In-situ tombstone metrics (#524)

Co-authored-by: svc-autorelease <[email protected]>
palantir · Oct 16, 2024 · c3ab8d6 · c3ab8d6
1 parent 7ca0451
commit c3ab8d6
Show file tree

Hide file tree

Showing 3 changed files with 40 additions and 29 deletions.
diff --git a/src/java/com/palantir/cassandra/utils/CountingCellIterator.java b/src/java/com/palantir/cassandra/utils/CountingCellIterator.java
@@ -20,6 +20,7 @@
 
 import org.apache.cassandra.db.Cell;
 import org.apache.cassandra.db.ExpiringCell;
+import org.apache.cassandra.metrics.ColumnFamilyMetrics;
 import org.apache.cassandra.utils.MergeIterator;
 
 /**
@@ -38,6 +39,7 @@
 public class CountingCellIterator implements Iterator<Cell> {
     private final Iterator<Cell> delegate;
 
+    private final ColumnFamilyMetrics metrics;
     /*
      * Given the way SliceQueryFilter is passed these values, per ColumnFamilyStore.getColumnFamily(QueryFilter filter),
      * it is generally expected that gcBefore will be equivalent to { (now / 1000) - gc_grace_seconds }
@@ -50,12 +52,13 @@ public class CountingCellIterator implements Iterator<Cell> {
     private int liveCells = 0;
     private int tombstones = 0;
 
-    public static CountingCellIterator wrapIterator(Iterator<Cell> delegate, long timestamp, long gcBefore) {
-        return new CountingCellIterator(delegate, timestamp, gcBefore);
+    public static CountingCellIterator wrapIterator(Iterator<Cell> delegate, ColumnFamilyMetrics metrics, long timestamp, long gcBefore) {
+        return new CountingCellIterator(delegate, metrics, timestamp, gcBefore);
     }
 
-    private CountingCellIterator(Iterator<Cell> delegate, long timestamp, long gcBefore) {
+    private CountingCellIterator(Iterator<Cell> delegate, ColumnFamilyMetrics metrics, long timestamp, long gcBefore) {
         this.delegate = delegate;
+        this.metrics = metrics;
         this.gcBeforeSeconds = gcBefore;
         this.nowMillis = timestamp;
     }
@@ -103,6 +106,9 @@ private void count(Cell cell) {
         } else if (ExpiringCell.class.isAssignableFrom(cell.getClass())) {
             droppableTtls++;
         } else if (cell.getLocalDeletionTime() < gcBeforeSeconds) {
+            if (metrics != null) {
+                metrics.droppableTombstones.mark();
+            }
             droppableTombstones++;
         } else {
             tombstones++;

diff --git a/src/java/org/apache/cassandra/db/ColumnFamilyStore.java b/src/java/org/apache/cassandra/db/ColumnFamilyStore.java
@@ -2060,6 +2060,10 @@ public ColumnFamily getColumnFamily(QueryFilter filter)
 
         ColumnFamily result = null;
 
+        if (filter.filter instanceof SliceQueryFilter) {
+            ((SliceQueryFilter) filter.filter).setMetrics(metric);
+        }
+
         long start = System.nanoTime();
 
         int readDelay = DatabaseDescriptor.getReadDelay();
@@ -2127,7 +2131,6 @@ public ColumnFamily getColumnFamily(QueryFilter filter)
     }
 
     private void recordMetrics(SliceQueryFilter filter) {
-        metric.droppableTombstones.mark(filter.lastReadDroppableTombstones());
         // Log the number of tombstones scanned on single key queries
         metric.tombstoneScannedHistogram.update(filter.lastTombstones());
         metric.liveScannedHistogram.update(filter.lastLive());
@@ -2136,21 +2139,6 @@ private void recordMetrics(SliceQueryFilter filter) {
         metric.liveReadHistogram.update(filter.lastReadLive());
         metric.tombstonesReadHistogram.update(filter.lastReadTombstones());
 
-        Optional<DeletionInfo> maybeDeletionInfo = filter.lastReadDeletionInfo();
-
-        if (maybeDeletionInfo.isPresent()) {
-            DeletionInfo deletionInfo = maybeDeletionInfo.get();
-            logger.trace("Ranged tombstones read {} and droppable {} for {}.{}",
-                         deletionInfo.getRangeTombstoneCounter().getNonDroppableCount(),
-                         deletionInfo.getRangeTombstoneCounter().getDroppableCount(), keyspace, getColumnFamilyName());
-
-            metric.rangeTombstonesReadHistogram.update(deletionInfo.getRangeTombstoneCounter().getNonDroppableCount());
-            metric.droppableRangeTombstonesReadHistogram.update(deletionInfo.getRangeTombstoneCounter().getDroppableCount());
-            metric.rangeTombstonesHistogram.update(deletionInfo.rangeCount());
-
-            metric.droppableTombstones.mark(deletionInfo.getRangeTombstoneCounter().getDroppableCount());
-        }
-
         if (filter.hitTombstoneWarnThreshold()) metric.tombstoneWarnings.inc();
     }
 
