HBASE-26649 Support meta replica LoadBalance mode for RegionLocator#getAllRegionLocations() (#4442) (#4464)

Signed-off-by: Duo Zhang <[email protected]>

Author: huaxiangsun

hbase-client/src/main/java/org/apache/hadoop/hbase/AsyncMetaTableAccessor.java

@@ -17,6 +17,7 @@
  */
 package org.apache.hadoop.hbase;
 
+import static org.apache.hadoop.hbase.client.RegionLocator.LOCATOR_META_REPLICAS_MODE;
 import static org.apache.hadoop.hbase.util.FutureUtils.addListener;
 
 import java.io.IOException;
@@ -29,6 +30,7 @@
 import java.util.Optional;
 import java.util.SortedMap;
 import java.util.concurrent.CompletableFuture;
+import java.util.concurrent.ThreadLocalRandom;
 import java.util.regex.Matcher;
 import java.util.regex.Pattern;
 import java.util.stream.Collectors;
@@ -295,7 +297,37 @@ private static CompletableFuture<Void> scanMeta(AsyncTable<AdvancedScanResultCon
     }
 
     CompletableFuture<Void> future = new CompletableFuture<Void>();
-    metaTable.scan(scan, new MetaTableScanResultConsumer(rowUpperLimit, visitor, future));
+    // Get the region locator's meta replica mode.
+    CatalogReplicaMode metaReplicaMode = CatalogReplicaMode.fromString(metaTable.getConfiguration()
+      .get(LOCATOR_META_REPLICAS_MODE, CatalogReplicaMode.NONE.toString()));
+
+    if (metaReplicaMode == CatalogReplicaMode.LOAD_BALANCE) {
+      addListener(metaTable.getDescriptor(), (desc, error) -> {
+        if (error != null) {
+          LOG.error("Failed to get meta table descriptor, error: ", error);
+          future.completeExceptionally(error);
+          return;
+        }
+
+        int numOfReplicas = desc.getRegionReplication();
+        if (numOfReplicas > 1) {
+          int replicaId = ThreadLocalRandom.current().nextInt(numOfReplicas);
+
+          // When the replicaId is 0, do not set to Consistency.TIMELINE
+          if (replicaId > 0) {
+            scan.setReplicaId(replicaId);
+            scan.setConsistency(Consistency.TIMELINE);
+          }
+        }
+        metaTable.scan(scan, new MetaTableScanResultConsumer(rowUpperLimit, visitor, future));
+      });
+    } else {
+      if (metaReplicaMode == CatalogReplicaMode.HEDGED_READ) {
+        scan.setConsistency(Consistency.TIMELINE);
+      }
+      metaTable.scan(scan, new MetaTableScanResultConsumer(rowUpperLimit, visitor, future));
+    }
+
     return future;
   }
 

…oop/hbase/client/CatalogReplicaMode.java …che/hadoop/hbase/CatalogReplicaMode.javahbase-client/src/main/java/org/apache/hadoop/hbase/client/CatalogReplicaMode.java renamed to hbase-client/src/main/java/org/apache/hadoop/hbase/CatalogReplicaMode.java hbase-client/src/main/java/org/apache/hadoop/hbase/client/CatalogReplicaMode.java renamed to hbase-client/src/main/java/org/apache/hadoop/hbase/CatalogReplicaMode.java

@@ -15,7 +15,7 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-package org.apache.hadoop.hbase.client;
+package org.apache.hadoop.hbase;
 
 import org.apache.yetus.audience.InterfaceAudience;
 
@@ -33,7 +33,7 @@
  * </ol>
  */
 @InterfaceAudience.Private
-enum CatalogReplicaMode {
+public enum CatalogReplicaMode {
   NONE {
     @Override
     public String toString() {

hbase-client/src/main/java/org/apache/hadoop/hbase/MetaTableAccessor.java

@@ -33,6 +33,7 @@
 import java.util.NavigableMap;
 import java.util.SortedMap;
 import java.util.TreeMap;
+import java.util.concurrent.ThreadLocalRandom;
 import java.util.regex.Matcher;
 import java.util.regex.Pattern;
 import java.util.stream.Collectors;
@@ -697,15 +698,21 @@ public static void fullScanMetaAndPrint(Connection connection) throws IOExceptio
     scanMeta(connection, null, null, QueryType.ALL, v);
   }
 
