044) * Increment inference stats counter for shard bulk inference calls (#129140) This change updates the inference stats counter to include chunked inference calls performed by the shard bulk inference filter on all semantic text fields. It ensures that usage of inference on semantic text fields is properly recorded in the stats. * Synthetic source: avoid storing multi fields of type text and match_only_text by default. (#129126) Don't store text and match_only_text field by default when source mode is synthetic and a field is a multi field or when there is a suitable multi field. Without this change, ES would store field otherwise twice in a multi-field configuration. For example: ``` ... "os": { "properties": { "name": { "ignore_above": 1024, "type": "keyword", "fields": { "text": { "type": "match_only_text" } } } ... ``` In this case, two stored fields were added, one in case for the `name` field and one for `name.text` multi-field. This change prevents this, and would never store a stored field when text or match_only_text field is a multi-field. * Adding `scheduled_report_id` field to kibana reporting template (#127827) * Adding scheduled_report_id field to kibana reporting template * Incrementing stack template registry version * ES|QL: Add FORK generative tests (#129135) * ES|QL Completion command syntax change (#129189) * Remove optional seed from ES|QL SAMPLE (#128887) * Remove optional seed from ES|QL SAMPLE * make it clear that seed is for testing * ES|QL Completion command syntax change (#129189) * Remove optional seed from ES|QL SAMPLE (#128887) * Remove optional seed from ES|QL SAMPLE * make it clear that seed is for testing * ES|QL Completion command syntax change (#129189) * Add Cluster Feature for L2 Norm (#129181) * propgating retrievers to inner retrievers * test feature taken care of * Small changes in concurrent multipart upload interfaces (#128977) Small changes in BlobContainer interface and wrapper. Relates ES-11815 * Unmute FollowingEngineTests#testProcessOnceOnPrimary() test (#129054) The reason the test fails is that operations contained _seq_no field with different doc value types (with no skippers and with skippers) and this isn't allowed, since field types need to be consistent in a Lucene index. The initial operations were generated not knowing about the fact the index mode was set to logsdb or time_series. Causing the operations to not have doc value skippers. However when replaying the operations via following engine, the operations did have doc value skippers. The fix is to set `index.seq_no.index_options` to `points_and_doc_values`, so that the initial operations are indexed without doc value skippers. This test doesn't gain anything from storing seqno with doc value skippers, so there is no loss of testing coverage. Closes #128541 * [Build] Add support for publishing to maven central (#128659) This ensures we package an aggregation zip with all artifacts we want to publish to maven central as part of a release. Running zipAggregation will produce a zip file in the build/nmcp/zip folder. The content of this zip is meant to match the maven artifacts we have currently declared as dra maven artifacts. * ESQL: Check for errors while loading blocks (#129016) Runs a sanity check after loading a block of values. Previously we were doing a quick check if assertions were enabled. Now we do two quick checks all the time. Better - we attach information about how a block was loaded when there's a problem. Relates to #128959 * Make `PhaseCacheManagementTests` project-aware (#129047) The functionality in `PhaseCacheManagement` was already project-aware, but these tests were still using deprecated methods. * Vector test tools (#128934) This adds some testing tools for verifying vector recall and latency directly without having to spin up an entire ES node and running a rally track. Its pretty barebones and