ESQL: LOOKUP JOIN with expressions (elastic#134098)

Add support for LOOKUP JOIN with expressions.

FROM index1
| LOOKUP JOIN lookup_index on left_field1 > right_field1 AND left_field2 <= right_field2 AND left_field3 == right_field1
--handle name conflict with rename
FROM index1
| RENAME id as id_left
| LOOKUP JOIN lookup_index on id_left >=id AND left_field2 <= right_field2
Add support for 1 or more join predicate expressions in the ON clause of the LOOKUP JOIN. If there is more than one predicate expression, the predicate expression must be connected with the AND operator. Each predicate is in the form field1 BINARY_OPERATOR field2. One of them should be from the left and the other from the right. BINARY_OPERATOR is one of ==, >,<, >=, <=, !=. The same field can be used multiple times in the join on condition. If the fields in the join on conditions have the same name, they should be renamed so that the fields in the join on condition have unique names.

This commit may include content that was generated or assisted by Cursor, Gemini CLI and/or Github Copilot.

Author: julian-elastic

docs/changelog/134098.yaml

@@ -0,0 +1,5 @@
+pr: 134098
+summary: LOOKUP JOIN with expressions
+area: ES|QL
+type: enhancement
+issues: []

docs/reference/query-languages/esql/_snippets/commands/types/lookup-join.md

@@ -5,17 +5,17 @@
 | field from the left index | field from the lookup index | 
 | --- | --- | 
 | boolean | boolean |
-| byte | half_float, float, double, scaled_float, byte, short, integer, long |
+| byte | byte, short, integer, long, half_float, float, double, scaled_float |
 | date | date |
 | date_nanos | date_nanos |
 | double | half_float, float, double, scaled_float, byte, short, integer, long |
 | float | half_float, float, double, scaled_float, byte, short, integer, long |
 | half_float | half_float, float, double, scaled_float, byte, short, integer, long |
-| integer | half_float, float, double, scaled_float, byte, short, integer, long |
+| integer | byte, short, integer, long, half_float, float, double, scaled_float |
 | ip | ip |
 | keyword | keyword |
-| long | half_float, float, double, scaled_float, byte, short, integer, long |
+| long | byte, short, integer, long, half_float, float, double, scaled_float |
 | scaled_float | half_float, float, double, scaled_float, byte, short, integer, long |
-| short | half_float, float, double, scaled_float, byte, short, integer, long |
+| short | byte, short, integer, long, half_float, float, double, scaled_float |
 | text | keyword |
 

server/src/main/java/org/elasticsearch/TransportVersions.java

@@ -327,6 +327,7 @@ static TransportVersion def(int id) {
     public static final TransportVersion TIMESERIES_DEFAULT_LIMIT = def(9_160_0_00);
     public static final TransportVersion INFERENCE_API_OPENAI_HEADERS = def(9_161_0_00);
     public static final TransportVersion NEW_SEMANTIC_QUERY_INTERCEPTORS = def(9_162_0_00);
+    public static final TransportVersion ESQL_LOOKUP_JOIN_ON_EXPRESSION = def(9_163_0_00);
 
     /*
      * STOP! READ THIS FIRST! No, really,

test/external-modules/esql-heap-attack/src/javaRestTest/java/org/elasticsearch/xpack/esql/heap_attack/HeapAttackIT.java

@@ -566,6 +566,7 @@ private Response query(String query, String filterPath) throws IOException {
                 .setRequestConfig(RequestConfig.custom().setSocketTimeout(Math.toIntExact(TimeValue.timeValueMinutes(6).millis())).build())
                 .setWarningsHandler(WarningsHandler.PERMISSIVE)
         );
+        logger.info("Running query:" + query);
         return runQuery(() -> client().performRequest(request));
     }
 
@@ -735,28 +736,48 @@ private Map<String, Object> fetchMvLongs() throws IOException {
     public void testLookupExplosion() throws IOException {
         int sensorDataCount = 400;
         int lookupEntries = 10000;
-        Map<?, ?> map = lookupExplosion(sensorDataCount, lookupEntries, 1, lookupEntries);
+        Map<?, ?> map = lookupExplosion(sensorDataCount, lookupEntries, 1, lookupEntries, false);
         assertMap(map, matchesMap().extraOk().entry("values", List.of(List.of(sensorDataCount * lookupEntries))));
     }
 
