HDFS-17210. Optimize AvailableSpaceBlockPlacementPolicy. (#6113). Contributed by GuoPhilipse.

Reviewed-by:  He Xiaoqiao <hexiaoqiao@apache.org>
Signed-off-by: Shuyan Zhang <zhangshuyan@apache.org>
This commit is contained in:
GuoPhilipse 2023-10-16 16:34:40 +08:00 committed by GitHub
parent 00f8cdcb0f
commit c8abca3004
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 124 additions and 7 deletions

View File

@ -1241,6 +1241,12 @@ public class DFSConfigKeys extends CommonConfigurationKeys {
public static final int
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_DEFAULT =
5;
public static final String
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_LIMIT_KEY =
"dfs.namenode.available-space-block-placement-policy.balanced-space-tolerance-limit";
public static final int
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_LIMIT_DEFAULT =
100;
public static final String
DFS_NAMENODE_AVAILABLE_SPACE_RACK_FAULT_TOLERANT_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_PREFERENCE_FRACTION_KEY =
"dfs.namenode.available-space-rack-fault-tolerant-block-placement-policy"

View File

@ -22,6 +22,8 @@
import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_PREFERENCE_FRACTION_DEFAULT;
import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_DEFAULT;
import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_KEY;
import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_LIMIT_KEY;
import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_LIMIT_DEFAULT;
import java.util.Collection;
import java.util.EnumMap;
@ -50,7 +52,10 @@ public class AvailableSpaceBlockPlacementPolicy extends
private int balancedPreference =
(int) (100 * DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_PREFERENCE_FRACTION_DEFAULT);
private int balancedSpaceTolerance =
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_DEFAULT;
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_DEFAULT;
private int balancedSpaceToleranceLimit =
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_LIMIT_DEFAULT;
private boolean optimizeLocal;
@Override
@ -59,8 +64,8 @@ public void initialize(Configuration conf, FSClusterStats stats,
super.initialize(conf, stats, clusterMap, host2datanodeMap);
float balancedPreferencePercent =
conf.getFloat(
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_PREFERENCE_FRACTION_KEY,
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_PREFERENCE_FRACTION_DEFAULT);
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_PREFERENCE_FRACTION_KEY,
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_PREFERENCE_FRACTION_DEFAULT);
LOG.info("Available space block placement policy initialized: "
+ DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_PREFERENCE_FRACTION_KEY
@ -71,6 +76,11 @@ public void initialize(Configuration conf, FSClusterStats stats,
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_KEY,
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_DEFAULT);
balancedSpaceToleranceLimit =
conf.getInt(
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_LIMIT_KEY,
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_LIMIT_DEFAULT);
optimizeLocal = conf.getBoolean(
DFSConfigKeys.DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCE_LOCAL_NODE_KEY,
DFSConfigKeys.DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCE_LOCAL_NODE_DEFAULT);
@ -87,6 +97,17 @@ public void initialize(Configuration conf, FSClusterStats stats,
+ " receive more block allocations.");
}
if (balancedSpaceToleranceLimit > 100 || balancedSpaceToleranceLimit < 0) {
LOG.warn("The value of "
+ DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_LIMIT_KEY
+ " is invalid, Current value is " + balancedSpaceToleranceLimit + ", Default value "
+ DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_LIMIT_DEFAULT
+ " will be used instead.");
balancedSpaceToleranceLimit =
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_LIMIT_DEFAULT;
}
if (balancedSpaceTolerance > 20 || balancedSpaceTolerance < 0) {
LOG.warn("The value of "
+ DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_KEY
@ -201,8 +222,12 @@ private DatanodeDescriptor select(DatanodeDescriptor a, DatanodeDescriptor b,
*/
protected int compareDataNode(final DatanodeDescriptor a,
final DatanodeDescriptor b, boolean isBalanceLocal) {
boolean toleranceLimit = Math.max(a.getDfsUsedPercent(), b.getDfsUsedPercent())
< balancedSpaceToleranceLimit;
if (a.equals(b)
|| Math.abs(a.getDfsUsedPercent() - b.getDfsUsedPercent()) < balancedSpaceTolerance || ((
|| (toleranceLimit && Math.abs(a.getDfsUsedPercent() - b.getDfsUsedPercent())
< balancedSpaceTolerance) || ((
isBalanceLocal && a.getDfsUsedPercent() < 50))) {
return 0;
}

View File

@ -5173,6 +5173,17 @@
</description>
</property>
<property>
<name>dfs.namenode.available-space-block-placement-policy.balanced-space-tolerance-limit</name>
<value>100</value>
<description>
Only used when the dfs.block.replicator.classname is set to
org.apache.hadoop.hdfs.server.blockmanagement.AvailableSpaceBlockPlacementPolicy.
Special value between 0 and 100, inclusive. if the value is set beyond the scope,
this value will be set as 100 by default.
</description>
</property>
<property>
<name>
dfs.namenode.available-space-block-placement-policy.balance-local-node

