Lucene: define and validate CAGRA parameter boundaries

java/cuvs-lucene/README.md

@@ -79,6 +79,55 @@ than an individual Lucene codec. Async allocation is optional for correctness an
 GPU workloads with repeated device allocations, especially concurrent or multi-stream searches.
 Applications that do not opt in use the default RMM device-memory resource.
 
+### Parameter bounds
+
+The public Lucene API rejects out-of-range CAGRA parameters in Java before they reach native CAGRA.
+Most of these checks run at construction time; the `SINGLE_CTA` `iTopK` limit is additionally
+re-checked at search time against the effective value actually sent to native CAGRA, since a filter
+can raise it after construction (see below). The table below lists what is enforced in Java; it is
+**not** a claim that every value in these ranges is supported by native CAGRA — see the caveats
+that follow.
+
+| Parameter | Java-level range enforced |
+| --- | --- |
+| GPU writer threads | 1–512 |
+| Intermediate graph degree | 2–512 |
+| Graph degree | 1–512 |
+| `GPUKnnFloatVectorQuery` `iTopK` | minimum 1; at most 512 with `SINGLE_CTA` |
+| `GPUKnnFloatVectorQuery` `searchWidth` | minimum 1; 4,194,303 numeric-safety ceiling |
+
+`graphDegree` must not exceed `intermediateGraphDegree` under the `CUSTOM` strategy. Under
+`HEURISTIC`, the configured `graphDegree`/`intermediateGraphDegree` pair is not what CAGRA
+actually builds with, so this relationship is not enforced on the configured pair: for
+`AcceleratedHNSWParams`, both degrees are derived from `maxConn`/`beamWidth` and the configured
+pair is ignored entirely; for `GPUSearchParams`, the configured `graphDegree` is passed into the
+dataset-size heuristic as an input (it is not ignored), while `intermediateGraphDegree` is ignored
+and the rest of the build parameters are derived from the heuristic's output.
+
+**Only the lower bound of 1 and the `SINGLE_CTA` `iTopK` maximum of 512 are genuine native
+limits.** `MAX_ITOPK` (`Integer.MAX_VALUE`) is simply the largest value representable by the
+public Java API, and `MAX_SEARCH_WIDTH` (4,194,303) only keeps CAGRA's result buffer within its
+unsigned 32-bit indexing limit. Neither is a promise that native CAGRA supports every value up
+to that ceiling, and in practice values anywhere near `MAX_ITOPK` are not usable. The true upper
+limit for a given search depends on the resolved CAGRA algorithm, `max_iterations`, graph degree,
+filtering, and available GPU memory. In particular, `MULTI_CTA` (which a normal one-query `AUTO`
+search resolves to) sizes an internal traversal hash table from `search_width`, `iTopK`,
+`max_iterations`, and the graph degree. This API does not replicate that calculation, since
+`max_iterations` is itself auto-derived from the graph degree and dataset size, values not known
+at query-construction time, so out-of-range combinations are caught by native CAGRA at search
+time rather than here.
+
+Note that native CAGRA only reports some of those combinations cleanly. Moderately oversized
+values raise a clear exception, but above roughly `iTopK` 1e9 the native hash-table sizing loop
+fails to terminate and the search hangs instead of returning an error
+([#2523](https://github.com/NVIDIA/cuvs/issues/2523)). Treat these constants as representational
+ceilings only, and size `iTopK`/`searchWidth` to what the workload actually needs.
+
+The query uses an effective `iTopK` equal to the greater of the configured value and the requested
+Lucene `k`; for `SINGLE_CTA`, this effective value is re-validated against the 512 limit again once
+the filtered per-segment search path finishes adjusting it, since a restrictive filter can raise it
+past what was checked at query construction time.
+
 In a Maven project that includes the `cuvs-lucene` dependency shown above, create `src/main/java/com/nvidia/cuvs/lucene/examples/HelloCuvsLucene.java`:
 
 ```java

java/cuvs-lucene/src/main/java/com/nvidia/cuvs/lucene/AcceleratedHNSWParams.java

@@ -31,11 +31,9 @@ public static enum Strategy {
     CUSTOM
   }
 
-  /*
-   * TODO: Update boundaries for all parameters when a consensus is reached.