     public void testLookupExplosionManyFields() throws IOException {
         int sensorDataCount = 400;
         int lookupEntries = 1000;
         int joinFieldsCount = 990;
-        Map<?, ?> map = lookupExplosion(sensorDataCount, lookupEntries, joinFieldsCount, lookupEntries);
+        Map<?, ?> map = lookupExplosion(sensorDataCount, lookupEntries, joinFieldsCount, lookupEntries, false);
+        assertMap(map, matchesMap().extraOk().entry("values", List.of(List.of(sensorDataCount * lookupEntries))));
+    }
+
+    public void testLookupExplosionExpression() throws IOException {
+        int sensorDataCount = 400;
+        int lookupEntries = 10000;
+        Map<?, ?> map = lookupExplosion(sensorDataCount, lookupEntries, 1, lookupEntries, true);
+        assertMap(map, matchesMap().extraOk().entry("values", List.of(List.of(sensorDataCount * lookupEntries))));
+    }
+
+    public void testLookupExplosionManyFieldsExpression() throws IOException {
+        int sensorDataCount = 400;
+        int lookupEntries = 1000;
+        int joinFieldsCount = 399;// only join on 399 columns due to max expression size of 400
+        Map<?, ?> map = lookupExplosion(sensorDataCount, lookupEntries, joinFieldsCount, lookupEntries, true);
         assertMap(map, matchesMap().extraOk().entry("values", List.of(List.of(sensorDataCount * lookupEntries))));
     }
 
     public void testLookupExplosionManyMatchesManyFields() throws IOException {
         // 1500, 10000 is enough locally, but some CI machines need more.
         int lookupEntries = 10000;
-        assertCircuitBreaks(attempt -> lookupExplosion(attempt * 1500, lookupEntries, 30, lookupEntries));
+        assertCircuitBreaks(attempt -> lookupExplosion(attempt * 1500, lookupEntries, 30, lookupEntries, false));
     }
 
     public void testLookupExplosionManyMatches() throws IOException {
         // 1500, 10000 is enough locally, but some CI machines need more.
         int lookupEntries = 10000;
-        assertCircuitBreaks(attempt -> lookupExplosion(attempt * 1500, lookupEntries, 1, lookupEntries));
+        assertCircuitBreaks(attempt -> lookupExplosion(attempt * 1500, lookupEntries, 1, lookupEntries, false));
+    }
+
+    public void testLookupExplosionManyMatchesExpression() throws IOException {
+        int lookupEntries = 10000;
+        assertCircuitBreaks(attempt -> lookupExplosion(attempt * 1500, lookupEntries, 1, lookupEntries, true));
     }
 
     public void testLookupExplosionManyMatchesFiltered() throws IOException {
@@ -768,9 +789,21 @@ public void testLookupExplosionManyMatchesFiltered() throws IOException {
         int reductionFactor = 1000; // reduce the number of matches by this factor
         // lookupEntries % reductionFactor must be 0 to ensure the number of rows returned matches the expected value
         assertTrue(0 == lookupEntries % reductionFactor);
-        Map<?, ?> map = lookupExplosion(sensorDataCount, lookupEntries, 1, lookupEntries / reductionFactor);
+        Map<?, ?> map = lookupExplosion(sensorDataCount, lookupEntries, 1, lookupEntries / reductionFactor, false);
         assertMap(map, matchesMap().extraOk().entry("values", List.of(List.of(sensorDataCount * lookupEntries / reductionFactor))));
+    }
 
+    public void testLookupExplosionManyMatchesFilteredExpression() throws IOException {
+        // This test will only work with the expanding join optimization
+        // that pushes the filter to the right side of the lookup.
+        // Without the optimization, it will fail with circuit_breaking_exception
+        int sensorDataCount = 10000;
+        int lookupEntries = 10000;
+        int reductionFactor = 1000; // reduce the number of matches by this factor
+        // lookupEntries % reductionFactor must be 0 to ensure the number of rows returned matches the expected value
+        assertTrue(0 == lookupEntries % reductionFactor);
+        Map<?, ?> map = lookupExplosion(sensorDataCount, lookupEntries, 1, lookupEntries / reductionFactor, true);
+        assertMap(map, matchesMap().extraOk().entry("values", List.of(List.of(sensorDataCount * lookupEntries / reductionFactor))));
     }
 
     public void testLookupExplosionNoFetch() throws IOException {
@@ -797,17 +830,33 @@ public void testLookupExplosionBigStringManyMatches() throws IOException {
         assertCircuitBreaks(attempt -> lookupExplosionBigString(attempt * 500, 1));
     }
 