takes inspiration from lucene-util, but I wanted access to our own formats and tooling to make our lives easier. Here is an example config file. This will build the initial index, run queries at num_candidates: 50, then again at num_candidates 100 (without reindexing, and re-using the cached nearest neighbors). ``` [{ "doc_vectors" : "path", "query_vectors" : "path", "num_docs" : 10000, "num_queries" : 10, "index_type" : "hnsw", "num_candidates" : 50, "k" : 10, "hnsw_m" : 16, "hnsw_ef_construction" : 200, "index_threads" : 4, "reindex" : true, "force_merge" : false, "vector_space" : "maximum_inner_product", "dimensions" : 768 }, { "doc_vectors" : "path", "query_vectors" : "path", "num_docs" : 10000, "num_queries" : 10, "index_type" : "hnsw", "num_candidates" : 100, "k" : 10, "hnsw_m" : 16, "hnsw_ef_construction" : 200, "vector_space" : "maximum_inner_product", "dimensions" : 768 } ] ``` To execute: ``` ./gradlew :qa:vector:checkVec --args="/Path/to/knn_tester_config.json" ``` Calling `./gradlew :qa:vector:checkVecHelp` gives some guidance on how to use it, additionally providing a way to run it via java directly (useful to bypass gradlew guff). * ES|QL: refactor generative tests (#129028) * Add a test of LOOKUP JOIN against a time series index (#129007) Add a spec test of `LOOKUP JOIN` against a time series index. * Make ILM `ClusterStateWaitStep` project-aware (#129042) This is part of an iterative process to make ILM project-aware. * Mute org.elasticsearch.xpack.esql.qa.mixed.MixedClusterEsqlSpecIT test {lookup-join.LookupJoinOnTimeSeriesIndex ASYNC} #129078 * Remove `ClusterState` param from ILM `AsyncBranchingStep` (#129076) The `ClusterState` parameter of the `asyncPredicate` is not used anywhere. * Mute org.elasticsearch.xpack.esql.qa.mixed.MixedClusterEsqlSpecIT test {lookup-join.LookupJoinOnTimeSeriesIndex SYNC} #129082 * Mute org.elasticsearch.upgrades.UpgradeClusterClientYamlTestSuiteIT test {p0=upgraded_cluster/70_ilm/Test Lifecycle Still There And Indices Are Still Managed} #129097 * Mute org.elasticsearch.upgrades.UpgradeClusterClientYamlTestSuiteIT test {p0=upgraded_cluster/90_ml_data_frame_analytics_crud/Get mixed cluster outlier_detection job} #129098 * Mute org.elasticsearch.packaging.test.DockerTests test081SymlinksAreFollowedWithEnvironmentVariableFiles #128867 * Threadpool merge executor is aware of available disk space (#127613) This PR introduces 3 new settings: indices.merge.disk.check_interval, indices.merge.disk.watermark.high, and indices.merge.disk.watermark.high.max_headroom that control if the threadpool merge executor starts executing new merges when the disk space is getting low. The intent of this change is to avoid the situation where in-progress merges exhaust the available disk space on the node's local filesystem. To this end, the thread pool merge executor periodically monitors the available disk space, as well as the current disk space estimates required by all in-progress (currently running) merges on the node, and will NOT schedule any new merges if the disk space is getting low (by default below the 5% limit of the total disk space, or 100 GB, whichever is smaller (same as the disk allocation flood stage level)). * Add option to include or exclude vectors from _source retrieval (#128735) This PR introduces a new include_vectors option to the _source retrieval context. When set to false, vectors are excluded from the returned _source. This is especially efficient when used with synthetic source, as it avoids loading vector fields entirely. By default, vectors remain included unless explicitly excluded. * Remove direct minScore propagation to inner retrievers * cleaned up skip * Mute org.elasticsearch.index.engine.ThreadPoolMergeExecutorServiceDiskSpaceTests testAvailableDiskSpaceMonitorWhenFileSystemStatErrors #129149 * Add transport version for ML inference Mistral chat completion (#129033) * Add transport version for ML inference Mistral chat completion * Add changelog for Mistral Chat Completion version fix * Revert "Add changelog for Mistral Chat Completion version fix" This reverts commit 7a57416. * Correct index path validation (#129144) All we care about is if reindex is true or false. We shouldn't worry about force merge. Because if reindex is true, we will create the directory, if its false, we won't. * Mute org.elasticsearch.index.engine.ThreadPoolMergeExecutorServiceDiskSpaceTests testUnavailableBudgetBlocksNewMergeTasksFromStartingExecution #129148 * Implemented completion task for Google VertexAI (#128694) * Google Vertex AI completion model, response entity and tests * Fixed GoogleVertexAiServiceTest for Service configuration * Changelog * Removed downcasting and using `moveToFirstToken` * Create GoogleVertexAiChatCompletionResponseHandler for streaming and non streaming responses * Added unit tests * PR feedback * Removed googlevertexaicompletion model. Using just GoogleVertexAiChatCompletionModel for completion and chat completion * Renamed uri -> nonStreamingUri. Added streamingUri and getters in GoogleVertexAiChatCompletionModel * Moved rateLimitGroupHashing to subclasses of GoogleVertexAiModel * Fixed rate limit has of GoogleVertexAiRerankModel and refactored uri for GoogleVertexAiUnifiedChatCompletionRequest --------- Co-authored-by: lhoet-google <lhoet@google.com> Co-authored-by: Jonathan Buttner <56361221+jonathan-buttner@users.noreply.github.com> * Added cluster feature to yaml * Node feature added * Duplicate line - result of merge removed * Update docs/changelog/129181.yaml * Update 129181.yaml --------- Co-authored-by: Tanguy Leroux <tlrx.dev@gmail.com> Co-authored-by: Martijn van Groningen <martijn.v.groningen@gmail.com> Co-authored-by: Rene Groeschke <rene@elastic.co> Co-authored-by: Nik Everett <nik9000@gmail.com> Co-authored-by: Niels Bauman <33722607+nielsbauman@users.noreply.github.com> Co-authored-by: Benjamin Trent <ben.w.trent@gmail.com> Co-authored-by: Luigi Dell'Aquila <luigi.dellaquila@gmail.com> Co-authored-by: Bogdan Pintea <bogdan.pintea@elastic.co> Co-authored-by: elasticsearchmachine <58790826+elasticsearchmachine@users.noreply.github.com> Co-authored-by: Albert Zaharovits <email+github@zalbert.me> Co-authored-by: Jim Ferenczi <jim.ferenczi@elastic.co> Co-authored-by: Jan-Kazlouski-elastic <jan.kazlouski@elastic.co> Co-authored-by: Leonardo Hoet <55866308+leo-hoet@users.noreply.github.com> Co-authored-by: lhoet-google <lhoet@google.com> Co-authored-by: Jonathan Buttner <56361221+jonathan-buttner@users.noreply.github.com> * Fix DRA dependenciesInfo task dependency resolution (#129209) * IVF Hierarchical KMeans Flush & Merge (#128675) added hierarchical kmeans as a clustering algorithm to better partitionin the space when running ivf on flush and merge * Mute org.elasticsearch.xpack.esql.qa.single_node.GenerativeForkIT test {lookup-join.EnrichLookupStatsBug ASYNC} #129228 * Mute org.elasticsearch.xpack.esql.qa.single_node.GenerativeForkIT test {lookup-join.EnrichLookupStatsBug SYNC} #129229 * [ES|QL] Specify population in StdDev docs (#129225) There are 2 types of Standard Deviation: population and sample, this commit clarifies that the existing is population. * Unmute IngestGeoIpClientYamlTestSuiteIT (#129178) * Fix an NPE in the ES|QL completion command. (#129235) * ESQL: fix bwc test by adding min required version (#129204) Closes #129093 Closes #129094 Closes #129095 Closes #129102 Closes #129103 * ESQL: Fix test by add excluding capability (#129202) Closes #129078 Closes #129082 * Fix vault field name (#129184) * Remove all usages of Metadata customs removal methods (#129043) This removes all non-test usage of ``` Metadata.Builder.removeProjectCustom(String) Metadata.Builder.removeProjectCustomIf(BiPredicate) ``` And replaces it with appropriate calls to the equivalent