Add cluster command for PPL

common/src/main/java/org/opensearch/sql/common/cluster/TextSimilarityClustering.java

@@ -0,0 +1,170 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package org.opensearch.sql.common.cluster;
+
+import java.util.HashMap;
+import java.util.Map;
+import java.util.concurrent.ConcurrentHashMap;
+import org.apache.commons.text.similarity.CosineSimilarity;
+
+/**
+ * Greedy single-pass text similarity clustering for grouping similar text values. Events are
+ * processed in order; each is compared to existing cluster representatives using cosine similarity.
+ * If the best match meets the threshold, the event joins that cluster; otherwise a new cluster is
+ * created.
+ *
+ * <p>Optimized for incremental processing with vector caching and memory-efficient operations.
+ */
+public class TextSimilarityClustering {
+
+  private static final CosineSimilarity COSINE = new CosineSimilarity();
+
+  // Cache vectorized representations to avoid recomputation
+  private final Map<String, Map<CharSequence, Integer>> vectorCache = new ConcurrentHashMap<>();
+  private static final int MAX_CACHE_SIZE = 10000;
+
+  private final double threshold;
+  private final String matchMode;
+  private final String delims;
+
+  public TextSimilarityClustering(double threshold, String matchMode, String delims) {
+    this.threshold = validateThreshold(threshold);
+    this.matchMode = validateMatchMode(matchMode);
+    this.delims = delims != null ? delims : " ";
+  }
+
+  private static double validateThreshold(double threshold) {
+    if (threshold <= 0.0 || threshold >= 1.0) {
+      throw new IllegalArgumentException(
+          "The threshold must be > 0.0 and < 1.0, got: " + threshold);
+    }
+    return threshold;
+  }
+
+  private static String validateMatchMode(String matchMode) {
+    if (matchMode == null) {
+      return "termlist";
+    }
+    switch (matchMode.toLowerCase()) {
+      case "termlist":
+      case "termset":
+      case "ngramset":
+        return matchMode.toLowerCase();
+      default:
+        throw new IllegalArgumentException(
+            "Invalid match mode: " + matchMode + ". Must be one of: termlist, termset, ngramset");
+    }
+  }
+
+  /**
+   * Compute similarity between two text values using the configured match mode. Used for
+   * incremental clustering against cluster representatives.
+   */
+  public double computeSimilarity(String text1, String text2) {
+    // Normalize nulls to empty strings
+    String normalizedText1 = (text1 == null) ? "" : text1;
+    String normalizedText2 = (text2 == null) ? "" : text2;
+
+    // Both are empty - perfect match
+    if (normalizedText1.isEmpty() && normalizedText2.isEmpty()) {
+      return 1.0;
+    }
+
+    // One is empty, other isn't - no match
+    if (normalizedText1.isEmpty() || normalizedText2.isEmpty()) {
+      return 0.0;
+    }
+
+    // Both non-empty - compute cosine similarity
+    Map<CharSequence, Integer> vector1 = vectorizeWithCache(normalizedText1);
+    Map<CharSequence, Integer> vector2 = vectorizeWithCache(normalizedText2);
+
+    return COSINE.cosineSimilarity(vector1, vector2);
+  }
+
+  private Map<CharSequence, Integer> vectorizeWithCache(String value) {
+    return vectorCache.computeIfAbsent(value, k -> {
+      if (vectorCache.size() > MAX_CACHE_SIZE) {
+        vectorCache.keySet().parallelStream()
+            .limit(MAX_CACHE_SIZE / 2)
+            .forEach(vectorCache::remove);
+      }
+      return vectorize(k);
+    });
+  }
+
+  private Map<CharSequence, Integer> vectorize(String value) {
+    if (value == null || value.isEmpty()) {
+      return Map.of();
+    }
+    return switch (matchMode) {
+      case "termset" -> vectorizeTermSet(value);
+      case "ngramset" -> vectorizeNgramSet(value);
+      default -> vectorizeTermList(value);
+    };
+  }
+
+  private static final java.util.regex.Pattern NUMERIC_PATTERN =
+      java.util.regex.Pattern.compile("^\\d+$");
+
+  private static String normalizeToken(String token) {
+    return NUMERIC_PATTERN.matcher(token).matches() ? "*" : token;
+  }
+
+  /** Positional term frequency — token order matters. */
+  private Map<CharSequence, Integer> vectorizeTermList(String value) {
+    String[] tokens = tokenize(value);
+    Map<CharSequence, Integer> vector = new HashMap<>((int) (tokens.length * 1.4));
+
+    for (int i = 0; i < tokens.length; i++) {
+      if (!tokens[i].isEmpty()) {
+        String key = i + "-" + normalizeToken(tokens[i]);
+        vector.merge(key, 1, Integer::sum);
+      }
+    }
+    return vector;
+  }
+
+  /** Bag-of-words term frequency — token order ignored. */
+  private Map<CharSequence, Integer> vectorizeTermSet(String value) {
+    String[] tokens = tokenize(value);
+    Map<CharSequence, Integer> vector = new HashMap<>((int) (tokens.length * 1.4));
+
+    for (String token : tokens) {
+      if (!token.isEmpty()) {
+        vector.merge(normalizeToken(token), 1, Integer::sum);
+      }
+    }
+    return vector;
+  }
+
+  /** Character trigram frequency. */
+  private Map<CharSequence, Integer> vectorizeNgramSet(String value) {
+    if (value.length() < 3) {
+      // For very short strings, fall back to character frequency
+      Map<CharSequence, Integer> vector = new HashMap<>();
+      for (char c : value.toCharArray()) {
+        vector.merge(String.valueOf(c), 1, Integer::sum);
+      }
+      return vector;
+    }
+
+    Map<CharSequence, Integer> vector = new HashMap<>((int) ((value.length() - 2) * 1.4));
+    for (int i = 0; i <= value.length() - 3; i++) {
+      String ngram = value.substring(i, i + 3);
+      vector.merge(ngram, 1, Integer::sum);
+    }
+    return vector;
+  }
+
+  private String[] tokenize(String value) {
+    if ("non-alphanumeric".equals(delims)) {
+      return value.split("[^a-zA-Z0-9_]+");
+    }
+    String pattern = "[" + java.util.regex.Pattern.quote(delims) + "]+";
+    return value.split(pattern);
+  }
+}