@@ -2586,6 +2574,11 @@ public List<Row> filter(AbstractScanIterator rowIterator, ExtendedFilter filter)
 
                 if (rowIterator.needsFiltering())
                 {
+                    IDiskAtomFilter queryFilter = filter.columnFilter(rawRow.key.getKey());
+                    if (queryFilter instanceof SliceQueryFilter) {
+                        ((SliceQueryFilter) queryFilter).setMetrics(metric);
+                    }
+
                     IDiskAtomFilter extraFilter = filter.getExtraFilter(rawRow.key, data);
                     if (extraFilter != null)
                     {
@@ -2603,7 +2596,6 @@ public List<Row> filter(AbstractScanIterator rowIterator, ExtendedFilter filter)
                     // cut the resultset back to what was requested, if necessary
                     data = filter.prune(rawRow.key, data);
 
-                    IDiskAtomFilter queryFilter = filter.columnFilter(rawRow.key.getKey());
                     if (queryFilter instanceof SliceQueryFilter) {
                         recordMetrics((SliceQueryFilter) queryFilter);
                     }

diff --git a/src/java/org/apache/cassandra/db/filter/SliceQueryFilter.java b/src/java/org/apache/cassandra/db/filter/SliceQueryFilter.java
@@ -30,6 +30,7 @@
 import org.apache.cassandra.config.CFMetaData;
 import org.apache.cassandra.db.composites.*;
 import org.apache.cassandra.io.util.FileUtils;
+import org.apache.cassandra.metrics.ColumnFamilyMetrics;
 import org.apache.cassandra.utils.Pair;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
@@ -65,15 +66,15 @@ public class SliceQueryFilter implements IDiskAtomFilter
     private boolean hitTombstoneFailureThreshold = false;
     private boolean hitTombstoneWarnThreshold = false;
 
-    private DeletionInfo deletionInfo;
-
     // Not serialized, just a ack for range slices to find the number of live column counted, even when we group
     private ColumnCounter columnCounter;
 
     private CountingCellIterator reducedCells;
 
     private RangeTombstoneCounter rangeTombstoneCounter = new RangeTombstoneCounter();
 
+    private volatile ColumnFamilyMetrics metrics;
+
     public SliceQueryFilter(Composite start, Composite finish, boolean reversed, int count)
     {
         this(new ColumnSlice(start, finish), reversed, count);
@@ -114,6 +115,10 @@ public SliceQueryFilter(ColumnSlice[] slices, boolean reversed, int count, int c
                                              : null;
     }
 
+    public void setMetrics(ColumnFamilyMetrics metrics) {
+        this.metrics = metrics;
+    }
+
     public SliceQueryFilter cloneShallow()
     {
         return new SliceQueryFilter(slices, reversed, count, compositesToGroup);
@@ -276,10 +281,22 @@ public Comparator<Cell> getColumnComparator(CellNameType comparator)
 
     public void collectReducedColumns(ColumnFamily container, Iterator<Cell> reducedColumns, DecoratedKey key, int gcBefore, long now)
     {
-        reducedCells = CountingCellIterator.wrapIterator(reducedColumns, now, gcBefore);
+        reducedCells = CountingCellIterator.wrapIterator(reducedColumns, metrics, now, gcBefore);
         columnCounter = columnCounter(container.getComparator(), now);
-        deletionInfo = container.deletionInfo();
-        DeletionInfo.InOrderTester tester = container.deletionInfo().inOrderTester(reversed);
+        DeletionInfo deletionInfo = container.deletionInfo();
+        logger.trace("Ranged tombstones read {} and droppable {} for {}",
+                     deletionInfo.getRangeTombstoneCounter().getNonDroppableCount(),
+                     deletionInfo.getRangeTombstoneCounter().getDroppableCount(), container.metadata().ksAndCFName);
+
+        if (metrics != null)
+        {
+            metrics.rangeTombstonesReadHistogram.update(deletionInfo.getRangeTombstoneCounter().getNonDroppableCount());
+            metrics.droppableRangeTombstonesReadHistogram.update(deletionInfo.getRangeTombstoneCounter().getDroppableCount());
+            metrics.rangeTombstonesHistogram.update(deletionInfo.rangeCount());
+
+            metrics.droppableTombstones.mark(deletionInfo.getRangeTombstoneCounter().getDroppableCount());
+        }
+        DeletionInfo.InOrderTester tester = deletionInfo.inOrderTester(reversed);
 
         boolean hasBreachedCollectionThreshold = false;
         long dataSizeCollected = 0;
@@ -500,10 +517,6 @@ public int lastReadTombstones()
         return reducedCells == null ? 0 : reducedCells.tombstones();
     }
 
-    public Optional<DeletionInfo> lastReadDeletionInfo() {
-        return Optional.ofNullable(deletionInfo);
-    }
-
     @Override
     public String toString()
     {