Use doc values skipper for _tsid in synthetic _id postings (#138568)

Instead of scanning all documents to find the first document that has a _tsid less than, or equal to, a given ordinal we can use a doc values skipper to skip as much as possible documents, and only then scan. the remaining docs.

When seeking a synthetic _id, we look up the _tsid ordinal, then use DV skipper to find a starting doc ID, then scan each doc to find the first doc ID matching the exact _tsid ordinal. Then we finally scan remaining docs to find the one matching the timestamp.

Relates ES-13604

Author: tlrx

docs/changelog/138568.yaml

@@ -0,0 +1,5 @@
+pr: 138568
+summary: Use doc values skipper for `_tsid` in synthetic `_id` postings
+area: TSDB
+type: enhancement
+issues: []

server/src/main/java/org/elasticsearch/index/codec/tsdb/TSDBSyntheticIdFieldsProducer.java

@@ -280,22 +280,44 @@ private BytesRef lookupTsIdOrd(int tsIdOrdinal) throws IOException {
         }
 
         /**
-         * Scan all documents to find the first document that has a _tsid equal or greater than the provided _tsid ordinal, returning its
-         * document ID. If no document is found, the method returns {@link DocIdSetIterator#NO_MORE_DOCS}.
+         * Use a doc values skipper to find a starting document ID for the provided _tsid ordinal. The returned document ID might have the
+         * exact _tsid ordinal provided, or a lower one.
          *
-         * Warning: This method is very slow because it potentially scans all documents in the segment.
+         * @param tsIdOrd the _tsid ordinal
+         * @return a docID to start scanning documents from in order to find the first document ID matching the provided _tsid
+         * @throws IOException if any I/O exception occurs
          */
-        private int slowScanToFirstDocWithTsIdOrdinalEqualOrGreaterThan(int tsIdOrd) throws IOException {
+        private int findStartDocIDForTsIdOrd(int tsIdOrd) throws IOException {
+            var skipper = docValuesProducer.getSkipper(tsIdFieldInfo);
+            assert skipper != null;
+            if (skipper.minValue() > tsIdOrd || tsIdOrd > skipper.maxValue()) {
+                return DocIdSetIterator.NO_MORE_DOCS;
+            }
+            skipper.advance(tsIdOrd, Long.MAX_VALUE);
+            return skipper.minDocID(0);
+        }
+
+        /**
+         * Find the first document that has a _tsid equal or greater than the provided _tsid ordinal, returning its document ID. If no
+         * document is found, the method returns {@link DocIdSetIterator#NO_MORE_DOCS}.
+         *
+         * Warning: This method can be slow because it potentially scans many documents in the segment.
+         */
+        private int findFirstDocWithTsIdOrdinalEqualOrGreaterThan(int tsIdOrd) throws IOException {
+            final int startDocId = findStartDocIDForTsIdOrd(tsIdOrd);
+            if (startDocId == DocIdSetIterator.NO_MORE_DOCS) {
+                return startDocId;
+            }
             // recreate even if doc values are already on the same ordinal, to ensure the method returns the first doc
-            if (tsIdDocValues == null || (cachedTsIdOrd != -1 && cachedTsIdOrd >= tsIdOrd)) {
+            if (tsIdDocValues == null || (cachedTsIdOrd != -1 && cachedTsIdOrd >= tsIdOrd) || tsIdDocValues.docID() > startDocId) {
                 tsIdDocValues = docValuesProducer.getSorted(tsIdFieldInfo);
                 cachedTsIdOrd = -1;
                 cachedTsId = null;
             }
             assert 0 <= tsIdOrd : tsIdOrd;
             assert tsIdOrd < tsIdDocValues.getValueCount() : tsIdOrd;
 
-            for (int docID = 0; docID != DocIdSetIterator.NO_MORE_DOCS; docID = tsIdDocValues.nextDoc()) {
+            for (int docID = startDocId; docID != DocIdSetIterator.NO_MORE_DOCS; docID = tsIdDocValues.nextDoc()) {
                 boolean found = tsIdDocValues.advanceExact(docID);
                 assert found : "No value found for field [" + tsIdFieldInfo.getName() + " and docID " + docID;
                 var ord = tsIdDocValues.ordValue();
@@ -313,22 +335,25 @@ private int slowScanToFirstDocWithTsIdOrdinalEqualOrGreaterThan(int tsIdOrd) thr
         }
 
