HDFS-17401. EC: Excess internal block may not be able to be deleted correctly when it's stored in fallback storage #6597
+77
−14
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
💔 -1 overall
This message was automatically generated. |
|
@jojochuang @tomscut Could you please take a look?The failed UT seems unrelated. |
|
@zhangshuyan0 @haiyang1987 Could you please take a look? |
zhangshuyan0
changed the title
zhangshuyan0
reviewed
Mar 6, 2024
| import org.apache.hadoop.hdfs.protocol.LocatedBlocks; | ||
| import org.apache.hadoop.hdfs.protocol.LocatedStripedBlock; | ||
| import org.apache.hadoop.hdfs.protocol.SystemErasureCodingPolicies; | ||
| import org.apache.hadoop.hdfs.protocol.*; |
There was a problem hiding this comment.
Please do not use * in import.
|
💔 -1 overall
This message was automatically generated. |
haiyang1987
reviewed
Mar 7, 2024
| fs.setStoragePolicy(dirPath, HdfsConstants.ALLSSD_STORAGE_POLICY_NAME); | ||
| DFSTestUtil.createFile(fs, filePath, | ||
| cellSize * dataBlocks * 2, (short) 1, 0L); | ||
| FSNamesystem fsn3 = cluster.getNamesystem(); |
There was a problem hiding this comment.
can remove unused code.
| DFSTestUtil.waitForReplication(fs, filePath, groupSize, 15 * 1000); | ||
|
|
||
| DatanodeInfo dnToStop2 = block.getLocations()[1]; | ||
| MiniDFSCluster.DataNodeProperties dnProp2 = |
haiyang1987
reviewed
Mar 7, 2024
| cellSize * dataBlocks * 2, (short) 1, 0L); | ||
| FSNamesystem fsn3 = cluster.getNamesystem(); | ||
| BlockManager bm3 = fsn3.getBlockManager(); | ||
| // stop a dn |
There was a problem hiding this comment.
The first letter should be uppercase~
|
💔 -1 overall
This message was automatically generated. |
|
please fix checkstytle, thanks~ |
|
💔 -1 overall
This message was automatically generated. |
|
In |
haiyang1987
reviewed
Mar 11, 2024
| @@ -4359,9 +4348,15 @@ private void chooseExcessRedundancyStriped(BlockCollection bc, | |||
| } | |||
| } | |||
| if (candidates.size() > 1) { | |||
| List<StorageType> internalExcessTypes = storagePolicy.chooseExcess( | |||
| (short) 1, DatanodeStorageInfo.toStorageTypes(candidates)); | |||
| if (internalExcessTypes.isEmpty()) { | |||
There was a problem hiding this comment.
About internalExcessTypes, could you explain which case will empty ? thanks~
haiyang1987
reviewed
Mar 11, 2024
| // Wait for reconstruction to happen | ||
| DFSTestUtil.waitForReplication(fs, filePath, groupSize, 15 * 1000); | ||
|
|
||
| DatanodeInfo dnToStop2 = block.getLocations()[1]; |
There was a problem hiding this comment.
here may be add cluster.stopDataNode(dnToStop2.getXferAddr());
This parameter is added in https:/apache/hadoop/pull/5474 (Also a patch I contributed), after the modification of the patch, I dont't think the parameter is neccessary any more, so we can remove it.The "empty exessType" log can be removed too, after the modification of this patch, the exessType is impossible to be null, becase the candidates is > 1 and the "replication"(first parameter of storagePolicy.chooseExcess) is constantly 1. |
|
💔 -1 overall
This message was automatically generated. |
|
💔 -1 overall
This message was automatically generated. |
|
💔 -1 overall
This message was automatically generated. |
haiyang1987
reviewed
Mar 13, 2024
|
|
||
| MiniDFSCluster.DataNodeProperties dnProp = | ||
| cluster.stopDataNode(dnToStop.getXferAddr()); | ||
| cluster.stopDataNode(dnToStop.getXferAddr()); |
There was a problem hiding this comment.
here repeated, need remove.
|
From my side seems well, except for some small issues in test UT. |
|
💔 -1 overall
This message was automatically generated. |
haiyang1987
reviewed
Apr 20, 2024
| cluster.waitActive(); | ||
| DFSTestUtil.verifyClientStats(conf, cluster); | ||
|
|
||
| // Currently namenode is able to track the missing block. And restart NN |
There was a problem hiding this comment.
here why need restart namenode?
For Line[629-639]
for (DataNode dn : cluster.getDataNodes()) {
DataNodeTestUtils.triggerBlockReport(dn);
}
BlockManager bm = cluster.getNamesystem().getBlockManager();
GenericTestUtils.waitFor(()
-> bm.getPendingDeletionBlocksCount() == 0,
10, 2000);
for (DataNode dn : cluster.getDataNodes()) {
DataNodeTestUtils.triggerHeartbeat(dn);
}
for (DataNode dn : cluster.getDataNodes()) {
DataNodeTestUtils.triggerDeletionReport(dn);
}
Maybe update this logic to avoid using Thread.sleep(3000), what do you think?
|
🎊 +1 overall
This message was automatically generated. |
|
💔 -1 overall
This message was automatically generated. |
|
💔 -1 overall
This message was automatically generated. |
|
Please fix checkstytle and the UT failed issue, thanks~ |
|
🎊 +1 overall
This message was automatically generated. |
@haiyang1987 Hi, just found out the failed UT is caused by the removal of Thread.sleep(3000). I think there is delay between the trigger of blockReport and the deletion of the block, so I will keep Thread.sleep(3000) in the UT. |
|
@haiyang1987 Could you please take a look? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description of PR
jira: https://issues.apache.org/jira/browse/HDFS-17401
Excess internal block can't be deleted correctly when it's stored in fallback storage.
Simple case:
EC-RS-6-3-1024k file is stored using ALL_SSD storage policy(SSD is default storage type and DISK is fallback storage type), if the block group is as follows
[0(SSD), 0(SSD), 1(SSD), 2(SSD), 3(SSD), 4(SSD), 5(SSD), 6(SSD), 7(SSD), 8(DISK)]
There are two index 0 internal blocks and one of them should be chosen to delete.But the current implement chooses the index 0 internal blocks as candidates but DISK as exess storage type.As a result, the exess storage type(DISK) can not correspond to the exess internal blocks' storage type(SSD) correctly, and the exess internal block can not be deleted correctly.