HDFS-7225. Remove stale block invalidation work when DN re-registers with different UUID. (Zhe Zhang and Andrew Wang)

This commit is contained in:
Andrew Wang 2014-11-18 22:14:04 -08:00
parent 79301e80d7
commit 406c09ad11
4 changed files with 165 additions and 46 deletions

View File

@ -453,6 +453,9 @@ Release 2.7.0 - UNRELEASED
HDFS-7406. SimpleHttpProxyHandler puts incorrect "Connection: Close"
header. (wheat9)
HDFS-7225. Remove stale block invalidation work when DN re-registers with
different UUID. (Zhe Zhang and Andrew Wang)
Release 2.6.0 - 2014-11-18
INCOMPATIBLE CHANGES

View File

@ -1111,6 +1111,18 @@ private void addToInvalidates(Block b) {
}
}
/**
* Remove all block invalidation tasks under this datanode UUID;
* used when a datanode registers with a new UUID and the old one
* is wiped.
*/
void removeFromInvalidates(final DatanodeInfo datanode) {
if (!namesystem.isPopulatingReplQueues()) {
return;
}
invalidateBlocks.remove(datanode);
}
/**
* Mark the block belonging to datanode as corrupt
* @param blk Block to be marked as corrupt
@ -3382,7 +3394,14 @@ private int invalidateWorkForOneNode(DatanodeInfo dn) {
return 0;
}
try {
toInvalidate = invalidateBlocks.invalidateWork(datanodeManager.getDatanode(dn));
DatanodeDescriptor dnDescriptor = datanodeManager.getDatanode(dn);
if (dnDescriptor == null) {
LOG.warn("DataNode " + dn + " cannot be found with UUID " +
dn.getDatanodeUuid() + ", removing block invalidation work.");
invalidateBlocks.remove(dn);
return 0;
}
toInvalidate = invalidateBlocks.invalidateWork(dnDescriptor);
if (toInvalidate == null) {
return 0;

View File

@ -600,6 +600,8 @@ private void wipeDatanode(final DatanodeID node) {
synchronized (datanodeMap) {
host2DatanodeMap.remove(datanodeMap.remove(key));
}
// Also remove all block invalidation tasks under this node
blockManager.removeFromInvalidates(new DatanodeInfo(node));
if (LOG.isDebugEnabled()) {
LOG.debug(getClass().getSimpleName() + ".wipeDatanode("
+ node + "): storage " + key