-   * Issue: https://github.com/rapidsai/cuvs-lucene/issues/99
-   */
+  /** Bounds for the public CAGRA and HNSW build parameters. */
   public static final int MIN_WRITER_THREADS = 1;
+
   public static final int MAX_WRITER_THREADS = 512;
   public static final int MIN_INT_GRAPH_DEG = 2;
   public static final int MAX_INT_GRAPH_DEG = 512;
@@ -342,6 +340,7 @@ public static class Builder {
      * @return instance of {@link Builder}
      */
     public Builder withWriterThreads(int writerThreads) {
+      validateRange("writerThreads", writerThreads, MIN_WRITER_THREADS, MAX_WRITER_THREADS);
       this.writerThreads = writerThreads;
       return this;
     }
@@ -356,6 +355,8 @@ public Builder withWriterThreads(int writerThreads) {
      * @return instance of {@link Builder}
      */
     public Builder withIntermediateGraphDegree(int intermediateGraphDegree) {
+      validateRange(
+          "intermediateGraphDegree", intermediateGraphDegree, MIN_INT_GRAPH_DEG, MAX_INT_GRAPH_DEG);
       this.intermediateGraphDegree = intermediateGraphDegree;
       return this;
     }
@@ -370,6 +371,7 @@ public Builder withIntermediateGraphDegree(int intermediateGraphDegree) {
      * @return instance of {@link Builder}
      */
     public Builder withGraphDegree(int graphDegree) {
+      validateRange("graphDegree", graphDegree, MIN_GRAPH_DEG, MAX_GRAPH_DEG);
       this.graphdegree = graphDegree;
       return this;
     }
@@ -520,6 +522,13 @@ public Builder withHnswHeuristicType(HnswHeuristicType hnswHeuristicType) {
       return this;
     }
 
+    private static void validateRange(String name, int value, int min, int max) {
+      if (value < min || value > max) {
+        throw new IllegalArgumentException(
+            name + " not in valid range. Valid range: [" + min + ", " + max + "]");
+      }
+    }
+
     /**
      * Validates the input parameters.
      *
@@ -551,6 +560,10 @@ private void validate() throws IllegalArgumentException {
                 + MAX_GRAPH_DEG
                 + "]");
       }
+      if (strategy == Strategy.CUSTOM && graphdegree > intermediateGraphDegree) {
+        throw new IllegalArgumentException(
+            "graphDegree must not be greater than intermediateGraphDegree.");
+      }
       if (hnswLayers < MIN_HNSW_LAYERS || hnswLayers > MAX_HNSW_LAYERS) {
         throw new IllegalArgumentException(
             "hnswLayers not in valid range. Valid range: ["

java/cuvs-lucene/src/main/java/com/nvidia/cuvs/lucene/CuVS2510GPUVectorsReader.java

@@ -501,9 +501,14 @@ public void search(String field, float[] target, KnnCollector knnCollector, Bits
         CagraSearchParams searchParams;
         if (knnCollector instanceof GPUPerLeafCuVSKnnCollector) {
           GPUPerLeafCuVSKnnCollector collector = (GPUPerLeafCuVSKnnCollector) knnCollector;
+          int effectiveITopK = Math.max(collector.getiTopK(), topK);
+          // topK may have been raised above the value validated at query construction time (see
+          // GPUKnnFloatVectorQuery.validateSearchParameters), e.g. by the filter-cardinality bump
+          // above. Re-validate against the final value actually sent to native CAGRA.
+          GPUKnnFloatVectorQuery.validateSingleCtaItopk(effectiveITopK, collector.getSearchAlgo());
           searchParams =
               new CagraSearchParams.Builder()
-                  .withItopkSize(Math.max(collector.getiTopK(), topK))
+                  .withItopkSize(effectiveITopK)
                   .withSearchWidth(collector.getSearchWidth())
                   .withThreadBlockSize(collector.getThreadBlockSize())
                   .withMaxIterations(collector.getMaxIterations())

java/cuvs-lucene/src/main/java/com/nvidia/cuvs/lucene/GPUKnnFloatVectorQuery.java

@@ -71,6 +71,46 @@
  */
 public class GPUKnnFloatVectorQuery extends KnnFloatVectorQuery {
 
+  /** Smallest supported CAGRA intermediate-result count. */
+  public static final int MIN_ITOPK = 1;
+
+  /**
+   * Largest intermediate-result count representable by the public Java API.
