LUCENE-8358: Relax assertion in IW#writeSomeDocValuesUpdates

This assertion is too strict since we can see this situation if for instance
a ReadersAndUpdates instance gets written to disk concurrently and
readerpooling is off. This change also simplifies ReaderPool#getReadersByRam and
adds a test for it.

lucene/core/src/java/org/apache/lucene/index/IndexWriter.java

@@ -29,7 +29,6 @@ import java.util.LinkedList;
 import java.util.List;
 import java.util.Locale;
 import java.util.Map;
-import java.util.PriorityQueue;
 import java.util.Queue;
 import java.util.Set;
 import java.util.concurrent.ConcurrentLinkedQueue;
@@ -592,25 +591,27 @@ public class IndexWriter implements Closeable, TwoPhaseCommit, Accountable,
             }
 
             // Sort by largest ramBytesUsed:
-            PriorityQueue<ReadersAndUpdates> queue = readerPool.getReadersByRam();
+            final List<ReadersAndUpdates> list = readerPool.getReadersByRam();
             int count = 0;
-            while (ramBytesUsed > 0.5 * ramBufferSizeMB * 1024 * 1024) {
-              ReadersAndUpdates rld = queue.poll();
-              if (rld == null) {
+            for (ReadersAndUpdates rld : list) {
+
+              if (ramBytesUsed <= 0.5 * ramBufferSizeMB * 1024 * 1024) {
                 break;
               }
-
               // We need to do before/after because not all RAM in this RAU is used by DV updates, and
               // not all of those bytes can be written here:
               long bytesUsedBefore = rld.ramBytesUsed.get();
-
+              if (bytesUsedBefore == 0) {
+                continue; // nothing to do here - lets not acquire the lock
+              }
               // Only acquire IW lock on each write, since this is a time consuming operation.  This way
               // other threads get a chance to run in between our writes.
               synchronized (this) {
                 // It's possible that the segment of a reader returned by readerPool#getReadersByRam
                 // is dropped before being processed here. If it happens, we need to skip that reader.
+                // this is also best effort to free ram, there might be some other thread writing this rld concurrently
+                // which wins and then if readerPooling is off this rld will be dropped.
                 if (readerPool.get(rld.info, false) == null) {
-                  assert segmentInfos.contains(rld.info) == false : "Segment [" + rld.info + "] is not dropped yet";
                   continue;
                 }
                 if (rld.writeFieldUpdates(directory, globalFieldNumberMap, bufferedUpdatesStream.getCompletedDelGen(), infoStream)) {

lucene/core/src/java/org/apache/lucene/index/ReaderPool.java

@@ -19,19 +19,22 @@ package org.apache.lucene.index;
 
 import java.io.Closeable;
 import java.io.IOException;
+import java.util.ArrayList;
 import java.util.Collection;
+import java.util.Collections;
 import java.util.HashMap;
 import java.util.HashSet;
 import java.util.Iterator;
 import java.util.List;
 import java.util.Map;
-import java.util.PriorityQueue;
 import java.util.Set;
 import java.util.concurrent.atomic.AtomicBoolean;
 import java.util.function.LongSupplier;
+import java.util.stream.Collectors;
 
 import org.apache.lucene.store.AlreadyClosedException;
 import org.apache.lucene.store.Directory;
+import org.apache.lucene.util.CollectionUtil;
 import org.apache.lucene.util.IOUtils;
 import org.apache.lucene.util.InfoStream;
 
@@ -243,16 +246,36 @@ final class ReaderPool implements Closeable {
     return any;
   }
 