-    private Map<String, Object> lookupExplosion(int sensorDataCount, int lookupEntries, int joinFieldsCount, int lookupEntriesToKeep)
-        throws IOException {
+    private Map<String, Object> lookupExplosion(
+        int sensorDataCount,
+        int lookupEntries,
+        int joinFieldsCount,
+        int lookupEntriesToKeep,
+        boolean expressionBasedJoin
+    ) throws IOException {
         try {
-            lookupExplosionData(sensorDataCount, lookupEntries, joinFieldsCount);
+            lookupExplosionData(sensorDataCount, lookupEntries, joinFieldsCount, expressionBasedJoin);
             StringBuilder query = startQuery();
             query.append("FROM sensor_data | LOOKUP JOIN sensor_lookup ON ");
-            for (int i = 0; i < joinFieldsCount; i++) {
-                if (i != 0) {
-                    query.append(",");
+            if (expressionBasedJoin) {
+                for (int i = 0; i < joinFieldsCount; i++) {
+                    if (i != 0) {
+                        query.append(" AND ");
+                    }
+                    query.append("id_left").append(i);
+                    query.append("==");
+                    query.append("id_right").append(i);
+                }
+            } else {
+                for (int i = 0; i < joinFieldsCount; i++) {
+                    if (i != 0) {
+                        query.append(",");
+                    }
+                    query.append("id").append(i);
                 }
-                query.append("id").append(i);
             }
             if (lookupEntries != lookupEntriesToKeep) {
                 // add a filter to reduce the number of matches
@@ -826,7 +875,7 @@ private Map<String, Object> lookupExplosion(int sensorDataCount, int lookupEntri
 
     private Map<String, Object> lookupExplosionNoFetch(int sensorDataCount, int lookupEntries) throws IOException {
         try {
-            lookupExplosionData(sensorDataCount, lookupEntries, 1);
+            lookupExplosionData(sensorDataCount, lookupEntries, 1, false);
             StringBuilder query = startQuery();
             query.append("FROM sensor_data | LOOKUP JOIN sensor_lookup ON id0 | STATS COUNT(*)\"}");
             return responseAsMap(query(query.toString(), null));
@@ -836,14 +885,15 @@ private Map<String, Object> lookupExplosionNoFetch(int sensorDataCount, int look
         }
     }
 
-    private void lookupExplosionData(int sensorDataCount, int lookupEntries, int joinFieldCount) throws IOException {
-        initSensorData(sensorDataCount, 1, joinFieldCount);
-        initSensorLookup(lookupEntries, 1, i -> "73.9857 40.7484", joinFieldCount);
+    private void lookupExplosionData(int sensorDataCount, int lookupEntries, int joinFieldCount, boolean expressionBasedJoin)
+        throws IOException {
+        initSensorData(sensorDataCount, 1, joinFieldCount, expressionBasedJoin);
+        initSensorLookup(lookupEntries, 1, i -> "73.9857 40.7484", joinFieldCount, expressionBasedJoin);
     }
 
     private Map<String, Object> lookupExplosionBigString(int sensorDataCount, int lookupEntries) throws IOException {
         try {
-            initSensorData(sensorDataCount, 1, 1);
+            initSensorData(sensorDataCount, 1, 1, false);
             initSensorLookupString(lookupEntries, 1, i -> {
                 int target = Math.toIntExact(ByteSizeValue.ofMb(1).getBytes());
                 StringBuilder str = new StringBuilder(Math.toIntExact(ByteSizeValue.ofMb(2).getBytes()));
@@ -876,7 +926,7 @@ public void testEnrichExplosionManyMatches() throws IOException {
 
     private Map<String, Object> enrichExplosion(int sensorDataCount, int lookupEntries) throws IOException {
         try {
-            initSensorData(sensorDataCount, 1, 1);
+            initSensorData(sensorDataCount, 1, 1, false);
             initSensorEnrich(lookupEntries, 1, i -> "73.9857 40.7484");
             try {
                 StringBuilder query = startQuery();
@@ -1050,7 +1100,7 @@ private void initMvLongsIndex(int docs, int fields, int fieldValues) throws IOEx
         initIndex("mv_longs", bulk.toString());
     }
 
