HDFS-9313. Possible NullPointerException in BlockManager if no excess replica can be chosen. (mingma)

This commit is contained in:
Ming Ma 2015-11-02 19:36:37 -08:00
parent 8e05dbf2bd
commit d565480da2
4 changed files with 45 additions and 3 deletions

View File

@ -2216,6 +2216,9 @@ Release 2.8.0 - UNRELEASED
HDFS-9329. TestBootstrapStandby#testRateThrottling is flaky because fsimage
size is smaller than IO buffer size. (zhz)
HDFS-9313. Possible NullPointerException in BlockManager if no excess
replica can be chosen. (mingma)
Release 2.7.2 - UNRELEASED
INCOMPATIBLE CHANGES

View File

@ -23,8 +23,6 @@
import java.util.Map;
import java.util.Set;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.apache.hadoop.classification.InterfaceAudience;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.StorageType;
@ -33,13 +31,17 @@
import org.apache.hadoop.net.NetworkTopology;
import org.apache.hadoop.net.Node;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
/**
* This interface is used for choosing the desired number of targets
* for placing block replicas.
*/
@InterfaceAudience.Private
public abstract class BlockPlacementPolicy {
static final Log LOG = LogFactory.getLog(BlockPlacementPolicy.class);
static final Logger LOG = LoggerFactory.getLogger(
BlockPlacementPolicy.class);
@InterfaceAudience.Private
public static class NotEnoughReplicasException extends Exception {

View File

@ -981,6 +981,12 @@ public List<DatanodeStorageInfo> chooseReplicasToDelete(
excessTypes);
}
firstOne = false;
if (cur == null) {
LOG.warn("No excess replica can be found. excessTypes: {}." +
" moreThanOne: {}. exactlyOne: {}.", excessTypes, moreThanOne,
exactlyOne);
break;
}
// adjust rackmap, moreThanOne, and exactlyOne
adjustSetsWithChosenReplica(rackMap, moreThanOne, exactlyOne, cur);

View File

@ -1000,6 +1000,14 @@ public void testChooseReplicasToDelete() throws Exception {
BlockStoragePolicySuite POLICY_SUITE = BlockStoragePolicySuite
.createDefaultSuite();
BlockStoragePolicy storagePolicy = POLICY_SUITE.getDefaultPolicy();
DatanodeStorageInfo excessSSD = DFSTestUtil.createDatanodeStorageInfo(
"Storage-excess-SSD-ID", "localhost",
storages[0].getDatanodeDescriptor().getNetworkLocation(),
"foo.com", StorageType.SSD);
updateHeartbeatWithUsage(excessSSD.getDatanodeDescriptor(),
2* HdfsServerConstants.MIN_BLOCKS_FOR_WRITE*BLOCK_SIZE, 0L,
2* HdfsServerConstants.MIN_BLOCKS_FOR_WRITE*BLOCK_SIZE, 0L, 0L, 0L, 0,
0);
// use delete hint case.
@ -1022,6 +1030,29 @@ public void testChooseReplicasToDelete() throws Exception {
excessReplicas = replicator.chooseReplicasToDelete(nonExcess, 3,
excessTypes, storages[3].getDatanodeDescriptor(), null);
assertTrue(excessReplicas.contains(excessStorage));
// The block was initially created on excessSSD(rack r1),
// storages[4](rack r3) and storages[5](rack r3) with
// ONESSD_STORAGE_POLICY_NAME storage policy.
// Right after balancer moves the block from storages[5] to
// storages[3](rack r2), the application changes the storage policy from
// ONESSD_STORAGE_POLICY_NAME to HOT_STORAGE_POLICY_ID. In this case,
// no replica can be chosen as the excessive replica as
// chooseReplicasToDelete only considers storages[4] and storages[5] that
// are the same rack. But neither's storage type is SSD.
// TODO BlockPlacementPolicyDefault should be able to delete excessSSD.
nonExcess.clear();
nonExcess.add(excessSSD);
nonExcess.add(storages[3]);
nonExcess.add(storages[4]);
nonExcess.add(storages[5]);
excessTypes = storagePolicy.chooseExcess((short) 3,
DatanodeStorageInfo.toStorageTypes(nonExcess));
excessReplicas = replicator.chooseReplicasToDelete(nonExcess, 3,
excessTypes, storages[3].getDatanodeDescriptor(),
storages[5].getDatanodeDescriptor());
assertTrue(excessReplicas.size() == 0);
}
@Test