[
https://issues.apache.org/jira/browse/HDFS-13915?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jiandan Yang updated HDFS-13915:
---------------------------------
Description:
Consider following situation:
1. create a file with ALLSSD policy
2. return [SSD,SSD,DISK] due to lack of SSD space
3. client call NameNodeRpcServer#getAdditionalDatanode when recovering write
pipeline and replacing bad datanode
4. BlockPlacementPolicyDefault#chooseTarget will call
StoragePolicy#chooseStorageTypes(3, [SSD,DISK], none, false), but
chooseStorageTypes return [SSD,SSD]
{code:java}
@Test
public void testAllSSDFallbackAndNonNewBlock() {
final BlockStoragePolicy allSSD = POLICY_SUITE.getPolicy(ALLSSD);
List<StorageType> storageTypes = allSSD.chooseStorageTypes((short) 3,
Arrays.asList(StorageType.DISK, StorageType.SSD),
EnumSet.noneOf(StorageType.class), false);
assertEquals(2, storageTypes.size());
assertEquals(StorageType.SSD, storageTypes.get(0));
assertEquals(StorageType.SSD, storageTypes.get(1));
}
{code}
5. do numOfReplicas = requiredStorageTypes.size() and numOfReplicas is set to 2
and choose additional two datanodes
6. BlockPlacementPolicyDefault#chooseTarget return four datanodes to client
7. DataStreamer#findNewDatanode find nodes.length != original.length + 1 and
throw IOException, and finally lead to write failed
{code:java}
private int findNewDatanode(final DatanodeInfo[] original
) throws IOException {
if (nodes.length != original.length + 1) {
throw new IOException(
"Failed to replace a bad datanode on the existing pipeline "
+ "due to no more good datanodes being available to try. "
+ "(Nodes: current=" + Arrays.asList(nodes)
+ ", original=" + Arrays.asList(original) + "). "
+ "The current failed datanode replacement policy is "
+ dfsClient.dtpReplaceDatanodeOnFailure
+ ", and a client may configure this via '"
+ BlockWrite.ReplaceDatanodeOnFailure.POLICY_KEY
+ "' in its configuration.");
}
for(int i = 0; i < nodes.length; i++) {
int j = 0;
for(; j < original.length && !nodes[i].equals(original[j]); j++);
if (j == original.length) {
return i;
}
}
throw new IOException("Failed: new datanode not found: nodes="
+ Arrays.asList(nodes) + ", original=" + Arrays.asList(original));
}
{code}
client warn logs is:
{code:java}
WARN [DataStreamer for file
/home/yarn/opensearch/in/data/120141286/0_65535/table/ucs_process/MANIFEST-093545
block BP-1742758844-11.138.8.184-1483707043031:blk_7086344902_6012765313]
org.apache.hadoop.hdfs.DFSClient: DataStreamer Exception
java.io.IOException: Failed to replace a bad datanode on the existing pipeline
due to no more good datanodes being available to try. (Nodes:
current=[DatanodeInfoWithStorage[11.138.5.4:50010,DS-04826cfc-1885-4213-a58b-8606845c5c42,SSD],
DatanodeInfoWithStorage[11.138.5.9:50010,DS-f6d8eb8b-2550-474b-a692-c991d7a6f6b3,SSD],
DatanodeInfoWithStorage[11.138.5.153:50010,DS-f5d77ca0-6fe3-4523-8ca8-5af975f845b6,SSD],
DatanodeInfoWithStorage[11.138.9.156:50010,DS-0d15ea12-1bad-4444-84f7-1a4917a1e194,DISK]],
original=[DatanodeInfoWithStorage[11.138.5.4:50010,DS-04826cfc-1885-4213-a58b-8606845c5c42,SSD],
DatanodeInfoWithStorage[11.138.9.156:50010,DS-0d15ea12-1bad-4444-84f7-1a4917a1e194,DISK]]).
The current failed datanode replacement policy is DEFAULT, and a client may
configure this via 'dfs.client.block.write.replace-datanode-on-failure.policy'
in its configuration.
{code}
was:
Consider following situation:
1. create a file with ALLSSD policy
2. return [SSD,SSD,DISK] due to lack of SSD space
3. client call NameNodeRpcServer#getAdditionalDatanode when recovering write
pipeline and replacing bad datanode
4. BlockPlacementPolicyDefault#chooseTarget will call
StoragePolicy#chooseStorageTypes(3, [SSD,DISK], none, false), but
chooseStorageTypes return [SSD,SSD]
5. do numOfReplicas = requiredStorageTypes.size() and numOfReplicas is set to 2
and choose additional two datanodes
6. BlockPlacementPolicyDefault#chooseTarget return four datanodes to client
7. DataStreamer#findNewDatanode find nodes.length != original.length + 1 and
throw IOException, and finally lead to write failed
client warn logs is:
{code:java}
WARN [DataStreamer for file
/home/yarn/opensearch/in/data/120141286/0_65535/table/ucs_process/MANIFEST-093545
block BP-1742758844-11.138.8.184-1483707043031:blk_7086344902_6012765313]
org.apache.hadoop.hdfs.DFSClient: DataStreamer Exception
java.io.IOException: Failed to replace a bad datanode on the existing pipeline
due to no more good datanodes being available to try. (Nodes:
current=[DatanodeInfoWithStorage[11.138.5.4:50010,DS-04826cfc-1885-4213-a58b-8606845c5c42,SSD],
DatanodeInfoWithStorage[11.138.5.9:50010,DS-f6d8eb8b-2550-474b-a692-c991d7a6f6b3,SSD],
DatanodeInfoWithStorage[11.138.5.153:50010,DS-f5d77ca0-6fe3-4523-8ca8-5af975f845b6,SSD],
DatanodeInfoWithStorage[11.138.9.156:50010,DS-0d15ea12-1bad-4444-84f7-1a4917a1e194,DISK]],
original=[DatanodeInfoWithStorage[11.138.5.4:50010,DS-04826cfc-1885-4213-a58b-8606845c5c42,SSD],
DatanodeInfoWithStorage[11.138.9.156:50010,DS-0d15ea12-1bad-4444-84f7-1a4917a1e194,DISK]]).