-    private void initSensorData(int docCount, int sensorCount, int joinFieldCount) throws IOException {
+    private void initSensorData(int docCount, int sensorCount, int joinFieldCount, boolean expressionBasedJoin) throws IOException {
         logger.info("loading sensor data");
         // We cannot go over 1000 fields, due to failed on parsing mappings on index creation
         // [sensor_data] java.lang.IllegalArgumentException: Limit of total fields [1000] has been exceeded
@@ -1061,8 +1111,9 @@ private void initSensorData(int docCount, int sensorCount, int joinFieldCount) t
                  "properties": {
                      "@timestamp": { "type": "date" },
             """);
+        String suffix = expressionBasedJoin ? "_left" : "";
         for (int i = 0; i < joinFieldCount; i++) {
-            createIndexBuilder.append("\"id").append(i).append("\": { \"type\": \"long\" },");
+            createIndexBuilder.append("\"id").append(suffix).append(i).append("\": { \"type\": \"long\" },");
         }
         createIndexBuilder.append("""
                     "value": { "type": "double" }
@@ -1083,7 +1134,7 @@ private void initSensorData(int docCount, int sensorCount, int joinFieldCount) t
                 {"create":{}}
                 {"timestamp":"%s",""", DateFieldMapper.DEFAULT_DATE_TIME_FORMATTER.formatMillis(i * 10L + firstDate)));
             for (int j = 0; j < joinFieldCount; j++) {
-                data.append(String.format(Locale.ROOT, "\"id%d\":%d, ", j, i % sensorCount));
+                data.append(String.format(Locale.ROOT, "\"id%s%d\":%d, ", suffix, j, i % sensorCount));
             }
             data.append(String.format(Locale.ROOT, "\"value\": %f}\n", i * 1.1));
             if (i % docsPerBulk == docsPerBulk - 1) {
@@ -1094,8 +1145,13 @@ private void initSensorData(int docCount, int sensorCount, int joinFieldCount) t
         initIndex("sensor_data", data.toString());
     }
 
-    private void initSensorLookup(int lookupEntries, int sensorCount, IntFunction<String> location, int joinFieldsCount)
-        throws IOException {
+    private void initSensorLookup(
+        int lookupEntries,
+        int sensorCount,
+        IntFunction<String> location,
+        int joinFieldsCount,
+        boolean expressionBasedJoin
+    ) throws IOException {
         logger.info("loading sensor lookup");
         // cannot go over 1000 fields, due to failed on parsing mappings on index creation
         // [sensor_data] java.lang.IllegalArgumentException: Limit of total fields [1000] has been exceeded
@@ -1105,8 +1161,9 @@ private void initSensorLookup(int lookupEntries, int sensorCount, IntFunction<St
             {
                 "properties": {
             """);
+        String suffix = expressionBasedJoin ? "_right" : "";
         for (int i = 0; i < joinFieldsCount; i++) {
-            createIndexBuilder.append("\"id").append(i).append("\": { \"type\": \"long\" },");
+            createIndexBuilder.append("\"id").append(suffix).append(i).append("\": { \"type\": \"long\" },");
         }
         createIndexBuilder.append("""
                     "location": { "type": "geo_point" },
@@ -1127,7 +1184,7 @@ private void initSensorLookup(int lookupEntries, int sensorCount, IntFunction<St
                 {"create":{}}
                 {"""));
             for (int j = 0; j < joinFieldsCount; j++) {
-                data.append(String.format(Locale.ROOT, "\"id%d\":%d, ", j, sensor));
+                data.append(String.format(Locale.ROOT, "\"id%s%d\":%d, ", suffix, j, sensor));
             }
             data.append(String.format(Locale.ROOT, """
                 "location": "POINT(%s)", "filter_key": %d}\n""", location.apply(sensor), i));
@@ -1165,7 +1222,7 @@ private void initSensorLookupString(int lookupEntries, int sensorCount, IntFunct
     }
 
     private void initSensorEnrich(int lookupEntries, int sensorCount, IntFunction<String> location) throws IOException {
-        initSensorLookup(lookupEntries, sensorCount, location, 1);
+        initSensorLookup(lookupEntries, sensorCount, location, 1, false);
         logger.info("loading sensor enrich");
 
         Request create = new Request("PUT", "/_enrich/policy/sensor");

x-pack/plugin/esql/compute/src/main/java/org/elasticsearch/compute/operator/lookup/QueryList.java