method on `ProjectMetadata.Builder` In most cases this _does not_ make the code project aware, but does reduce the number of deprecated methods in use. * Replace tuple with record (#128976) * improve support for bytecode patching signed jars (#128613) * improve support for bytecode patching signed jars * Update docs/changelog/128613.yaml --------- Co-authored-by: elasticsearchmachine <infra-root+elasticsearchmachine@elastic.co> Co-authored-by: Johannes Freden Jansson <johannes.freden@elastic.co> * rename ES|QL sample capability (#129193) * ESQL: Mute GenerativeForkIT for some LOOKUP JOIN tests (#129248) * ESQL: Extend `RENAME` syntax to allow a `new = old` syntax (#129212) This extends RENAME's grammar to allow a new syntax: `| RENAME new_name = old_name` This is supported along the existing `... old_name AS new_name` syntax. Closes #129208 * [DOCS] Adds preview tag to the CHANGE_POINT ES|QL command in the command list. (#129247) * ESQL: Skip unused STATS groups by adding a Top N BlockHash implementation (#127148) - Add a new `LongTopNBlockHash` implementation taking care of skipping unused values. - Add a `TopNUniqueSet` to take care of storing the top N values (without nulls). - Add a `TopNMultivalueDedupeLong` class helping with it (An adaptation of the existing `MultivalueDedupeLong`). - Add some tests to `HashAggregationOperator`. It wasn't changed much, but helps a bit with the E2E. - Add MicroBenchmarks for TopN groupings, to ensure we're actually improving things with this. * Add "Searchable Snapshots" to changelog validation schema (#129180) We created a new ":Distributed Indexing/Searchable Snapshots" label recently on Github, so I think it makes sense to also have a "Searchable Snapshots" label in the changelog. It also makes sense since there is automatic changelog generation based on the pull request label. * ESQL: Fix FieldAttribute name usage in InferNonNullAggConstraint (#128910) * Fix InferNonNullAggConstraint with union types * Begin fixing LucenePushdownPredicates with union types * Introduce a dedicated wrapper record FieldName to be used where field names are really required. The fixes consist of using FieldAttribute.fieldName() instead of .name() or .field().name(). .name() can be some temporary string unrelated to the actual name of the Lucene index field, whereas .field().name() doesn't know about parent fields; .fieldName() gives the full field name (from the root of the document). The biggest offender of such misuse is SearchStats; make this always require a FieldName, not a String - and make FieldAttribute#fieldName handily return an instance of FieldName so users of SearchStats don't accidentally use the return value of FieldAttribute#name. * Remove usages of `Metadata.Builder#indexGraveyard` (#129041) And replace it with appropriate calls to the equivalent method on `ProjectMetadata.Builder`. In most cases this _does not_ make the code project aware, but does reduce the number of deprecated methods in use. Concerns both the getter and the setter. * Mute org.elasticsearch.compute.data.sort.LongTopNSetTests testCrankyBreaker #129257 * Enable Shard-Level Search-load rate metric (#128660) Introduces a new search load metric to the stats infrastructure, measured and tracked on a per-shard basis. The metric represents the Exponentially Weighted Moving Rate (EWMR) of search operations, calculated using the "took" time from each completed search phase. * [ESQL] Fix typo in search-functions.md (#129260) ^^ * ESQL: Log partial failures (#129164) Now that ESQL has `allow_partial_results` we can reply with a `200` even though some nodes failed to run ESQL. This could happen because the node is restarting. Or because of a bug. Or a disconnect. All kinds of things. This logs those partial failures so an operator can look at them and get a sense of why they are happening. * Update Gradle wrapper to 8.14.2 (#129179) * Fix ivf nodestats impl for getOffHeapByteSize (#129259) This fixes a silly bug where we didn't override `OffHeapStats` for IVF. * feat: enable date_detection for all apm data streams (#128913) * feat: enable date_detection for all apm data streams * Update resources.yaml * Create 128913.yml --------- Co-authored-by: Carson Ip <carsonip@users.noreply.github.com> * [BC Upgrage] Fix incorrect version parsing in tests (#129243) This PR introduces several fixes to various IT tests, related to the use and misuse of the version identifier for the start cluster: wherever we can, we replace of versions in test code with features where we can't, we make sure we use the actual stack version (the one provided by -Dtests.bwc.main.version and not the bogus "0.0.0" version string) when requesting the cluster version we make sure we do use the "unresolved" version identifier (the value of the tests.old_cluster_version system property e.g. 0.0.0 ) so we resolve the right distribution These changes enabled the tests to be used in BC upgrade tests (and potentially in serverless upgrade tests too, where they would have also failed) Relates to ES-12010 Precedes #128614, #128823 and #128983 * [Build] Build maven aggregation zip as part of DRA build (#129175) * [Build] Build maven aggregation zip as part of DRA build * Update path for aggregation zip * Throttle indexing when disk IO throttling is disabled (#129245) The threadpool-based merge scheduler triggers indexing throttling if merges are still getting enqueued faster than they're executed, while they are also disk IO unthrottled. This PR fixes the case where indexing throttling was incorrectly NOT triggered when disk IO throttling was disabled via the index settings. * Register match_phrase as a function not a snapshot function (#129255) * Register match_phrase as a function not a snapshot function * Update usage * [Gradle] Spotless plugin update (#115750) - provides better configuration cache support - requires some rework due to changed defaults * Adding support to exclude semantic_text subfields (#127664) * Adding support to exclude semantic_text subfields * Update docs/changelog/127664.yaml * Updating changelog file * remove duplicate test from yaml file * Adding support to exclude semantic_text subfields from mapper builders * Adding support for generic field types * refactoring to use builder and setting exclude value from semantic_text mapper * update in semantic_text mapper and fetcher to incorporate the support functionality * Fix code style issue * adding node feature for yaml tests * Adding more restrictive checks on yaml tests and few refactoring * Returns metadata fields from metadata mappers * returns all source fields for fieldcaps * gather all fields and iterate to process for fieldcaps api * revert back all changes from MappedFieldtype and subclasses * revert back exclude logic from semantic_text mapper * fix lint issues * fix lint issues * Adding runtime fields into fieldCaps * Fix linting issue * removing unused functions that used in previous implementation * fix multifield tests failure * getting alias fields for field caps * adding support for query time runtime fields * [CI] Auto commit changes from spotless * Fix empty mapping fieldCaps call * Address passthrough behavior for mappers * Fix SearchAsYoutype mapper failures * rename abstract method to have more meaningful name * Rename mapper function to match its functionality * Adding filtering for infernece subfields * revert back previous implementation changes * Adding yaml test for field caps not filtering multi-field * Fixing yaml test * Adding comment why .infernece filter is added --------- Co-authored-by: elasticsearchmachine <infra-root+elasticsearchmachine@elastic.co> Co-authored-by: Elastic Machine <elasticmachine@users.noreply.github.com> * Revert "[Gradle] Spotless plugin update (#115750)" This reverts