The current failed datanode replacement policy is DEFAULT, and a client may
configure this via 'dfs.client.block.write.replace-datanode-on-failure.policy'
in its configuration.
{code}
> replace datanode failed because of NameNodeRpcServer#getAdditionalDatanode
> returning excessive datanodeInfo
> ------------------------------------------------------------------------------------------------------------
>
> Key: HDFS-13915
> URL: https://issues.apache.org/jira/browse/HDFS-13915
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: hdfs
> Environment:
> Reporter: Jiandan Yang
> Assignee: Jiandan Yang
> Priority: Major
>
> Consider following situation:
> 1. create a file with ALLSSD policy
> 2. return [SSD,SSD,DISK] due to lack of SSD space
> 3. client call NameNodeRpcServer#getAdditionalDatanode when recovering write
> pipeline and replacing bad datanode
> 4. BlockPlacementPolicyDefault#chooseTarget will call
> StoragePolicy#chooseStorageTypes(3, [SSD,DISK], none, false), but
> chooseStorageTypes return [SSD,SSD]
> {code:java}
> @Test
> public void testAllSSDFallbackAndNonNewBlock() {
> final BlockStoragePolicy allSSD = POLICY_SUITE.getPolicy(ALLSSD);
> List<StorageType> storageTypes = allSSD.chooseStorageTypes((short) 3,
> Arrays.asList(StorageType.DISK, StorageType.SSD),
> EnumSet.noneOf(StorageType.class), false);
> assertEquals(2, storageTypes.size());
> assertEquals(StorageType.SSD, storageTypes.get(0));
> assertEquals(StorageType.SSD, storageTypes.get(1));
> }
> {code}
> 5. do numOfReplicas = requiredStorageTypes.size() and numOfReplicas is set to
> 2 and choose additional two datanodes
> 6. BlockPlacementPolicyDefault#chooseTarget return four datanodes to client
> 7. DataStreamer#findNewDatanode find nodes.length != original.length + 1 and
> throw IOException, and finally lead to write failed
> {code:java}
> private int findNewDatanode(final DatanodeInfo[] original
> ) throws IOException {
> if (nodes.length != original.length + 1) {
> throw new IOException(
> "Failed to replace a bad datanode on the existing pipeline "
> + "due to no more good datanodes being available to try. "
> + "(Nodes: current=" + Arrays.asList(nodes)
> + ", original=" + Arrays.asList(original) + "). "
> + "The current failed datanode replacement policy is "
> + dfsClient.dtpReplaceDatanodeOnFailure
> + ", and a client may configure this via '"
> + BlockWrite.ReplaceDatanodeOnFailure.POLICY_KEY
> + "' in its configuration.");
> }
> for(int i = 0; i < nodes.length; i++) {
> int j = 0;
> for(; j < original.length && !nodes[i].equals(original[j]); j++);
> if (j == original.length) {
> return i;
> }
> }
> throw new IOException("Failed: new datanode not found: nodes="
> + Arrays.asList(nodes) + ", original=" + Arrays.asList(original));
> }
> {code}
> client warn logs is:
> {code:java}
> WARN [DataStreamer for file
> /home/yarn/opensearch/in/data/120141286/0_65535/table/ucs_process/MANIFEST-093545
> block BP-1742758844-11.138.8.184-1483707043031:blk_7086344902_6012765313]
> org.apache.hadoop.hdfs.DFSClient: DataStreamer Exception
> java.io.IOException: Failed to replace a bad datanode on the existing
> pipeline due to no more good datanodes being available to try. (Nodes:
> current=[DatanodeInfoWithStorage[11.138.5.4:50010,DS-04826cfc-1885-4213-a58b-8606845c5c42,SSD],
>
> DatanodeInfoWithStorage[11.138.5.9:50010,DS-f6d8eb8b-2550-474b-a692-c991d7a6f6b3,SSD],
>
> DatanodeInfoWithStorage[11.138.5.153:50010,DS-f5d77ca0-6fe3-4523-8ca8-5af975f845b6,SSD],
>
> DatanodeInfoWithStorage[11.138.9.156:50010,DS-0d15ea12-1bad-4444-84f7-1a4917a1e194,DISK]],
>
> original=[DatanodeInfoWithStorage[11.138.5.4:50010,DS-04826cfc-1885-4213-a58b-8606845c5c42,SSD],
>
> DatanodeInfoWithStorage[11.138.9.156:50010,DS-0d15ea12-1bad-4444-84f7-1a4917a1e194,DISK]]).
> The current failed datanode replacement policy is DEFAULT, and a client may
> configure this via
> 'dfs.client.block.write.replace-datanode-on-failure.policy' in its
> configuration.
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