core/src/main/java/org/opensearch/sql/ast/AbstractNodeVisitor.java

@@ -54,6 +54,7 @@
 import org.opensearch.sql.ast.tree.Bin;
 import org.opensearch.sql.ast.tree.Chart;
 import org.opensearch.sql.ast.tree.CloseCursor;
+import org.opensearch.sql.ast.tree.Cluster;
 import org.opensearch.sql.ast.tree.Convert;
 import org.opensearch.sql.ast.tree.Dedupe;
 import org.opensearch.sql.ast.tree.Eval;
@@ -432,6 +433,10 @@ public T visitPatterns(Patterns patterns, C context) {
     return visitChildren(patterns, context);
   }
 
+  public T visitCluster(Cluster node, C context) {
+    return visitChildren(node, context);
+  }
+
   public T visitWindow(Window window, C context) {
     return visitChildren(window, context);
   }

core/src/main/java/org/opensearch/sql/ast/tree/Cluster.java

@@ -0,0 +1,53 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package org.opensearch.sql.ast.tree;
+
+import com.google.common.collect.ImmutableList;
+import java.util.List;
+import lombok.AllArgsConstructor;
+import lombok.EqualsAndHashCode;
+import lombok.Getter;
+import lombok.RequiredArgsConstructor;
+import lombok.Setter;
+import lombok.ToString;
+import org.opensearch.sql.ast.AbstractNodeVisitor;
+import org.opensearch.sql.ast.expression.UnresolvedExpression;
+
+/** AST node for the PPL cluster command. */
+@Getter
+@Setter
+@ToString
+@EqualsAndHashCode(callSuper = false)
+@RequiredArgsConstructor
+@AllArgsConstructor
+public class Cluster extends UnresolvedPlan {
+
+  private final UnresolvedExpression sourceField;
+  private final double threshold;
+  private final String matchMode;
+  private final String labelField;
+  private final String countField;
+  private final boolean labelOnly;
+  private final boolean showCount;
+  private final String delims;
+  private UnresolvedPlan child;
+
+  @Override
+  public Cluster attach(UnresolvedPlan child) {
+    this.child = child;
+    return this;
+  }
+
+  @Override
+  public List<UnresolvedPlan> getChild() {
+    return this.child == null ? ImmutableList.of() : ImmutableList.of(this.child);
+  }
+
+  @Override
+  public <T, C> T accept(AbstractNodeVisitor<T, C> nodeVisitor, C context) {
+    return nodeVisitor.visitCluster(this, context);
+  }
+}

core/src/main/java/org/opensearch/sql/calcite/CalciteRelNodeVisitor.java

@@ -2478,6 +2478,93 @@ public RelNode visitKmeans(Kmeans node, CalcitePlanContext context) {
     throw new CalciteUnsupportedException("Kmeans command is unsupported in Calcite");
   }
 