+  public static void scanMetaForTableRegions(Connection connection, Visitor visitor,
+    TableName tableName, CatalogReplicaMode metaReplicaMode) throws IOException {
+    scanMeta(connection, tableName, QueryType.REGION, Integer.MAX_VALUE, visitor, metaReplicaMode);
+  }
+
   public static void scanMetaForTableRegions(Connection connection, Visitor visitor,
     TableName tableName) throws IOException {
-    scanMeta(connection, tableName, QueryType.REGION, Integer.MAX_VALUE, visitor);
+    scanMetaForTableRegions(connection, visitor, tableName, CatalogReplicaMode.NONE);
   }
 
   private static void scanMeta(Connection connection, TableName table, QueryType type, int maxRows,
-    final Visitor visitor) throws IOException {
+    final Visitor visitor, CatalogReplicaMode metaReplicaMode) throws IOException {
     scanMeta(connection, getTableStartRowForMeta(table, type), getTableStopRowForMeta(table, type),
-      type, maxRows, visitor);
+      type, null, maxRows, visitor, metaReplicaMode);
+
   }
 
   private static void scanMeta(Connection connection, @Nullable final byte[] startRow,
@@ -749,12 +756,12 @@ public static void scanMeta(Connection connection, final Visitor visitor,
   static void scanMeta(Connection connection, @Nullable final byte[] startRow,
     @Nullable final byte[] stopRow, QueryType type, int maxRows, final Visitor visitor)
     throws IOException {
-    scanMeta(connection, startRow, stopRow, type, null, maxRows, visitor);
+    scanMeta(connection, startRow, stopRow, type, null, maxRows, visitor, CatalogReplicaMode.NONE);
   }
 
   private static void scanMeta(Connection connection, @Nullable final byte[] startRow,
     @Nullable final byte[] stopRow, QueryType type, @Nullable Filter filter, int maxRows,
-    final Visitor visitor) throws IOException {
+    final Visitor visitor, CatalogReplicaMode metaReplicaMode) throws IOException {
     int rowUpperLimit = maxRows > 0 ? maxRows : Integer.MAX_VALUE;
     Scan scan = getMetaScan(connection.getConfiguration(), rowUpperLimit);
 
@@ -779,6 +786,25 @@ private static void scanMeta(Connection connection, @Nullable final byte[] start
 
     int currentRow = 0;
     try (Table metaTable = getMetaHTable(connection)) {
+      switch (metaReplicaMode) {
+        case LOAD_BALANCE:
+          int numOfReplicas = metaTable.getDescriptor().getRegionReplication();
+          if (numOfReplicas > 1) {
+            int replicaId = ThreadLocalRandom.current().nextInt(numOfReplicas);
+
+            // When the replicaId is 0, do not set to Consistency.TIMELINE
+            if (replicaId > 0) {
+              scan.setReplicaId(replicaId);
+              scan.setConsistency(Consistency.TIMELINE);
+            }
+          }
+          break;
+        case HEDGED_READ:
+          scan.setConsistency(Consistency.TIMELINE);
+          break;
+        default:
+          // Do nothing
+      }
       try (ResultScanner scanner = metaTable.getScanner(scan)) {
         Result data;
         while ((data = scanner.next()) != null) {
@@ -2056,7 +2082,7 @@ public static List<String> getTableEncodedRegionNamesForSerialReplication(Connec
       new FirstKeyOnlyFilter(), Integer.MAX_VALUE, r -> {
         list.add(RegionInfo.encodeRegionName(r.getRow()));
         return true;
-      });
+      }, CatalogReplicaMode.NONE);
     return list;
   }
 

hbase-client/src/main/java/org/apache/hadoop/hbase/client/AsyncNonMetaRegionLocator.java

@@ -49,6 +49,7 @@
 import java.util.concurrent.ConcurrentSkipListMap;
 import java.util.concurrent.TimeUnit;
 import org.apache.commons.lang3.ObjectUtils;
+import org.apache.hadoop.hbase.CatalogReplicaMode;
 import org.apache.hadoop.hbase.HBaseIOException;
 import org.apache.hadoop.hbase.HConstants;
 import org.apache.hadoop.hbase.HRegionLocation;

hbase-client/src/main/java/org/apache/hadoop/hbase/client/ConnectionImplementation.java

