You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@hbase.apache.org by te...@apache.org on 2011/07/12 07:15:58 UTC

svn commit: r1145459 - in /hbase/trunk: CHANGES.txt src/main/java/org/apache/hadoop/hbase/mapreduce/LoadIncrementalHFiles.java

Author: tedyu
Date: Tue Jul 12 05:15:58 2011
New Revision: 1145459

URL: http://svn.apache.org/viewvc?rev=1145459&view=rev
Log:
HBASE-3871  Speedup LoadIncrementalHFiles by parallelizing HFile splitting

Modified:
    hbase/trunk/CHANGES.txt
    hbase/trunk/src/main/java/org/apache/hadoop/hbase/mapreduce/LoadIncrementalHFiles.java

Modified: hbase/trunk/CHANGES.txt
URL: http://svn.apache.org/viewvc/hbase/trunk/CHANGES.txt?rev=1145459&r1=1145458&r2=1145459&view=diff
==============================================================================
--- hbase/trunk/CHANGES.txt (original)
+++ hbase/trunk/CHANGES.txt Tue Jul 12 05:15:58 2011
@@ -313,6 +313,7 @@ Release 0.91.0 - Unreleased
                (Aaron T. Myers via todd)
    HBASE-4054  Usability improvement to HTablePool (Daniel Iancu)
    HBASE-4079  HTableUtil - helper class for loading data (Doug Meil via Ted Yu)
+   HBASE-3871  Speedup LoadIncrementalHFiles by parallelizing HFile splitting
 
   TASKS
    HBASE-3559  Move report of split to master OFF the heartbeat channel

Modified: hbase/trunk/src/main/java/org/apache/hadoop/hbase/mapreduce/LoadIncrementalHFiles.java
URL: http://svn.apache.org/viewvc/hbase/trunk/src/main/java/org/apache/hadoop/hbase/mapreduce/LoadIncrementalHFiles.java?rev=1145459&r1=1145458&r2=1145459&view=diff
==============================================================================
--- hbase/trunk/src/main/java/org/apache/hadoop/hbase/mapreduce/LoadIncrementalHFiles.java (original)
+++ hbase/trunk/src/main/java/org/apache/hadoop/hbase/mapreduce/LoadIncrementalHFiles.java Tue Jul 12 05:15:58 2011
@@ -25,6 +25,7 @@ import java.util.ArrayList;
 import java.util.Arrays;
 import java.util.Deque;
 import java.util.HashSet;
+import java.util.Iterator;
 import java.util.LinkedList;
 import java.util.Map;
 import java.util.Set;
@@ -36,6 +37,7 @@ import java.util.concurrent.Future;
 import java.util.concurrent.LinkedBlockingQueue;
 import java.util.concurrent.ThreadPoolExecutor;
 import java.util.concurrent.TimeUnit;
+import java.util.concurrent.TimeoutException;
 import java.util.concurrent.atomic.AtomicLong;
 
 import org.apache.commons.logging.Log;
@@ -85,6 +87,7 @@ public class LoadIncrementalHFiles exten
   private HBaseAdmin hbAdmin;
   private Configuration cfg;
   private Set<Future> futures = new HashSet<Future>();
+  private Set<Future> futuresForSplittingHFile = new HashSet<Future>();
 
   public static String NAME = "completebulkload";
 
@@ -188,13 +191,30 @@ public class LoadIncrementalHFiles exten
     try {
       queue = discoverLoadQueue(hfofDir);
       // outer loop picks up LoadQueueItem due to HFile split
-      while (!queue.isEmpty()) {
+      while (!queue.isEmpty() || futuresForSplittingHFile.size() > 0) {
         Pair<byte[][],byte[][]> startEndKeys = table.getStartEndKeys();
         // inner loop groups callables
         while (!queue.isEmpty()) {
           LoadQueueItem item = queue.remove();
           tryLoad(item, conn, table, queue, startEndKeys, pool);
         }
+        Iterator<Future> iter = futuresForSplittingHFile.iterator();
+        while (iter.hasNext()) {
+          boolean timeoutSeen = false;
+          Future future = iter.next();
+          try {
+            future.get(20, TimeUnit.MILLISECONDS);
+            break;  // we have at least two new HFiles to process
+          } catch (ExecutionException ee) {
+            LOG.error(ee);
+          } catch (InterruptedException ie) {
+            LOG.error(ie);
+          } catch (TimeoutException te) {
+              timeoutSeen = true;
+          } finally {
+            if (!timeoutSeen) iter.remove();
+          }
+        }
       }
       for (Future<Void> future : futures) {
         try {
@@ -247,8 +267,10 @@ public class LoadIncrementalHFiles exten
 
     // Add these back at the *front* of the queue, so there's a lower
     // chance that the region will just split again before we get there.
-    queue.addFirst(new LoadQueueItem(item.family, botOut));
-    queue.addFirst(new LoadQueueItem(item.family, topOut));
+    synchronized (queue) {
+      queue.addFirst(new LoadQueueItem(item.family, botOut));
+      queue.addFirst(new LoadQueueItem(item.family, topOut));
+    }
     LOG.info("Successfully split into new HFiles " + botOut + " and " + topOut);
   }
 
@@ -260,7 +282,8 @@ public class LoadIncrementalHFiles exten
    */
   private boolean tryLoad(final LoadQueueItem item,
       final HConnection conn, final HTable table,
-      final Deque<LoadQueueItem> queue, Pair<byte[][],byte[][]> startEndKeys,
+      final Deque<LoadQueueItem> queue,
+      final Pair<byte[][],byte[][]> startEndKeys,
       ExecutorService pool)
   throws IOException {
     final Path hfilePath = item.hfilePath;
@@ -292,12 +315,21 @@ public class LoadIncrementalHFiles exten
     if (idx < 0) {
       idx = -(idx+1)-1;
     }
+    final int indexForCallable = idx;
     boolean lastKeyInRange =
       Bytes.compareTo(last, startEndKeys.getSecond()[idx]) < 0 ||
       Bytes.equals(startEndKeys.getSecond()[idx], HConstants.EMPTY_BYTE_ARRAY);
     if (!lastKeyInRange) {
-      splitStoreFileAndRequeue(item, queue, table,
-          startEndKeys.getFirst()[idx], startEndKeys.getSecond()[idx]);
+      Callable<Void> callable = new Callable<Void>() {
+        public Void call() throws Exception {
+          splitStoreFileAndRequeue(item, queue, table,
+              startEndKeys.getFirst()[indexForCallable],
+              startEndKeys.getSecond()[indexForCallable]);
+          return (Void)null;
+        }
+      };
+      futuresForSplittingHFile.add(pool.submit(callable));
+
       return true;
     }