@@ -125,7 +125,7 @@ public final Query getQuery(int position) {
      * Returns the query at the given position.
      */
     @Nullable
-    abstract Query doGetQuery(int position, int firstValueIndex, int valueCount);
+    public abstract Query doGetQuery(int position, int firstValueIndex, int valueCount);
 
     private Query wrapSingleValueQuery(Query query) {
         assert onlySingleValueParams != null : "Requested to wrap single value query without single value params";
@@ -155,17 +155,11 @@ private Query wrapSingleValueQuery(Query query) {
     }
 
     /**
-     * Returns a list of term queries for the given field and the input block
-     * using only the {@link ElementType} of the {@link Block} to determine the
-     * query.
+     * Returns a function that reads values from the given block. The function
+     * takes the offset of the value to read and returns the value as an {@link Object}.
      */
-    public static QueryList rawTermQueryList(
-        MappedFieldType field,
-        SearchExecutionContext searchExecutionContext,
-        AliasFilter aliasFilter,
-        Block block
-    ) {
-        IntFunction<Object> blockToJavaObject = switch (block.elementType()) {
+    public static IntFunction<Object> createBlockValueReader(Block block) {
+        return switch (block.elementType()) {
             case BOOLEAN -> {
                 BooleanBlock booleanBlock = (BooleanBlock) block;
                 yield booleanBlock::getBoolean;
@@ -196,7 +190,20 @@ public static QueryList rawTermQueryList(
             case AGGREGATE_METRIC_DOUBLE -> throw new IllegalArgumentException("can't read values from [aggregate metric double] block");
             case UNKNOWN -> throw new IllegalArgumentException("can't read values from [" + block + "]");
         };
-        return new TermQueryList(field, searchExecutionContext, aliasFilter, block, null, blockToJavaObject);
+    }
+
+    /**
+     * Returns a list of term queries for the given field and the input block
+     * using only the {@link ElementType} of the {@link Block} to determine the
+     * query.
+     */
+    public static QueryList rawTermQueryList(
+        MappedFieldType field,
+        SearchExecutionContext searchExecutionContext,
+        AliasFilter aliasFilter,
+        Block block
+    ) {
+        return new TermQueryList(field, searchExecutionContext, aliasFilter, block, null, createBlockValueReader(block));
     }
 
     /**
@@ -297,7 +304,7 @@ public TermQueryList onlySingleValues(Warnings warnings, String multiValueWarnin
         }
 
         @Override
-        Query doGetQuery(int position, int firstValueIndex, int valueCount) {
+        public Query doGetQuery(int position, int firstValueIndex, int valueCount) {
             return switch (valueCount) {
                 case 0 -> null;
                 case 1 -> field.termQuery(blockValueReader.apply(firstValueIndex), searchExecutionContext);
@@ -360,7 +367,7 @@ public DateNanosQueryList onlySingleValues(Warnings warnings, String multiValueW
         }
 
         @Override
-        Query doGetQuery(int position, int firstValueIndex, int valueCount) {
+        public Query doGetQuery(int position, int firstValueIndex, int valueCount) {
             return switch (valueCount) {
                 case 0 -> null;
                 case 1 -> dateFieldType.equalityQuery(blockValueReader.apply(firstValueIndex), searchExecutionContext);
@@ -412,7 +419,7 @@ public GeoShapeQueryList onlySingleValues(Warnings warnings, String multiValueWa
         }
 
         @Override
-        Query doGetQuery(int position, int firstValueIndex, int valueCount) {
+        public Query doGetQuery(int position, int firstValueIndex, int valueCount) {
             return switch (valueCount) {
                 case 0 -> null;
                 case 1 -> shapeQuery.apply(firstValueIndex);
@@ -453,5 +460,5 @@ private IntFunction<Query> shapeQuery() {
         }
     }
 
-    protected record OnlySingleValueParams(Warnings warnings, String multiValueWarningMessage) {}
+    public record OnlySingleValueParams(Warnings warnings, String multiValueWarningMessage) {}
 }

x-pack/plugin/esql/compute/test/src/main/java/org/elasticsearch/compute/test/OperatorTestCase.java

@@ -97,7 +97,7 @@ protected ByteSizeValue enoughMemoryForSimple() {
      * asserting both that this throws a {@link CircuitBreakingException} and releases
      * all pages.
      */
-    public final void testSimpleCircuitBreaking() {
+    public void testSimpleCircuitBreaking() {
         /*
          * Build the input before building `simple` to handle the rare
          * cases where `simple` need some state from the input - mostly