HBASE-25709 Close region may stuck when region is compacting and skipped most cells read

Author: sunhelly

hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/NoLimitScannerContext.java

@@ -36,7 +36,7 @@
 public class NoLimitScannerContext extends ScannerContext {
 
   public NoLimitScannerContext() {
-    super(false, null, false);
+    super(false, null, false, false);
   }
 
   /**

hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/ScannerContext.java

@@ -110,7 +110,13 @@ public class ScannerContext {
    */
   final ServerSideScanMetrics metrics;
 
-  ScannerContext(boolean keepProgress, LimitFields limitsToCopy, boolean trackMetrics) {
+  /**
+   * Set this to true will prevent the looping skip of heap cells, see HBASE-25709.
+   */
+  private boolean preventLoopReadEnabled = false;
+
+  ScannerContext(boolean keepProgress, LimitFields limitsToCopy, boolean trackMetrics,
+    boolean preventLoopReadEnabled) {
     this.limits = new LimitFields();
     if (limitsToCopy != null) {
       this.limits.copy(limitsToCopy);
@@ -122,12 +128,17 @@ public class ScannerContext {
     this.keepProgress = keepProgress;
     this.scannerState = DEFAULT_STATE;
     this.metrics = trackMetrics ? new ServerSideScanMetrics() : null;
+    this.preventLoopReadEnabled = preventLoopReadEnabled;
   }
 
   public boolean isTrackingMetrics() {
     return this.metrics != null;
   }
 
+  public boolean isPreventLoopReadEnabled() {
+    return preventLoopReadEnabled;
+  }
+
   /**
    * Get the metrics instance. Should only be called after a call to {@link #isTrackingMetrics()}
    * has been made to confirm that metrics are indeed being tracked.
@@ -372,6 +383,7 @@ public static final class Builder {
     boolean keepProgress = DEFAULT_KEEP_PROGRESS;
     boolean trackMetrics = false;
     LimitFields limits = new LimitFields();
+    boolean preventLoopReadEnabled = false;
 
     private Builder() {
     }
@@ -408,8 +420,13 @@ public Builder setBatchLimit(int batchLimit) {
       return this;
     }
 
+    public Builder setPreventLoopReadEnabled(boolean enabled) {
+      this.preventLoopReadEnabled = enabled;
+      return this;
+    }
+
     public ScannerContext build() {
-      return new ScannerContext(keepProgress, limits, trackMetrics);
+      return new ScannerContext(keepProgress, limits, trackMetrics, preventLoopReadEnabled);
     }
   }
 

hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/StoreScanner.java

@@ -751,6 +751,10 @@ public boolean next(List<Cell> outResult, ScannerContext scannerContext) throws
           default:
             throw new RuntimeException("UNEXPECTED");
         }
+        // when reach heartbeat check, try to avoid loop read cells
+        if (kvsScanned % cellsPerHeartbeatCheck == 0 && scannerContext.isPreventLoopReadEnabled()) {
+          return scannerContext.setScannerState(NextState.MORE_VALUES).hasMoreValues();
+        }
       } while ((cell = this.heap.peek()) != null);
 
       if (count > 0) {

hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/compactions/Compactor.java

@@ -401,8 +401,10 @@ protected boolean performCompaction(FileDetails fd, InternalScanner scanner, Cel
     String compactionName = ThroughputControlUtil.getNameForThrottling(store, "compaction");
     long now = 0;
     boolean hasMore;
-    ScannerContext scannerContext =
-          ScannerContext.newBuilder().setBatchLimit(compactionKVMax).build();
+    ScannerContext scannerContext = ScannerContext.newBuilder()
+      .setBatchLimit(compactionKVMax)
+      .setPreventLoopReadEnabled(true)
+      .build();
 
     throughputController.start(compactionName);
     KeyValueScanner kvs = (scanner instanceof KeyValueScanner) ? (KeyValueScanner) scanner : null;