commit 6370d60. * Switch IVF Writer to ES Logger (#129224) update to use ES logger instead of infostream and fixing native access warnings * Add heap usage estimate to ClusterInfo (#128723) Co-authored-by: ywangd <yang.wang@elastic.co> Co-authored-by: rjernst <ryan@elastic.co> Relates: ES-11445 * Revert "Use IndexOrDocValuesQuery in NumberFieldType#termQuery implementations (#128293)" (#129206) This reverts commit de7c91c. * Delegated authorization using Microsoft Graph (SDK) (#128396) * Delegated authorization using Microsoft Graph (SDK) --------- Co-authored-by: elasticsearchmachine <infra-root+elasticsearchmachine@elastic.co> Co-authored-by: Johannes Freden Jansson <johannes.freden@elastic.co> Co-authored-by: Johannes Fredén <109296772+jfreden@users.noreply.github.com> * Add `none` chunking strategy to disable automatic chunking for inference endpoints (#129150) This introduces a `none` chunking strategy that disables automatic chunking when using an inference endpoint. It enables users to provide pre-chunked input directly to a `semantic_text` field without any additional splitting. The chunking strategy can be configured either on the inference endpoint or directly in the `semantic_text` field definition. **Example:** ```json PUT test-index { "mappings": { "properties": { "my_semantic_field": { "type": "semantic_text", "chunking_settings": { "strategy": "none" <1> } } } } } ``` <1> Disables automatic chunking on `my_semantic_field`. ```json PUT test-index/_doc/1 { "my_semantic_field": ["my first chunk", "my second chunk", ...] <1> ... } ``` <1> Pre-chunked input provided as an array of strings. Each array element represents a single chunk that will be sent directly to the inference service without further processing. * Fix broken bwc logic in text field mapper introduced by #129126 (#129308) A missing condition in the bwc logic caused a text field to be a stored, while before #129126, this wasn't the case. * [ESQL] Fix SpatialDocValuesExtraction rule replacing TimeSeries agg node (#129273) `TimeSeriesAggregateExec` (TS) node inherits from `AggregateExec` (STATS). The `SpatialDocValuesExtraction` rule was replacing all `AggregateExec` instances with another `AggregateExec`, whether the same class or not. * Make `TransportMoveToStepAction` project-aware (#129252) Future work is necessary to make the YAML tests pass in MP mode. * [DOCS] Adds term vectors API examples (#129328) * [DOCS] Adds term vectors API examples. * Update docs/reference/elasticsearch/rest-apis/term-vectors-examples.md Co-authored-by: Liam Thompson <32779855+leemthompo@users.noreply.github.com> * [DOCS] Addresses feedback. * [DOCS] Fixes link. --------- Co-authored-by: Liam Thompson <32779855+leemthompo@users.noreply.github.com> * [ESQL] Fix TopNSetTestCase test and unmute it (#129327) Closes #129257 * ESQL: Change queries ID to be the same as the async (#127472) This PR changes the list and query API for ESQL, such that the ID now follows the same format as async query IDs. This is saved as part of the task status. For async queries, this is easy, but for sync queries, this is slightly more complicated, since when creating them, we don't have access to a node ID. So instead, the status itself is just the doc ID portion of the async execution ID, which is used for salting, since this part needs to be consistent, so that when we list the queries, we can compute the async execution ID correctly. Also, I've removed the individual ID, node, and data node tags, as mentioned in the ticket. In addition, I've changed the accept and content-type to be JSON for lists. Resolves #127187 * Adjust unpromotable shard refresh request validation to allow RefreshResult.NO_REFRESH (#129176) When a primary shard uses the read-only engine, it always returns a RefreshResult.NO_REFRESH for refreshes. Since #93600 we added an extra roundtrip to hook unpromotable shard refresh