-  PriorityQueue<ReadersAndUpdates> getReadersByRam() {
-    // Sort by largest ramBytesUsed:
-    PriorityQueue<ReadersAndUpdates> queue = new PriorityQueue<>(readerMap.size(),
-        (a, b) -> Long.compare(b.ramBytesUsed.get(), a.ramBytesUsed.get()));
+  /**
+   * Returns a list of all currently maintained ReadersAndUpdates sorted by it's ram consumption largest to smallest.
+   * This list can also contain readers that don't consume any ram at this point ie. don't have any updates buffered.
+   */
+  synchronized List<ReadersAndUpdates> getReadersByRam() {
+    class RamRecordingHolder {
+      final ReadersAndUpdates updates;
+      final long ramBytesUsed;
+      RamRecordingHolder(ReadersAndUpdates updates) {
+        this.updates = updates;
+        this.ramBytesUsed = updates.ramBytesUsed.get();
+      }
+    }
+    final ArrayList<RamRecordingHolder> readersByRam;
     synchronized (this) {
+      if (readerMap.isEmpty()) {
+        return Collections.emptyList();
+      }
+      readersByRam = new ArrayList<>(readerMap.size());
       for (ReadersAndUpdates rld : readerMap.values()) {
-        queue.add(rld);
+        // we have to record the ram usage once and then sort
+        // since the ram usage can change concurrently and that will confuse the sort or hit an assertion
+        // the we can acquire here is not enough we would need to lock all ReadersAndUpdates to make sure it doesn't
+        // change
+        readersByRam.add(new RamRecordingHolder(rld));
       }
     }
-    return queue;
+    // Sort this outside of the lock by largest ramBytesUsed:
+    CollectionUtil.introSort(readersByRam, (a, b) -> Long.compare(b.ramBytesUsed, a.ramBytesUsed));
+    return Collections.unmodifiableList(readersByRam.stream().map(h -> h.updates).collect(Collectors.toList()));
   }
 
 

lucene/core/src/test/org/apache/lucene/index/TestBinaryDocValuesUpdates.java

@@ -88,7 +88,7 @@ public class TestBinaryDocValuesUpdates extends LuceneTestCase {
     return doc;
   }
   
-  public void testUpdatesAreFlushed() throws IOException {
+  public void testUpdatesAreFlushed() throws IOException, InterruptedException {
     Directory dir = newDirectory();
     IndexWriter writer = new IndexWriter(dir, newIndexWriterConfig(new MockAnalyzer(random(), MockTokenizer.WHITESPACE, false))
                                                 .setRAMBufferSizeMB(0.00000001));

lucene/core/src/test/org/apache/lucene/index/TestReaderPool.java

@@ -19,6 +19,7 @@ package org.apache.lucene.index;
 
 import java.io.IOException;
 import java.util.Collections;
+import java.util.List;
 import java.util.concurrent.CountDownLatch;
 import java.util.concurrent.atomic.AtomicBoolean;
 
@@ -30,6 +31,7 @@ import org.apache.lucene.document.StringField;
 import org.apache.lucene.search.DocIdSetIterator;
 import org.apache.lucene.store.Directory;
 import org.apache.lucene.store.IOContext;
+import org.apache.lucene.util.BytesRef;
 import org.apache.lucene.util.IOSupplier;
 import org.apache.lucene.util.IOUtils;
 import org.apache.lucene.util.LuceneTestCase;
@@ -278,4 +280,34 @@ public class TestReaderPool extends LuceneTestCase {
     writer.close();
     return writer.globalFieldNumberMap;
   }
+
+  public void testGetReaderByRam() throws IOException {
+    Directory directory = newDirectory();
+    FieldInfos.FieldNumbers fieldNumbers = buildIndex(directory);
+    StandardDirectoryReader reader = (StandardDirectoryReader) DirectoryReader.open(directory);
+    SegmentInfos segmentInfos = reader.segmentInfos.clone();
+    ReaderPool pool = new ReaderPool(directory, directory, segmentInfos, fieldNumbers, () -> 0l,
+        new NullInfoStream(), null, null);
+    assertEquals(0, pool.getReadersByRam().size());
+
+    int ord = 0;
+    for (SegmentCommitInfo commitInfo : segmentInfos) {
+      ReadersAndUpdates readersAndUpdates = pool.get(commitInfo, true);
+      BinaryDocValuesFieldUpdates test = new BinaryDocValuesFieldUpdates(0, "test", commitInfo.info.maxDoc());
+      test.add(0, new BytesRef(new byte[ord++]));
+      test.finish();
+      readersAndUpdates.addDVUpdate(test);
+    }
+
+    List<ReadersAndUpdates> readersByRam = pool.getReadersByRam();
+    assertEquals(segmentInfos.size(), readersByRam.size());
+    long previousRam = Long.MAX_VALUE;
+    for (ReadersAndUpdates rld : readersByRam) {
+      assertTrue("previous: " + previousRam + " now: " + rld.ramBytesUsed.get(), previousRam >= rld.ramBytesUsed.get());
+      previousRam = rld.ramBytesUsed.get();
+      rld.dropChanges();
+      pool.drop(rld.info);
+    }
+    IOUtils.close(pool, reader, directory);
+  }
 }