@@ -54,6 +54,7 @@
 import java.util.stream.Collectors;
 import org.apache.hadoop.conf.Configuration;
 import org.apache.hadoop.hbase.AuthUtil;
+import org.apache.hadoop.hbase.CatalogReplicaMode;
 import org.apache.hadoop.hbase.ChoreService;
 import org.apache.hadoop.hbase.DoNotRetryIOException;
 import org.apache.hadoop.hbase.HBaseServerException;

hbase-client/src/main/java/org/apache/hadoop/hbase/client/HRegionLocator.java

@@ -31,6 +31,7 @@
 import java.util.function.Function;
 import java.util.function.Supplier;
 import java.util.stream.Collectors;
+import org.apache.hadoop.hbase.CatalogReplicaMode;
 import org.apache.hadoop.hbase.HConstants;
 import org.apache.hadoop.hbase.HRegionLocation;
 import org.apache.hadoop.hbase.MetaTableAccessor;
@@ -143,7 +144,9 @@ public boolean visitInternal(Result result) throws IOException {
         return true;
       }
     };
-    MetaTableAccessor.scanMetaForTableRegions(connection, visitor, tableName);
+    CatalogReplicaMode metaReplicaMode = CatalogReplicaMode.fromString(connection.getConfiguration()
+      .get(LOCATOR_META_REPLICAS_MODE, CatalogReplicaMode.NONE.toString()));
+    MetaTableAccessor.scanMetaForTableRegions(connection, visitor, tableName, metaReplicaMode);
     return regions;
   }
 

hbase-server/src/test/java/org/apache/hadoop/hbase/client/TestAsyncNonMetaRegionLocator.java

@@ -37,6 +37,7 @@
 import java.util.concurrent.ThreadLocalRandom;
 import java.util.stream.IntStream;
 import org.apache.hadoop.conf.Configuration;
+import org.apache.hadoop.hbase.CatalogReplicaMode;
 import org.apache.hadoop.hbase.HBaseClassTestRule;
 import org.apache.hadoop.hbase.HBaseTestingUtility;
 import org.apache.hadoop.hbase.HRegionLocation;

hbase-server/src/test/java/org/apache/hadoop/hbase/replication/regionserver/TestMetaRegionReplicaReplicationEndpoint.java

@@ -516,11 +516,8 @@ private void primaryMayIncreaseReplicaNoChange(final long[] before, final long[]
   }
 
   private void primaryIncreaseReplicaIncrease(final long[] before, final long[] after) {
-    // There are read requests increase for primary meta replica.
-    assertTrue(after[RegionInfo.DEFAULT_REPLICA_ID] > before[RegionInfo.DEFAULT_REPLICA_ID]);
-
-    // There are read requests incrase for meta replica regions.
-    for (int i = 1; i < after.length; i++) {
+    // There are read requests increase for all meta replica regions,
+    for (int i = 0; i < after.length; i++) {
       assertTrue(after[i] > before[i]);
     }
   }
@@ -541,6 +538,7 @@ public void testHBaseMetaReplicaGets() throws Exception {
     final Region[] metaRegions = getAllRegions(TableName.META_TABLE_NAME, numOfMetaReplica);
     long[] readReqsForMetaReplicas = new long[numOfMetaReplica];
     long[] readReqsForMetaReplicasAfterGet = new long[numOfMetaReplica];
+    long[] readReqsForMetaReplicasAfterGetAllLocations = new long[numOfMetaReplica];
     long[] readReqsForMetaReplicasAfterMove = new long[numOfMetaReplica];
     long[] readReqsForMetaReplicasAfterSecondMove = new long[numOfMetaReplica];
     long[] readReqsForMetaReplicasAfterThirdGet = new long[numOfMetaReplica];
@@ -588,6 +586,16 @@ public void testHBaseMetaReplicaGets() throws Exception {
       // There are more reads against all meta replica regions, including the primary region.
       primaryIncreaseReplicaIncrease(readReqsForMetaReplicas, readReqsForMetaReplicasAfterGet);
 
+      RegionLocator locator = tableForGet.getRegionLocator();
+
+      for (int j = 0; j < numOfMetaReplica * 3; j++) {
+        locator.getAllRegionLocations();
+      }
+
+      getMetaReplicaReadRequests(metaRegions, readReqsForMetaReplicasAfterGetAllLocations);
+      primaryIncreaseReplicaIncrease(readReqsForMetaReplicasAfterGet,
+        readReqsForMetaReplicasAfterGetAllLocations);
+
       // move one of regions so it meta cache may be invalid.
       HTU.moveRegionAndWait(userRegion.getRegionInfo(), destRs.getServerName());
 
@@ -597,7 +605,7 @@ public void testHBaseMetaReplicaGets() throws Exception {
 
       // There are read requests increase for primary meta replica.
       // For rest of meta replicas, there is no change as regionMove will tell the new location
-      primaryIncreaseReplicaNoChange(readReqsForMetaReplicasAfterGet,
+      primaryIncreaseReplicaNoChange(readReqsForMetaReplicasAfterGetAllLocations,
         readReqsForMetaReplicasAfterMove);
       // Move region again.
       HTU.moveRegionAndWait(userRegion.getRegionInfo(), srcRs.getServerName());