Bloom-filter fast-path for the dense tag store (O(1) insertion)

internal-api/src/main/java/datadog/trace/api/TagMap.java

@@ -1225,11 +1225,11 @@ static final class EmptyHolder {
 
   /**
    * Dense known-tag store (dense-tagmap-design §5). Values for KNOWN tags (those {@link
-   * KnownTagCodec#keyOf} resolves to a stored id) live in these INSERTION-ORDERED parallel arrays with
-   * NO per-tag {@link Entry} object — the allocation win. Lazily allocated on the first known-tag
-   * write ({@code null} until then, so all-unknown maps pay nothing) and grown x2 from {@link
-   * #KNOWN_INIT_CAP}. Matched by globalSerial via a linear scan ({@link #knownIndexOf}); reads
-   * aren't hot, so O(knownCount) is fine and positional indexing is deferred. Dormant until a
+   * KnownTagCodec#keyOf} resolves to a stored id) live in these INSERTION-ORDERED parallel arrays
+   * with NO per-tag {@link Entry} object — the allocation win. Lazily allocated on the first
+   * known-tag write ({@code null} until then, so all-unknown maps pay nothing) and grown x2 from
+   * {@link #KNOWN_INIT_CAP}. Matched by globalSerial via a linear scan ({@link #knownIndexOf});
+   * reads aren't hot, so O(knownCount) is fine and positional indexing is deferred. Dormant until a
    * resolver is registered: {@code keyOf} returns 0, so nothing routes here and production is
    * byte-identical.
    *
@@ -1247,6 +1247,15 @@ static final class EmptyHolder {
   private Object[] knownValues;
   private int knownCount;
 
+  /**
+   * Bloom-style presence filter over the dense store: a set bit means {@code tagId} MAY be present
+   * (scan to confirm), a clear bit means it is DEFINITELY absent (skip the scan, append in O(1) —
+   * the common per-build case). A superset of the present ids' bits: set on every add, never
+   * cleared on remove (a stale bit only costs a scan, never a wrong answer). So correctness never
+   * depends on the position→bit collision rate; only the fast-path hit rate does.
+   */
+  private long knownBloom;
+
   private static final int KNOWN_INIT_CAP = 8;
 
   /**
@@ -1552,27 +1561,45 @@ private void ensureKnownCapacity() {
     }
   }
 
+  /**
+   * Presence-filter bit for {@code tagId}. Crude position→bit map ({@code fieldPos} mod 64) to
+   * start; a collision-minimizing per-type coloring later only raises the hit rate — correctness
+   * never depends on it, because the scan is authoritative.
+   */
+  private static long knownBloomBit(long tagId) {
+    return 1L << (KnownTagCodec.fieldPos(tagId) & 63);
+  }
+
   /**
    * Stores a known tag's value densely (no {@link Entry} alloc). Overwrites in place when present
    * (returning the prior value materialized as an Entry, per the {@code Map} contract — usually
-   * discarded by {@code set}); otherwise appends, growing x2 as needed.
+   * discarded by {@code set}); otherwise appends, growing x2 as needed. The bloom filter skips the
+   * {@link #knownIndexOf} scan when the tag is definitely absent (the common per-build case), so an
+   * append is O(1) instead of O(n).
    */
   private Entry putKnownValue(long tagId, Object value) {
-    int i = this.knownIndexOf(tagId);
-    if (i >= 0) {
-      Object prior = this.knownValues[i];
-      this.knownValues[i] = value;
-      return materializeKnown(tagId, prior);
+    long bit = knownBloomBit(tagId);
+    if ((this.knownBloom & bit) != 0) {
+      // maybe present -> scan to overwrite in place
+      int i = this.knownIndexOf(tagId);
+      if (i >= 0) {
+        Object prior = this.knownValues[i];
+        this.knownValues[i] = value;
+        return materializeKnown(tagId, prior);
+      }
+      // bloom false positive (collision) -> fall through to append
     }
     this.ensureKnownCapacity();
     int slot = this.knownCount++;
     this.knownIds[slot] = tagId;
     this.knownValues[slot] = value;
+    this.knownBloom |= bit;
     return null;
   }
 
   /** Raw dense value for {@code tagId}, or {@code null} when absent (no Entry, no boxing). */
   private Object knownRawValue(long tagId) {
+    if ((this.knownBloom & knownBloomBit(tagId)) == 0) return null; // definitely absent, no scan
     int i = this.knownIndexOf(tagId);
     return i < 0 ? null : this.knownValues[i];
   }
@@ -1581,6 +1608,7 @@ private Object knownRawValue(long tagId) {
    * Removes a known tag from the dense store (swap-with-last), returning the prior Entry or null.
    */
   private Entry removeKnown(long tagId) {
+    if ((this.knownBloom & knownBloomBit(tagId)) == 0) return null; // definitely absent
     int i = this.knownIndexOf(tagId);
     if (i < 0) return null;
     Object prior = this.knownValues[i];
@@ -1589,6 +1617,8 @@ private Entry removeKnown(long tagId) {
     this.knownValues[i] = this.knownValues[last];
     this.knownIds[last] = 0L;
     this.knownValues[last] = null;
+    // knownBloom intentionally NOT cleared: a stale-set bit only costs a scan; clearing could drop
+    // a bit still shared (via collision) by a present id -> false negative.
     return materializeKnown(tagId, prior);
   }
 
@@ -1603,7 +1633,8 @@ private static Entry materializeKnown(long tagId, Object value) {
    * local bucket entry — known tags never bucket — so no bucket check is needed here.)
    */
   private boolean parentDenseHidden(long tagId) {
-    if (this.knownIndexOf(tagId) >= 0) return true; // shadowed by a local dense entry
+    // shadowed by a local dense entry (bloom prunes the scan when definitely absent)
+    if ((this.knownBloom & knownBloomBit(tagId)) != 0 && this.knownIndexOf(tagId) >= 0) return true;
     return this.removedFromParent != null
         && this.removedFromParent.contains(KnownTagCodec.nameOf(tagId)); // tombstoned
   }
@@ -2006,6 +2037,7 @@ private void putAllIntoEmptyMap(OptimizedTagMap that) {
       this.knownIds = Arrays.copyOf(that.knownIds, that.knownIds.length);
       this.knownValues = Arrays.copyOf(that.knownValues, that.knownValues.length);
       this.knownCount = that.knownCount;
+      this.knownBloom = that.knownBloom;
     }
   }
 
@@ -2048,7 +2080,8 @@ public void fillStringMap(Map<? super String, ? super String> stringMap) {
     }
     for (int i = 0; i < this.knownCount; ++i) {
       stringMap.put(
-          KnownTagCodec.nameOf(this.knownIds[i]), TagValueConversions.toString(this.knownValues[i]));
+          KnownTagCodec.nameOf(this.knownIds[i]),
+          TagValueConversions.toString(this.knownValues[i]));
     }
   }
 
@@ -2393,6 +2426,7 @@ public void clear() {
     this.knownIds = null;
     this.knownValues = null;
     this.knownCount = 0;
+    this.knownBloom = 0L;
   }
 
   public OptimizedTagMap freeze() {