logic. This hook is always executed, even if there are no unpromotable shards, but the UnpromotableShardRefreshRequest would fail if the primary shard returns a RefreshResult.NO_REFRESH result. Fix to be backported to several versions as it's annoying. Closes #129036 * Add a Multi-Project Search Rest Test (#128657) This commit adds a Rest IT specifically for search in MultiProject. Everything was already working as expected, but we were a bit light on explicit testing for search, which as _the_ core capability of Elasticsearch is worth testing thoroughly and clearly. * Modified LinearRetriever to include minScore * cleaned up * Made the same changes we did in textSimilarity * Fixed a minor error * cleaned up * Minscore is working :) * chore: empty commit to trigger PR update * Update docs/changelog/129359.yaml * Update 10_linear_retriever.yml * [CI] Auto commit changes from spotless --------- Co-authored-by: Tanguy Leroux <tlrx.dev@gmail.com> Co-authored-by: Martijn van Groningen <martijn.v.groningen@gmail.com> Co-authored-by: Rene Groeschke <rene@elastic.co> Co-authored-by: Nik Everett <nik9000@gmail.com> Co-authored-by: Niels Bauman <33722607+nielsbauman@users.noreply.github.com> Co-authored-by: Benjamin Trent <ben.w.trent@gmail.com> Co-authored-by: Luigi Dell'Aquila <luigi.dellaquila@gmail.com> Co-authored-by: Bogdan Pintea <bogdan.pintea@elastic.co> Co-authored-by: elasticsearchmachine <58790826+elasticsearchmachine@users.noreply.github.com> Co-authored-by: Albert Zaharovits <email+github@zalbert.me> Co-authored-by: Jim Ferenczi <jim.ferenczi@elastic.co> Co-authored-by: Jan-Kazlouski-elastic <jan.kazlouski@elastic.co> Co-authored-by: Leonardo Hoet <55866308+leo-hoet@users.noreply.github.com> Co-authored-by: lhoet-google <lhoet@google.com> Co-authored-by: Jonathan Buttner <56361221+jonathan-buttner@users.noreply.github.com> Co-authored-by: Carlos Delgado <6339205+carlosdelest@users.noreply.github.com> Co-authored-by: Jan Kuipers <148754765+jan-elastic@users.noreply.github.com> Co-authored-by: Tim Grein <tim.grein@elastic.co> Co-authored-by: Ievgen Degtiarenko <ievgen.degtiarenko@elastic.co> Co-authored-by: elasticsearchmachine <infra-root+elasticsearchmachine@elastic.co> Co-authored-by: Ignacio Vera <ignacio.vera@elastic.co> Co-authored-by: Mike Pellegrini <mike.pellegrini@elastic.co> Co-authored-by: Moritz Mack <mmack@apache.org> Co-authored-by: Ying Mao <ying.mao@elastic.co> Co-authored-by: Ioana Tagirta <ioanatia@users.noreply.github.com> Co-authored-by: Aurélien FOUCRET <aurelien.foucret@gmail.com> Co-authored-by: John Wagster <john.wagster@elastic.co> Co-authored-by: Larisa Motova <larisa.motova@elastic.co> Co-authored-by: Sam Xiao <sam.xiao@elastic.co> Co-authored-by: Richard Dennehy <richard.dennehy@elastic.co> Co-authored-by: Johannes Freden Jansson <johannes.freden@elastic.co> Co-authored-by: Alexander Spies <alexander.spies@elastic.co> Co-authored-by: István Zoltán Szabó <istvan.szabo@elastic.co> Co-authored-by: Iván Cea Fontenla <ivancea96@outlook.com> Co-authored-by: Dimitris Rempapis <dimitris.rempapis@elastic.co> Co-authored-by: Liam Thompson <32779855+leemthompo@users.noreply.github.com> Co-authored-by: kruskall <99559985+kruskall@users.noreply.github.com> Co-authored-by: Carson Ip <carsonip@users.noreply.github.com> Co-authored-by: Lorenzo Dematté <lorenzo.dematte@elastic.co> Co-authored-by: Kathleen DeRusso <kathleen.derusso@elastic.co> Co-authored-by: Samiul Monir <150824886+Samiul-TheSoccerFan@users.noreply.github.com> Co-authored-by: Elastic Machine <elasticmachine@users.noreply.github.com> Co-authored-by: Nick Tindall <nick.tindall@elastic.co> Co-authored-by: ywangd <yang.wang@elastic.co> Co-authored-by: rjernst <ryan@elastic.co> Co-authored-by: Johannes Fredén <109296772+jfreden@users.noreply.github.com> Co-authored-by: Gal Lalouche <gal.lalouche@elastic.co> Co-authored-by: Tim Vernum <tim@adjective.org>

