Reuse TantivySearcher across queries via searcher pool

Author: chenghuichen

paimon-tantivy/paimon-tantivy-index/src/main/java/org/apache/paimon/tantivy/index/ArchiveLayout.java

@@ -18,7 +18,6 @@
 
 package org.apache.paimon.tantivy.index;
 
-/** Parsed archive header: file names, byte offsets, and lengths within the archive stream. */
 class ArchiveLayout {
     final String[] fileNames;
     final long[] fileOffsets;

paimon-tantivy/paimon-tantivy-index/src/main/java/org/apache/paimon/tantivy/index/TantivyFullTextGlobalIndexerFactory.java

@@ -28,14 +28,6 @@ public class TantivyFullTextGlobalIndexerFactory implements GlobalIndexerFactory
 
     public static final String IDENTIFIER = "tantivy-fulltext";
 
-    /**
-     * Shared pool across all indexers created by this factory. This factory instance is a JVM-level
-     * singleton (loaded once via {@link java.util.ServiceLoader}), so the pool naturally survives
-     * across queries and scanners.
-     *
-     * <p>The pool is initialized lazily on the first {@link #create} call so that the pool size can
-     * be read from user-supplied options.
-     */
     private volatile TantivySearcherPool searcherPool;
 
     @Override

paimon-tantivy/paimon-tantivy-index/src/main/java/org/apache/paimon/tantivy/index/TantivySearcherPool.java

@@ -35,12 +35,10 @@
  * Pool of {@link TantivySearcher} instances keyed by index file identity ({@code filePath@size}).
  *
  * <p>Each searcher holds the Tantivy index open in Rust memory (including the FST term dictionary).
- * Pooling avoids the repeated cost of loading the index on every query.
  *
  * <p>At most one idle searcher is kept per key. Under concurrent queries on the same shard, the
- * last entry to be returned wins; the others are closed immediately. Entries idle for more than
- * {@link #EXPIRE_AFTER_ACCESS_MINUTES} minutes are evicted and closed automatically. The total
- * number of idle entries across all keys is bounded by {@code maxSize}.
+ * last entry to be returned wins; the others are closed immediately. The total number of idle
+ * entries across all keys is bounded by {@code maxSize}.
  *
  * <p>Thread-safe. Borrow/return semantics guarantee at most one thread uses a given entry at a
  * time.