+   *
+   * <p>This is a representational limit only. It is emphatically not a supported maximum: values
+   * anywhere near it are rejected by native CAGRA in practice. Native CAGRA sizes internal
+   * traversal hash tables from a combination of itopk_size, search_width, max_iterations, and
+   * (for MULTI_CTA, which a normal one-query {@code AUTO} search resolves to) the graph degree
+   * and dataset size, none of which are all known at query-construction time, so this class does
+   * not attempt to replicate that sizing logic.
+   *
+   * <p>Moderately oversized combinations are rejected by native CAGRA with a clear exception (see
+   * {@link Utils#handleThrowable}). Very large values are not: above roughly 1e9, native CAGRA's
+   * hash-table sizing loop fails to terminate and the search hangs instead of returning an error.
+   * See <a href="https://github.com/NVIDIA/cuvs/issues/2523">#2523</a>. Callers should treat
+   * itopk_size as bounded by what their algorithm and dataset actually support, not by this
+   * constant.
+   */
+  public static final int MAX_ITOPK = Integer.MAX_VALUE;
+
+  /** Largest intermediate-result count supported by CAGRA's SINGLE_CTA search algorithm. */
+  public static final int MAX_SINGLE_CTA_ITOPK = 512;
+
+  /** Smallest supported number of CAGRA search entry points. */
+  public static final int MIN_SEARCH_WIDTH = 1;
+
+  /**
+   * Largest search width that keeps CAGRA's result buffer within its unsigned 32-bit indexing
+   * limit at the maximum graph degree and aligned {@link #MAX_ITOPK}.
+   *
+   * <p>This bound alone does not guarantee a given (iTopK, searchWidth) pair is supported: as
+   * with {@link #MAX_ITOPK}, native CAGRA may still reject a combination that exceeds its
+   * traversal hash table's capacity (e.g. the MULTI_CTA path used by a normal one-query {@code
+   * AUTO} search), since that capacity also depends on max_iterations, graph degree, and dataset
+   * size, which are not known here.
+   */
+  public static final int MAX_SEARCH_WIDTH = 4_194_303;
+
   private final int iTopK;
   private final int searchWidth;
   private final int threadBlockSize;
@@ -117,13 +157,56 @@ public GPUKnnFloatVectorQuery(
       int maxIterations,
       CagraSearchParams.SearchAlgo searchAlgo) {
     super(field, target, k, filter);
+    validateSearchParameters(iTopK, searchWidth, k, searchAlgo);
     this.iTopK = iTopK;
     this.searchWidth = searchWidth;
     this.threadBlockSize = threadBlockSize;
     this.maxIterations = maxIterations;
     this.searchAlgo = searchAlgo;
   }
 
+  private static void validateSearchParameters(
+      int iTopK, int searchWidth, int k, CagraSearchParams.SearchAlgo searchAlgo) {
+    validateRange("iTopK", iTopK, MIN_ITOPK, MAX_ITOPK);
+    validateRange("searchWidth", searchWidth, MIN_SEARCH_WIDTH, MAX_SEARCH_WIDTH);
+    // This is a lower bound on the effective iTopK actually sent to native CAGRA: the filtered
+    // per-segment fallback path (see CuVS2510GPUVectorsReader) can raise topK further based on
+    // filter cardinality, so a later, authoritative check is required at that point too — see
+    // validateSingleCtaItopk below.
+    validateSingleCtaItopk(Math.max(iTopK, k), searchAlgo);
+  }
+
+  /**
+   * Validates that {@code effectiveITopK} — the itopk_size value actually about to be sent to
+   * native CAGRA — does not exceed the SINGLE_CTA algorithm's limit.
