Apache
/
hadoop
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

HDFS-16182.numOfReplicas is given the wrong value in BlockPlacementPolicyDefault$chooseTarget can cause DataStreamer to fail with Heterogeneous Storage. (#3320)

This commit is contained in:
Neil 2021-11-16 18:21:50 +08:00 committed by GitHub
parent c9f95b01ef
commit 5626734a36
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 60 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockPlacementPolicyDefault.java
View File

@ -435,7 +435,7 @@ public class BlockPlacementPolicyDefault extends BlockPlacementPolicy {
* @param storageTypes storage type to be considered for target
* @return local node of writer (not chosen node)
*/
private Node chooseTarget(int numOfReplicas,
private Node chooseTarget(final int numOfReplicas,
Node writer,
final Set<Node> excludedNodes,
final long blocksize,
@ -469,7 +469,7 @@ public class BlockPlacementPolicyDefault extends BlockPlacementPolicy {
LOG.trace("storageTypes={}", storageTypes);
try {
if ((numOfReplicas = requiredStorageTypes.size()) == 0) {
if (requiredStorageTypes.size() == 0) {
throw new NotEnoughReplicasException(
"All required storage types are unavailable: "
+ " unavailableStorages=" + unavailableStorages
@ -498,10 +498,10 @@ public class BlockPlacementPolicyDefault extends BlockPlacementPolicy {
for (DatanodeStorageInfo resultStorage : results) {
addToExcludedNodes(resultStorage.getDatanodeDescriptor(), oldExcludedNodes);
}
// Set numOfReplicas, since it can get out of sync with the result list
// Set newNumOfReplicas, since it can get out of sync with the result list
// if the NotEnoughReplicasException was thrown in chooseRandom().
numOfReplicas = totalReplicasExpected - results.size();
return chooseTarget(numOfReplicas, writer, oldExcludedNodes, blocksize,
int newNumOfReplicas = totalReplicasExpected - results.size();
return chooseTarget(newNumOfReplicas, writer, oldExcludedNodes, blocksize,
maxNodesPerRack, results, false, storagePolicy, unavailableStorages,
newBlock, null);
}
@ -520,8 +520,8 @@ public class BlockPlacementPolicyDefault extends BlockPlacementPolicy {
addToExcludedNodes(resultStorage.getDatanodeDescriptor(),
oldExcludedNodes);
}
numOfReplicas = totalReplicasExpected - results.size();
return chooseTarget(numOfReplicas, writer, oldExcludedNodes, blocksize,
int newNumOfReplicas = totalReplicasExpected - results.size();
return chooseTarget(newNumOfReplicas, writer, oldExcludedNodes, blocksize,
maxNodesPerRack, results, false, storagePolicy, unavailableStorages,
newBlock, null);
}

53
hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestBlockStoragePolicy.java
View File

@ -1337,6 +1337,59 @@ public class TestBlockStoragePolicy {
Assert.assertEquals(StorageType.DISK, targets[1].getStorageType());
}
/**
* Consider a File with All_SSD storage policy.
* 1. Choose 3 DISK DNs for pipeline because SSD DNs no enough at
* the beginning.
* 2. One of DISK DNs fails And it need choose one new DN for existing.
* pipeline {@link DataStreamer addDatanode2ExistingPipeline()}.
* Make sure the number of target DNs are 3.
* see HDFS-16182.
*/
@Test
public void testAddDatanode2ExistingPipelineInSsd() throws Exception {
BlockStoragePolicy policy = POLICY_SUITE.getPolicy(ALLSSD);
final String[] racks = {"/d1/r1", "/d2/r2", "/d3/r3", "/d4/r4", "/d5/r5",
"/d6/r6", "/d7/r7"};
final String[] hosts = {"host1", "host2", "host3", "host4", "host5",
"host6", "host7"};
final StorageType[] disks = {StorageType.DISK, StorageType.DISK, StorageType.DISK};
final DatanodeStorageInfo[] diskStorages
= DFSTestUtil.createDatanodeStorageInfos(7, racks, hosts, disks);
final DatanodeDescriptor[] dataNodes
= DFSTestUtil.toDatanodeDescriptor(diskStorages);
for (int i = 0; i < dataNodes.length; i++) {
BlockManagerTestUtil.updateStorage(dataNodes[i],
new DatanodeStorage("ssd" + i + 1, DatanodeStorage.State.NORMAL,
StorageType.SSD));
}
FileSystem.setDefaultUri(conf, "hdfs://localhost:0");
conf.set(DFSConfigKeys.DFS_NAMENODE_HTTP_ADDRESS_KEY, "0.0.0.0:0");
File baseDir = PathUtils.getTestDir(TestReplicationPolicy.class);
conf.set(DFSConfigKeys.DFS_NAMENODE_NAME_DIR_KEY,
new File(baseDir, "name").getPath());
DFSTestUtil.formatNameNode(conf);
NameNode namenode = new NameNode(conf);
final BlockManager bm = namenode.getNamesystem().getBlockManager();
BlockPlacementPolicy replicator = bm.getBlockPlacementPolicy();
NetworkTopology cluster = bm.getDatanodeManager().getNetworkTopology();
for (DatanodeDescriptor datanode : dataNodes) {
cluster.add(datanode);
}
// chsenDs are DISK StorageType to simulate not enough SDD Storage
List<DatanodeStorageInfo> chsenDs = new ArrayList<>();
chsenDs.add(diskStorages[0]);
chsenDs.add(diskStorages[1]);
DatanodeStorageInfo[] targets = replicator.chooseTarget("/foo", 1,
null, chsenDs, true,
new HashSet<Node>(), 0, policy, null);
Assert.assertEquals(3, targets.length);
}
@Test
public void testGetFileStoragePolicyAfterRestartNN() throws Exception {
//HDFS8219