From a6d0e9e56a7eef226c8a891cf741ab22d8205e1c Mon Sep 17 00:00:00 2001
From: Walter Su <waltersu4549@apache.org>
Date: Mon, 18 Apr 2016 09:28:02 +0800
Subject: [PATCH] HDFS-9412. getBlocks occupies FSLock and takes too long to
 complete. Contributed by He Tianyi.

(cherry picked from commit 67523ffcf491f4f2db5335899c00a174d0caaa9b)

 Conflicts:
	hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestGetBlocks.java
---
 .../server/blockmanagement/BlockManager.java  | 17 ++++++++++++
 .../org/apache/hadoop/hdfs/TestGetBlocks.java | 26 ++++++-------------
 2 files changed, 25 insertions(+), 18 deletions(-)

diff --git a/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java b/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java
index 4f68c28cc8b..ee9ca98d3ed 100644
--- a/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java
+++ b/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java
@@ -279,6 +279,14 @@ public int getPendingDataNodeMessageCount() {
    * processed again after aquiring lock again.
    */
   private int numBlocksPerIteration;
+
+  /**
+   * Minimum size that a block can be sent to Balancer through getBlocks.
+   * And after HDFS-8824, the small blocks are unused anyway, so there's no
+   * point to send them to balancer.
+   */
+  private long getBlocksMinBlockSize = -1;
+
   /**
    * Progress of the Replication queues initialisation.
    */
@@ -373,6 +381,9 @@ public BlockManager(final Namesystem namesystem, boolean haEnabled,
     this.numBlocksPerIteration = conf.getInt(
         DFSConfigKeys.DFS_BLOCK_MISREPLICATION_PROCESSING_LIMIT,
         DFSConfigKeys.DFS_BLOCK_MISREPLICATION_PROCESSING_LIMIT_DEFAULT);
+    this.getBlocksMinBlockSize = conf.getLongBytes(
+        DFSConfigKeys.DFS_BALANCER_GETBLOCKS_MIN_BLOCK_SIZE_KEY,
+        DFSConfigKeys.DFS_BALANCER_GETBLOCKS_MIN_BLOCK_SIZE_DEFAULT);
     this.blockReportLeaseManager = new BlockReportLeaseManager(conf);
 
     bmSafeMode = new BlockManagerSafeMode(this, namesystem, haEnabled, conf);
@@ -1068,6 +1079,9 @@ public BlocksWithLocations getBlocksWithLocations(final DatanodeID datanode,
     while(totalSize<size && iter.hasNext()) {
       curBlock = iter.next();
       if(!curBlock.isComplete())  continue;
+      if (curBlock.getNumBytes() < getBlocksMinBlockSize) {
+        continue;
+      }
       totalSize += addBlock(curBlock, results);
     }
     if(totalSize<size) {
@@ -1075,6 +1089,9 @@ public BlocksWithLocations getBlocksWithLocations(final DatanodeID datanode,
       for(int i=0; i<startBlock&&totalSize<size; i++) {
         curBlock = iter.next();
         if(!curBlock.isComplete())  continue;
+        if (curBlock.getNumBytes() < getBlocksMinBlockSize) {
+          continue;
+        }
         totalSize += addBlock(curBlock, results);
       }
     }
diff --git a/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestGetBlocks.java b/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestGetBlocks.java
index 9abf754acd3..b5164bff511 100644
--- a/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestGetBlocks.java
+++ b/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestGetBlocks.java
@@ -179,29 +179,19 @@ public void testGetBlocks() throws Exception {
 
     final short REPLICATION_FACTOR = (short) 2;
     final int DEFAULT_BLOCK_SIZE = 1024;
-    final Random r = new Random();
 
     CONF.setLong(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, DEFAULT_BLOCK_SIZE);
+    CONF.setLong(DFSConfigKeys.DFS_BALANCER_GETBLOCKS_MIN_BLOCK_SIZE_KEY,
+      DEFAULT_BLOCK_SIZE);
+
     MiniDFSCluster cluster = new MiniDFSCluster.Builder(CONF).numDataNodes(
         REPLICATION_FACTOR).build();
     try {
       cluster.waitActive();
-
-      // create a file with two blocks
-      FileSystem fs = cluster.getFileSystem();
-      FSDataOutputStream out = fs.create(new Path("/tmp.txt"),
-          REPLICATION_FACTOR);
-      byte[] data = new byte[1024];
-      long fileLen = 2 * DEFAULT_BLOCK_SIZE;
-      long bytesToWrite = fileLen;
-      while (bytesToWrite > 0) {
-        r.nextBytes(data);
-        int bytesToWriteNext = (1024 < bytesToWrite) ? 1024
-            : (int) bytesToWrite;
-        out.write(data, 0, bytesToWriteNext);
-        bytesToWrite -= bytesToWriteNext;
-      }
-      out.close();
+      // the third block will not be visible to getBlocks
+      long fileLen = 2 * DEFAULT_BLOCK_SIZE + 1;
+      DFSTestUtil.createFile(cluster.getFileSystem(), new Path("/tmp.txt"),
+          fileLen, REPLICATION_FACTOR, 0L);
 
       // get blocks & data nodes
       List<LocatedBlock> locatedBlocks;
@@ -212,7 +202,7 @@ public void testGetBlocks() throws Exception {
             DFSUtilClient.getNNAddress(CONF), CONF);
         locatedBlocks = dfsclient.getNamenode()
             .getBlockLocations("/tmp.txt", 0, fileLen).getLocatedBlocks();
-        assertEquals(2, locatedBlocks.size());
+        assertEquals(3, locatedBlocks.size());
         notWritten = false;
         for (int i = 0; i < 2; i++) {
           dataNodes = locatedBlocks.get(i).getLocations();