+   *
+   * <p>Callers that can further increase itopk_size after construction (e.g. the filtered
+   * per-segment fallback path, which raises topK based on filter cardinality) must call this
+   * again with the final, post-adjustment value immediately before building {@link
+   * CagraSearchParams}.
+   *
+   * @param effectiveITopK the itopk_size value about to be sent to native CAGRA
+   * @param searchAlgo the CAGRA search algorithm the query will run under
+   */
+  static void validateSingleCtaItopk(int effectiveITopK, CagraSearchParams.SearchAlgo searchAlgo) {
+    if (searchAlgo == CagraSearchParams.SearchAlgo.SINGLE_CTA
+        && effectiveITopK > MAX_SINGLE_CTA_ITOPK) {
+      throw new IllegalArgumentException(
+          "effective iTopK must not exceed "
+              + MAX_SINGLE_CTA_ITOPK
+              + " for SINGLE_CTA search, but was "
+              + effectiveITopK
+              + ".");
+    }
+  }
+
+  private static void validateRange(String name, int value, int min, int max) {
+    if (value < min || value > max) {
+      throw new IllegalArgumentException(
+          name + " not in valid range. Valid range: [" + min + ", " + max + "]");
+    }
+  }
+
   // -------------------------------------------------------------------------
   // Optimized multi-segment path
   // -------------------------------------------------------------------------

java/cuvs-lucene/src/main/java/com/nvidia/cuvs/lucene/GPUSearchParams.java

@@ -30,11 +30,9 @@ public static enum Strategy {
     CUSTOM
   }
 
-  /*
-   * TODO: Update boundaries for all parameters when a consensus is reached.
-   * Issue: https://github.com/rapidsai/cuvs-lucene/issues/99
-   */
+  /** Bounds for the public CAGRA build parameters. */
   public static final int MIN_WRITER_THREADS = 1;
+
   public static final int MAX_WRITER_THREADS = 512;
   public static final int MIN_INT_GRAPH_DEG = 2;
   public static final int MAX_INT_GRAPH_DEG = 512;
@@ -256,6 +254,7 @@ public static class Builder {
      * @return instance of {@link Builder}
      */
     public Builder withWriterThreads(int writerThreads) {
+      validateRange("writerThreads", writerThreads, MIN_WRITER_THREADS, MAX_WRITER_THREADS);
       this.writerThreads = writerThreads;
       return this;
     }
@@ -269,6 +268,8 @@ public Builder withWriterThreads(int writerThreads) {
      * @return instance of {@link Builder}
      */
     public Builder withIntermediateGraphDegree(int intermediateGraphDegree) {
+      validateRange(
+          "intermediateGraphDegree", intermediateGraphDegree, MIN_INT_GRAPH_DEG, MAX_INT_GRAPH_DEG);
       this.intermediateGraphDegree = intermediateGraphDegree;
       return this;
     }
@@ -282,6 +283,7 @@ public Builder withIntermediateGraphDegree(int intermediateGraphDegree) {
      * @return instance of {@link Builder}
      */
     public Builder withGraphDegree(int graphDegree) {
+      validateRange("graphDegree", graphDegree, MIN_GRAPH_DEG, MAX_GRAPH_DEG);
       this.graphdegree = graphDegree;
       return this;
     }
@@ -381,6 +383,13 @@ public Builder withBuildQuality(int buildQuality) {
       return this;
     }
 
+    private static void validateRange(String name, int value, int min, int max) {
+      if (value < min || value > max) {
+        throw new IllegalArgumentException(
+            name + " not in valid range. Valid range: [" + min + ", " + max + "]");
+      }
+    }
+
     /**
      * Validates the input parameters.
      *
@@ -412,6 +421,10 @@ private void validate() throws IllegalArgumentException {
                 + MAX_GRAPH_DEG
                 + "]");
       }
+      if (strategy == Strategy.CUSTOM && graphdegree > intermediateGraphDegree) {
+        throw new IllegalArgumentException(
+            "graphDegree must not be greater than intermediateGraphDegree.");
+      }
       if (Objects.isNull(cagraGraphBuildAlgo)) {
         throw new IllegalArgumentException("cagraGraphBuildAlgo cannot be null.");
       }