View File

@ -58,9 +58,12 @@ public class TestAvailableSpaceBlockPlacementPolicy {
@BeforeClass
public static void setupCluster() throws Exception {
conf = new HdfsConfiguration();
conf.setFloat(
DFSConfigKeys.DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_PREFERENCE_FRACTION_KEY,
0.6f);
conf.setFloat(DFSConfigKeys.
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_PREFERENCE_FRACTION_KEY,
0.6f);
conf.setInt(DFSConfigKeys.
DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_TOLERANCE_LIMIT_KEY,
93);
String[] racks = new String[numRacks];
for (int i = 0; i < numRacks; i++) {
racks[i] = "/rack" + i;
@ -219,6 +222,78 @@ public void testChooseSimilarDataNode() {
tolerateDataNodes[0], false) == 1);
}
@Test
public void testCompareDataNode() {
DatanodeDescriptor[] tolerateDataNodes;
DatanodeStorageInfo[] tolerateStorages;
int capacity = 5;
Collection<Node> allTolerateNodes = new ArrayList<>(capacity);
String[] ownerRackOfTolerateNodes = new String[capacity];
for (int i = 0; i < capacity; i++) {
ownerRackOfTolerateNodes[i] = "rack"+i;
}
tolerateStorages = DFSTestUtil.createDatanodeStorageInfos(ownerRackOfTolerateNodes);
tolerateDataNodes = DFSTestUtil.toDatanodeDescriptor(tolerateStorages);
Collections.addAll(allTolerateNodes, tolerateDataNodes);
final BlockManager bm = namenode.getNamesystem().getBlockManager();
AvailableSpaceBlockPlacementPolicy toleratePlacementPolicy =
(AvailableSpaceBlockPlacementPolicy)bm.getBlockPlacementPolicy();
//96.6%
updateHeartbeatWithUsage(tolerateDataNodes[0],
30 * HdfsServerConstants.MIN_BLOCKS_FOR_WRITE * blockSize,
29 * HdfsServerConstants.MIN_BLOCKS_FOR_WRITE * blockSize,
HdfsServerConstants.MIN_BLOCKS_FOR_WRITE
* blockSize, 0L, 0L, 0L, 0, 0);
//93.3%
updateHeartbeatWithUsage(tolerateDataNodes[1],
30 * HdfsServerConstants.MIN_BLOCKS_FOR_WRITE * blockSize,
28 * HdfsServerConstants.MIN_BLOCKS_FOR_WRITE * blockSize,
HdfsServerConstants.MIN_BLOCKS_FOR_WRITE
* blockSize, 0L, 0L, 0L, 0, 0);
//90.0%
updateHeartbeatWithUsage(tolerateDataNodes[2],
30 * HdfsServerConstants.MIN_BLOCKS_FOR_WRITE * blockSize,
27 * HdfsServerConstants.MIN_BLOCKS_FOR_WRITE * blockSize,
HdfsServerConstants.MIN_BLOCKS_FOR_WRITE
* blockSize, 0L, 0L, 0L, 0, 0);
//86.6%
updateHeartbeatWithUsage(tolerateDataNodes[3],
30 * HdfsServerConstants.MIN_BLOCKS_FOR_WRITE * blockSize,
26 * HdfsServerConstants.MIN_BLOCKS_FOR_WRITE * blockSize,
HdfsServerConstants.MIN_BLOCKS_FOR_WRITE
* blockSize, 0L, 0L, 0L, 0, 0);
//83.3%
updateHeartbeatWithUsage(tolerateDataNodes[4],
30 * HdfsServerConstants.MIN_BLOCKS_FOR_WRITE * blockSize,
25 * HdfsServerConstants.MIN_BLOCKS_FOR_WRITE * blockSize,
HdfsServerConstants.MIN_BLOCKS_FOR_WRITE
* blockSize, 0L, 0L, 0L, 0, 0);
assertTrue(toleratePlacementPolicy.compareDataNode(tolerateDataNodes[0],
tolerateDataNodes[1], false) == 1);
assertTrue(toleratePlacementPolicy.compareDataNode(tolerateDataNodes[1],
tolerateDataNodes[0], false) == -1);
assertTrue(toleratePlacementPolicy.compareDataNode(tolerateDataNodes[1],
tolerateDataNodes[2], false) == 1);
assertTrue(toleratePlacementPolicy.compareDataNode(tolerateDataNodes[2],
tolerateDataNodes[1], false) == -1);
assertTrue(toleratePlacementPolicy.compareDataNode(tolerateDataNodes[2],
tolerateDataNodes[3], false) == 0);
assertTrue(toleratePlacementPolicy.compareDataNode(tolerateDataNodes[3],
tolerateDataNodes[2], false) == 0);
assertTrue(toleratePlacementPolicy.compareDataNode(tolerateDataNodes[2],
tolerateDataNodes[4], false) == 1);
assertTrue(toleratePlacementPolicy.compareDataNode(tolerateDataNodes[4],
tolerateDataNodes[2], false) == -1);
}
@AfterClass
public static void teardownCluster() {
if (namenode != null) {