         /**
-         * Scan all documents to find the first document that has a _tsid equal to the provided _tsid ordinal, returning its
-         * document ID. If no document is found, the method returns {@link DocIdSetIterator#NO_MORE_DOCS}.
+         * Find the first document that has a _tsid equal to the provided _tsid ordinal, returning its document ID. If no document is found,
+         * the method returns {@link DocIdSetIterator#NO_MORE_DOCS}.
          *
-         * Warning: This method is very slow because it potentially scans all documents in the segment.
+         * Warning: This method can be slow because it potentially scans many documents in the segment.
          */
-        private int slowScanToFirstDocWithTsIdOrdinalEqualTo(int tsIdOrd) throws IOException {
+        private int findFirstDocWithTsIdOrdinalEqualTo(int tsIdOrd) throws IOException {
+            final int startDocId = findStartDocIDForTsIdOrd(tsIdOrd);
+            assert startDocId != DocIdSetIterator.NO_MORE_DOCS : startDocId;
+
             // recreate even if doc values are already on the same ordinal, to ensure the method returns the first doc
-            if (tsIdDocValues == null || (cachedTsIdOrd != -1 && cachedTsIdOrd >= tsIdOrd)) {
+            if (tsIdDocValues == null || (cachedTsIdOrd != -1 && cachedTsIdOrd >= tsIdOrd) || tsIdDocValues.docID() > startDocId) {
                 tsIdDocValues = docValuesProducer.getSorted(tsIdFieldInfo);
                 cachedTsIdOrd = -1;
                 cachedTsId = null;
             }
             assert 0 <= tsIdOrd : tsIdOrd;
             assert tsIdOrd < tsIdDocValues.getValueCount() : tsIdOrd;
 
-            for (int docID = 0; docID != DocIdSetIterator.NO_MORE_DOCS; docID = tsIdDocValues.nextDoc()) {
+            for (int docID = startDocId; docID != DocIdSetIterator.NO_MORE_DOCS; docID = tsIdDocValues.nextDoc()) {
                 boolean found = tsIdDocValues.advanceExact(docID);
                 assert found : "No value found for field [" + tsIdFieldInfo.getName() + " and docID " + docID;
                 var ord = tsIdDocValues.ordValue();
@@ -441,7 +466,7 @@ public SeekStatus seekCeil(BytesRef id) throws IOException {
                 tsIdOrd = -tsIdOrd - 1;
                 // set the terms enum on the first non-matching document
                 if (tsIdOrd < docValues.getTsIdValueCount()) {
-                    int docID = docValues.slowScanToFirstDocWithTsIdOrdinalEqualOrGreaterThan(tsIdOrd);
+                    int docID = docValues.findFirstDocWithTsIdOrdinalEqualOrGreaterThan(tsIdOrd);
                     if (docID != DocIdSetIterator.NO_MORE_DOCS) {
                         current = new SyntheticTerm(
                             docID,
@@ -461,8 +486,8 @@ public SeekStatus seekCeil(BytesRef id) throws IOException {
             // _tsid found, extract the timestamp
             final long timestamp = TsidExtractingIdFieldMapper.extractTimestampFromSyntheticId(id);
 
-            // Slow scan to the first document matching the _tsid
-            final int startDocID = docValues.slowScanToFirstDocWithTsIdOrdinalEqualTo(tsIdOrd);
+            // Find the first document matching the _tsid
+            final int startDocID = docValues.findFirstDocWithTsIdOrdinalEqualTo(tsIdOrd);
             assert startDocID >= 0 : startDocID;
 
             int docID = startDocID;

server/src/main/java/org/elasticsearch/index/mapper/ParsedDocument.java

@@ -75,6 +75,7 @@ public static ParsedDocument noopTombstone(SeqNoFieldMapper.SeqNoIndexOptions se
      * The returned document consists only _uid, _seqno, _term and _version fields; other metadata fields are excluded.
      * @param id                the id of the deleted document
      */
+    // used by tests
     public static ParsedDocument deleteTombstone(SeqNoFieldMapper.SeqNoIndexOptions seqNoIndexOptions, String id) {
         return deleteTombstone(seqNoIndexOptions, false /* ignored */, false, id, null /* ignored */);
     }
@@ -101,6 +102,9 @@ public static ParsedDocument deleteTombstone(
             // Use a synthetic _id field which is not indexed nor stored
             document.add(IdFieldMapper.syntheticIdField(id));
 
+            // Add doc values fields that are used to synthesize the synthetic _id.
+            // Note: It is not strictly required for tombstones documents but we decided to add them so that iterating and seeking synthetic
+            // _id terms over tombstones also work as if a regular _id field was present.
             var timeSeriesId = TsidExtractingIdFieldMapper.extractTimeSeriesIdFromSyntheticId(uid);
             var timestamp = TsidExtractingIdFieldMapper.extractTimestampFromSyntheticId(uid);
             var routingHash = TsidExtractingIdFieldMapper.extractRoutingHashBytesFromSyntheticId(uid);