ivancea added 10 commits April 16, 2025 14:29

Add initial blockhash with TreeSet and basic operator test

3322f34

Fix dedupe for nulls and multivalues

fcdfbd4

Added tests for null order and sort order

7282e98

Merge branch 'main' into esql-top-n-agg-grouping

b6d7265

Move generic BlockHash code to base TestCase

4950f5f

Initial TopNBlockHash test structure, WIP

a425c5d

Added TopNBlockHash test without nulls, and fixed bug in ords

665ace3

Fix BlockHashTestCase keys assertion early return not checking all keys

6616fdf

Added nulls tests, fixed dedupe lookup nulls handling and nulls error…

d838057

… in hash

Added multivalue tests and fixed dedupe with multivalues and no matches

e78fac5

ivancea added >enhancement Team:Analytics Meta label for analytical engine team (ESQL/Aggs/Geo) :Analytics/ES|QL AKA ESQL labels Apr 22, 2025

elasticsearchmachine added the v9.1.0 label Apr 22, 2025

ivancea and others added 3 commits April 22, 2025 14:59

Update docs/changelog/127148.yaml

143d86f

[CI] Auto commit changes from spotless

fa3a779

Remove TopNBlockHash abstract class

4b212cd

ivancea commented Apr 22, 2025

View reviewed changes

nik9000 reviewed Apr 22, 2025

View reviewed changes

ivancea and others added 7 commits April 23, 2025 10:57

Remove unused class

87eb135

Merge branch 'main' into esql-top-n-agg-grouping

c9635f9

Fix SeenGroupIds and added extra tests for it

29fa4d1

Included TopNLongBlockHash into BlockHash.ubuild() and GroupSpec logi…

7ca3ce3

…cs, and added it to benchmark

[CI] Auto commit changes from spotless

13900f6

Keep the last value in the top while using TreeSet, for performance

c778b3b

Merge branch 'main' into esql-top-n-agg-grouping

270eb6b

ivancea and others added 6 commits May 27, 2025 14:37

Merge branch 'main' into esql-top-n-agg-grouping-tests

9c12c35

Merge branch 'main' into esql-top-n-agg-grouping

714aed9

Merge branch 'esql-top-n-agg-grouping' into esql-top-n-agg-grouping-t…

9d8f749

…ests

Make operator tests not depend directly on the BlockHash implementation

ed18e7c

Restore original BucketedSort

16c88eb

Improve migration and remove unused BucketedSort test

2b1f868

ivancea commented May 28, 2025

View reviewed changes

ivancea added 5 commits May 28, 2025 17:24

Remove fixed TO-DOs

f0349f4

Fix potential memory leak on circuit breaker error

e7ebd05

Merge branch 'main' into esql-top-n-agg-grouping

40b7682

Renamed topN set and added tests

76291a6

Remove unused code from TopNMultivalueDedupeLong

dd48d97

ivancea commented May 29, 2025

View reviewed changes

Fix missing success=true in constructor

e63b73a

ivancea marked this pull request as ready for review May 29, 2025 13:11

ivancea requested a review from a team as a code owner May 29, 2025 13:11

ivancea requested a review from nik9000 May 29, 2025 14:23

nik9000 approved these changes Jun 10, 2025

View reviewed changes

Merge branch 'main' into esql-top-n-agg-grouping

b2c42e4

ivancea enabled auto-merge (squash) June 11, 2025 10:52

ivancea merged commit d405d3a into elastic:main Jun 11, 2025
16 of 18 checks passed

ivancea deleted the esql-top-n-agg-grouping branch June 11, 2025 12:00

luigidellaquila mentioned this pull request Jun 11, 2025

[CI] LongTopNSetTests testCrankyBreaker failing #129257

Closed

ivancea mentioned this pull request Jun 18, 2025

[ESQL] Fix TopN grouping aggregates bug with non-qualifying intermediate states #129633

Draft


		import static org.hamcrest.Matchers.equalTo;

		public abstract class TopNSetTestCase<T extends Releasable, V extends Comparable<V>> extends ESTestCase {

ESQL: Skip unused STATS groups by adding a Top N BlockHash implementation #127148

ESQL: Skip unused STATS groups by adding a Top N BlockHash implementation #127148

Uh oh!

Conversation

ivancea commented Apr 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Current behaviour

New behaviour

In this PR

Benchmarks

Uh oh!

elasticsearchmachine commented Apr 22, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ivancea Apr 22, 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!

Choose a reason for hiding this comment

Uh oh!

elasticsearchmachine commented May 29, 2025

Uh oh!

nik9000 left a comment

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!

Uh oh!

Uh oh!

ivancea commented Apr 22, 2025 •

edited

Loading

ivancea Apr 22, 2025 •

edited

Loading