View File

@ -17,66 +17,161 @@
*/
package org.apache.hadoop.hdfs.server.blockmanagement;
import java.util.UUID;
import static org.junit.Assert.assertEquals;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.hdfs.DFSTestUtil;
import org.apache.hadoop.hdfs.DistributedFileSystem;
import org.apache.hadoop.hdfs.HdfsConfiguration;
import org.apache.hadoop.hdfs.MiniDFSCluster;
import org.apache.hadoop.hdfs.protocol.Block;
import org.apache.hadoop.hdfs.protocol.DatanodeID;
import org.apache.hadoop.hdfs.security.token.block.ExportedBlockKeys;
import org.apache.hadoop.hdfs.server.common.GenerationStamp;
import org.apache.hadoop.hdfs.server.common.HdfsServerConstants;
import org.apache.hadoop.hdfs.server.common.StorageInfo;
import org.apache.hadoop.hdfs.server.datanode.DataNode;
import org.apache.hadoop.hdfs.server.namenode.FSNamesystem;
import org.apache.hadoop.hdfs.server.protocol.DatanodeRegistration;
import org.apache.hadoop.util.VersionInfo;
import org.junit.After;
import org.junit.Assert;
import org.junit.Before;
import org.junit.Test;
import org.mockito.internal.util.reflection.Whitebox;
/**
* Test if FSNamesystem handles heartbeat right
*/
public class TestComputeInvalidateWork {
/**
* Test if {@link FSNamesystem#computeInvalidateWork(int)}
* can schedule invalidate work correctly
*/
@Test
public void testCompInvalidate() throws Exception {
final Configuration conf = new HdfsConfiguration();
final int NUM_OF_DATANODES = 3;
final MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).numDataNodes(NUM_OF_DATANODES).build();
try {
cluster.waitActive();
final FSNamesystem namesystem = cluster.getNamesystem();
final BlockManager bm = namesystem.getBlockManager();
final int blockInvalidateLimit = bm.getDatanodeManager().blockInvalidateLimit;
final DatanodeDescriptor[] nodes = bm.getDatanodeManager(
).getHeartbeatManager().getDatanodes();
assertEquals(nodes.length, NUM_OF_DATANODES);
namesystem.writeLock();
try {
for (int i=0; i<nodes.length; i++) {
for(int j=0; j<3*blockInvalidateLimit+1; j++) {
Block block = new Block(i*(blockInvalidateLimit+1)+j, 0,
GenerationStamp.LAST_RESERVED_STAMP);
bm.addToInvalidates(block, nodes[i]);
}
}
assertEquals(blockInvalidateLimit*NUM_OF_DATANODES,
bm.computeInvalidateWork(NUM_OF_DATANODES+1));
assertEquals(blockInvalidateLimit*NUM_OF_DATANODES,
bm.computeInvalidateWork(NUM_OF_DATANODES));
assertEquals(blockInvalidateLimit*(NUM_OF_DATANODES-1),
bm.computeInvalidateWork(NUM_OF_DATANODES-1));
int workCount = bm.computeInvalidateWork(1);
if (workCount == 1) {
assertEquals(blockInvalidateLimit+1, bm.computeInvalidateWork(2));
} else {
assertEquals(workCount, blockInvalidateLimit);
assertEquals(2, bm.computeInvalidateWork(2));
}
} finally {
namesystem.writeUnlock();
}
} finally {
private Configuration conf;
private final int NUM_OF_DATANODES = 3;
private MiniDFSCluster cluster;
private FSNamesystem namesystem;
private BlockManager bm;
private DatanodeDescriptor[] nodes;
@Before
public void setup() throws Exception {
conf = new HdfsConfiguration();
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(NUM_OF_DATANODES)
.build();
cluster.waitActive();
namesystem = cluster.getNamesystem();
bm = namesystem.getBlockManager();
nodes = bm.getDatanodeManager().getHeartbeatManager().getDatanodes();
assertEquals(nodes.length, NUM_OF_DATANODES);
}
@After
public void teardown() throws Exception {
if (cluster != null) {
cluster.shutdown();
}
}
/**
* Test if {@link BlockManager#computeInvalidateWork(int)}
* can schedule invalidate work correctly
*/
@Test(timeout=120000)
public void testCompInvalidate() throws Exception {
final int blockInvalidateLimit = bm.getDatanodeManager()
.blockInvalidateLimit;
namesystem.writeLock();
try {
for (int i=0; i<nodes.length; i++) {
for(int j=0; j<3*blockInvalidateLimit+1; j++) {
Block block = new Block(i*(blockInvalidateLimit+1)+j, 0,
GenerationStamp.LAST_RESERVED_STAMP);
bm.addToInvalidates(block, nodes[i]);
}
}
assertEquals(blockInvalidateLimit*NUM_OF_DATANODES,
bm.computeInvalidateWork(NUM_OF_DATANODES+1));
assertEquals(blockInvalidateLimit*NUM_OF_DATANODES,
bm.computeInvalidateWork(NUM_OF_DATANODES));
assertEquals(blockInvalidateLimit*(NUM_OF_DATANODES-1),
bm.computeInvalidateWork(NUM_OF_DATANODES-1));
int workCount = bm.computeInvalidateWork(1);
if (workCount == 1) {
assertEquals(blockInvalidateLimit+1, bm.computeInvalidateWork(2));
} else {
assertEquals(workCount, blockInvalidateLimit);
assertEquals(2, bm.computeInvalidateWork(2));
}
} finally {
namesystem.writeUnlock();
}
}
/**
* Reformatted DataNodes will replace the original UUID in the
* {@link DatanodeManager#datanodeMap}. This tests if block
* invalidation work on the original DataNode can be skipped.
*/
@Test(timeout=120000)
public void testDatanodeReformat() throws Exception {
namesystem.writeLock();
try {
Block block = new Block(0, 0, GenerationStamp.LAST_RESERVED_STAMP);
bm.addToInvalidates(block, nodes[0]);
// Change the datanode UUID to emulate a reformation
nodes[0].setDatanodeUuidForTesting("fortesting");
// Since UUID has changed, the invalidation work should be skipped
assertEquals(0, bm.computeInvalidateWork(1));
assertEquals(0, bm.getPendingDeletionBlocksCount());
} finally {
namesystem.writeUnlock();
}
}
@Test(timeout=12000)
public void testDatanodeReRegistration() throws Exception {
// Create a test file
final DistributedFileSystem dfs = cluster.getFileSystem();
final Path path = new Path("/testRR");
// Create a file and shutdown the DNs, which populates InvalidateBlocks
DFSTestUtil.createFile(dfs, path, dfs.getDefaultBlockSize(),
(short) NUM_OF_DATANODES, 0xED0ED0);
for (DataNode dn : cluster.getDataNodes()) {
dn.shutdown();
}
dfs.delete(path, false);
namesystem.writeLock();
InvalidateBlocks invalidateBlocks;
int expected = NUM_OF_DATANODES;
try {
invalidateBlocks = (InvalidateBlocks) Whitebox
.getInternalState(cluster.getNamesystem().getBlockManager(),
"invalidateBlocks");
assertEquals("Expected invalidate blocks to be the number of DNs",
(long) expected, invalidateBlocks.numBlocks());
} finally {
namesystem.writeUnlock();
}
// Re-register each DN and see that it wipes the invalidation work
for (DataNode dn : cluster.getDataNodes()) {
DatanodeID did = dn.getDatanodeId();
did.setDatanodeUuidForTesting(UUID.randomUUID().toString());
DatanodeRegistration reg = new DatanodeRegistration(did,
new StorageInfo(HdfsServerConstants.NodeType.DATA_NODE),
new ExportedBlockKeys(),
VersionInfo.getVersion());
namesystem.writeLock();
try {
bm.getDatanodeManager().registerDatanode(reg);
expected--;
assertEquals("Expected number of invalidate blocks to decrease",
(long) expected, invalidateBlocks.numBlocks());
} finally {
namesystem.writeUnlock();
}
}
}
}