Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

HDFS-16479. EC: NameNode should not send a reconstruction work when the source datanodes are insufficient #4138

Merged
merged 6 commits into from
Apr 14, 2022
Merged

HDFS-16479. EC: NameNode should not send a reconstruction work when the source datanodes are insufficient #4138

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
8a4daad
HDFS-16479. EC: NameNode should not send a reconstruction work when t…
tasanuma Apr 5, 2022
1b40cf5
addressing short data blocks
tasanuma Apr 5, 2022
7784d02
use BlockInfoStriped.getRealDataBlockNum()
tasanuma Apr 12, 2022
f6c9d6b
add NameNode.getNameNodeMetrics().incNumTimesReReplicationNotScheduled()
tasanuma Apr 12, 2022
494a921
fix typo comment
tasanuma Apr 12, 2022
a0d5756
update comments and variable names
tasanuma Apr 12, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 9 additions & 0 deletions ...hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2163,6 +2163,15 @@ BlockReconstructionWork scheduleReconstruction(BlockInfo block,
return null;
}

// skip if source datanodes for reconstructing ec block are not enough
if (block.isStriped()) {
BlockInfoStriped stripedBlock = (BlockInfoStriped) block;
if (stripedBlock.getDataBlockNum() > srcNodes.length) {
Copy link
Member

@ayushtkn ayushtkn Apr 5, 2022 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Had a very quick look.
Just thinking about a scenario with say RS-6-3-1024k, and we just write 1 mb, in that case the total number of blocks available will be 1 Datablock + 3 Parity. In that case BG itself will have total 4 Blocks. Will this code start returning null? Not sure if getRealDataBlockNum helps here or not. If it is actually a problem

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@ayushtkn Thanks for your review. You're right, it's a problem.
I updated the PR to calculate the real data block number. It is the same logic used in StripedReader. I also added one more unit test to cover the case.

LOG.debug("Block {} cannot be reconstructed due to shortage of source datanodes ", block);
return null;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we increment the metrics before returning null

    NameNode.getNameNodeMetrics().incNumTimesReReplicationNotScheduled();

}
}

// liveReplicaNodes can include READ_ONLY_SHARED replicas which are
// not included in the numReplicas.liveReplicas() count
assert liveReplicaNodes.size() >= numReplicas.liveReplicas();
Expand Down
43 changes: 43 additions & 0 deletions ...op-hdfs/src/test/java/org/apache/hadoop/hdfs/server/blockmanagement/TestBlockManager.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -852,6 +852,49 @@ public void testChooseSrcDNWithDupECInDecommissioningNode() throws Exception {
0, numReplicas.redundantInternalBlocks());
}

@Test
public void testSkipReconstructionWithManyBusyNodes() {
long blockId = -9223372036854775776L; // real ec block id
// RS-3-2 EC policy
ErasureCodingPolicy ecPolicy =
SystemErasureCodingPolicies.getPolicies().get(1);
// striped blockInfo
Block aBlock = new Block(blockId, ecPolicy.getCellSize() * ecPolicy.getNumDataUnits(), 0);
BlockInfoStriped aBlockInfoStriped = new BlockInfoStriped(aBlock, ecPolicy);
// ec storageInfo
DatanodeStorageInfo ds1 = DFSTestUtil.createDatanodeStorageInfo(
"storage1", "1.1.1.1", "rack1", "host1");
DatanodeStorageInfo ds2 = DFSTestUtil.createDatanodeStorageInfo(
"storage2", "2.2.2.2", "rack2", "host2");
DatanodeStorageInfo ds3 = DFSTestUtil.createDatanodeStorageInfo(
"storage3", "3.3.3.3", "rack3", "host3");
DatanodeStorageInfo ds4 = DFSTestUtil.createDatanodeStorageInfo(
"storage4", "4.4.4.4", "rack4", "host4");

// link block with storage
aBlockInfoStriped.addStorage(ds1, aBlock);
aBlockInfoStriped.addStorage(ds2, new Block(blockId + 1, 0, 0));
aBlockInfoStriped.addStorage(ds3, new Block(blockId + 2, 0, 0));
aBlockInfoStriped.addStorage(ds4, new Block(blockId + 3, 0, 0));

addEcBlockToBM(blockId, ecPolicy);
aBlockInfoStriped.setBlockCollectionId(mockINodeId);

// reconstruction should be scheduled
BlockReconstructionWork work = bm.scheduleReconstruction(aBlockInfoStriped, 3);
assertNotNull(work);

// simulate the 3 nodes reach maxReplicationStreams
for(int i = 0; i < bm.maxReplicationStreams; i++){
ds3.getDatanodeDescriptor().incrementPendingReplicationWithoutTargets();
ds4.getDatanodeDescriptor().incrementPendingReplicationWithoutTargets();
}

// reconstruction should be skipped since the number of non-busy nodes are not enough
work = bm.scheduleReconstruction(aBlockInfoStriped, 3);
assertNull(work);
}

@Test
public void testFavorDecomUntilHardLimit() throws Exception {
bm.maxReplicationStreams = 0;
Expand Down