+  @Override
+  public RelNode visitCluster(
+      org.opensearch.sql.ast.tree.Cluster node, CalcitePlanContext context) {
+    visitChildren(node, context);
+
+    // Resolve clustering as a window function over all rows (unbounded frame).
+    // The window function buffers all rows, runs the greedy clustering algorithm,
+    // and returns an array of cluster labels (one per input row, in order).
+    List<UnresolvedExpression> funcParams = new ArrayList<>();
+    funcParams.add(node.getSourceField());
+    funcParams.add(AstDSL.doubleLiteral(node.getThreshold()));
+    funcParams.add(AstDSL.stringLiteral(node.getMatchMode()));
+    funcParams.add(AstDSL.stringLiteral(node.getDelims()));
+
+    RexNode clusterWindow =
+        rexVisitor.analyze(
+            new WindowFunction(
+                new Function(
+                    BuiltinFunctionName.INTERNAL_CLUSTER_LABEL.getName().getFunctionName(),
+                    funcParams),
+                List.of(),
+                List.of()),
+            context);
+    String arrayAlias = "_cluster_labels_array";
+    context.relBuilder.projectPlus(context.relBuilder.alias(clusterWindow, arrayAlias));
+
+    // Add ROW_NUMBER to index into the array (1-based).
+    String rowNumAlias = "_cluster_row_idx";
+    RexNode rowNum =
+        context
+            .relBuilder
+            .aggregateCall(SqlStdOperatorTable.ROW_NUMBER)
+            .over()
+            .rowsBetween(RexWindowBounds.UNBOUNDED_PRECEDING, RexWindowBounds.CURRENT_ROW)
+            .as(rowNumAlias);
+    context.relBuilder.projectPlus(rowNum);
+
+    // Extract the label for this row: array[row_number] (ITEM access is 1-based).
+    RexNode rowIdxAsInt =
+        context.rexBuilder.makeCast(
+            context.rexBuilder.getTypeFactory().createSqlType(SqlTypeName.INTEGER),
+            context.relBuilder.field(rowNumAlias));
+    RexNode labelExpr =
+        context.rexBuilder.makeCall(
+            SqlStdOperatorTable.ITEM, context.relBuilder.field(arrayAlias), rowIdxAsInt);
+    context.relBuilder.projectPlus(context.relBuilder.alias(labelExpr, node.getLabelField()));
+
+    // Remove the temporary array and row index columns.
+    context.relBuilder.projectExcept(
+        context.relBuilder.field(arrayAlias), context.relBuilder.field(rowNumAlias));
+
+    if (node.isShowCount()) {
+      // cluster_count = COUNT(*) OVER (PARTITION BY cluster_label)
+      RexNode countWindow =
+          context
+              .relBuilder
+              .aggregateCall(SqlStdOperatorTable.COUNT)
+              .over()
+              .partitionBy(context.relBuilder.field(node.getLabelField()))
+              .rowsBetween(RexWindowBounds.UNBOUNDED_PRECEDING, RexWindowBounds.UNBOUNDED_FOLLOWING)
+              .as(node.getCountField());
+      context.relBuilder.projectPlus(countWindow);
+    }
+
+    if (!node.isLabelOnly()) {
+      // Filter to representative rows only: keep the first event per cluster.
+      String convergenceRowNum = "_cluster_convergence_row_num";
+      RexNode convergenceRn =
+          context
+              .relBuilder
+              .aggregateCall(SqlStdOperatorTable.ROW_NUMBER)
+              .over()
+              .partitionBy(context.relBuilder.field(node.getLabelField()))
+              .rowsTo(RexWindowBounds.CURRENT_ROW)
+              .as(convergenceRowNum);
+      context.relBuilder.projectPlus(convergenceRn);
+      context.relBuilder.filter(
+          context.rexBuilder.makeCall(
+              SqlStdOperatorTable.EQUALS,
+              context.relBuilder.field(convergenceRowNum),
+              context.rexBuilder.makeExactLiteral(java.math.BigDecimal.ONE)));
+      context.relBuilder.projectExcept(context.relBuilder.field(convergenceRowNum));
+    }
+
+    return context.relBuilder.peek();
+  }
+
   @Override
   public RelNode visitRareTopN(RareTopN node, CalcitePlanContext context) {
     visitChildren(node, context);