HDFS-9412. getBlocks occupies FSLock and takes too long to complete. Contributed by He Tianyi. Backport HDFS-11855 by Vinitha Reddy Gankidi.

(cherry picked from commit 67523ffcf491f4f2db5335899c00a174d0caaa9b)
(cherry picked from commit a6d0e9e56a7eef226c8a891cf741ab22d8205e1c)

hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt

@@ -105,6 +105,9 @@ Release 2.7.4 - UNRELEASED
     HDFS-11384. Balancer disperses getBlocks calls to avoid NameNode's rpc queue
     saturation. (shv)
 
+    HDFS-9412. getBlocks occupies FSLock and takes too long to complete.
+    Contributed by He Tianyi. Backport HDFS-11855 by Vinitha Reddy Gankidi.
+
   BUG FIXES
    
     HDFS-8307. Spurious DNS Queries from hdfs shell. (Andres Perez via aengineer)

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java

@@ -260,6 +260,14 @@ public class BlockManager {
    * 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.
    */
@@ -350,7 +358,10 @@ public class BlockManager {
     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);
+
     LOG.info("defaultReplication         = " + defaultReplication);
     LOG.info("maxReplication             = " + maxReplication);
     LOG.info("minReplication             = " + minReplication);
@@ -1069,6 +1080,9 @@ public class BlockManager {
     while(totalSize<size && iter.hasNext()) {
       curBlock = iter.next();
       if(!curBlock.isComplete())  continue;
+      if (curBlock.getNumBytes() < getBlocksMinBlockSize) {
+        continue;
+      }
       totalSize += addBlock(curBlock, results);
     }
     if(totalSize<size) {
@@ -1076,6 +1090,9 @@ public class BlockManager {
       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);
       }
     }

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestGetBlocks.java

@@ -188,6 +188,9 @@ public class TestGetBlocks {
     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)
               .storagesPerDatanode(4)
@@ -200,7 +203,7 @@ public class TestGetBlocks {
       FSDataOutputStream out = fs.create(new Path("/tmp.txt"),
           REPLICATION_FACTOR);
       byte[] data = new byte[1024];
-      long fileLen = 12 * DEFAULT_BLOCK_SIZE;
+      long fileLen = 12 * DEFAULT_BLOCK_SIZE + 1;
       long bytesToWrite = fileLen;
       while (bytesToWrite > 0) {
         r.nextBytes(data);
@@ -220,7 +223,8 @@ public class TestGetBlocks {
       do {
         locatedBlocks = dfsclient.getNamenode()
             .getBlockLocations("/tmp.txt", 0, fileLen).getLocatedBlocks();
-        assertEquals(12, locatedBlocks.size());
+        assertEquals(13, locatedBlocks.size());
+
         notWritten = false;
         for (int i = 0; i < 2; i++) {
           dataNodes = locatedBlocks.get(i).getLocations();