[
https://issues.apache.org/jira/browse/HDFS-14768?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16928962#comment-16928962
]
guojh commented on HDFS-14768:
------------------------------
[~zhaoyim] Yes, after you sleep 3 min, NN will reconstruct a new correct block
for I stop the DN on the block with index 6. But in product, because exist two
index 6 block, NN may delete the correct one, if data block is missing, then
the incorrect block will encode to generate the data block, after that, the
file is corruption and is not useable.
> In some cases, erasure blocks are corruption when they are reconstruct.
> ------------------------------------------------------------------------
>
> Key: HDFS-14768
> URL: https://issues.apache.org/jira/browse/HDFS-14768
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: datanode, erasure-coding, hdfs, namenode
> Affects Versions: 3.0.2
> Reporter: guojh
> Assignee: guojh
> Priority: Major
> Labels: patch
> Fix For: 3.3.0
>
> Attachments: 1568275810244.jpg, 1568276338275.jpg,
> HDFS-14768.000.patch, guojh_UT_after_deomission.txt,
> guojh_UT_before_deomission.txt, zhaoyiming_UT_after_deomission.txt,
> zhaoyiming_UT_beofre_deomission.txt
>
>
> Policy is RS-6-3-1024K, version is hadoop 3.0.2;
> We suppose a file's block Index is [0,1,2,3,4,5,6,7,8], And decommission
> index[3,4], increase the index 6 datanode's
> pendingReplicationWithoutTargets that make it large than
> replicationStreamsHardLimit(we set 14). Then, After the method
> chooseSourceDatanodes of BlockMananger, the liveBlockIndices is
> [0,1,2,3,4,5,7,8], Block Counter is, Live:7, Decommission:2.
> In method scheduleReconstruction of BlockManager, the additionalReplRequired
> is 9 - 7 = 2. After Namenode choose two target Datanode, will assign a
> erasureCode task to target datanode.
> When datanode get the task will build targetIndices from liveBlockIndices
> and target length. the code is blow.
> {code:java}
> // code placeholder
> targetIndices = new short[targets.length];
> private void initTargetIndices() {
> BitSet bitset = reconstructor.getLiveBitSet();
> int m = 0; hasValidTargets = false;
> for (int i = 0; i < dataBlkNum + parityBlkNum; i++) {
> if (!bitset.get) {
> if (reconstructor.getBlockLen > 0) {
> if (m < targets.length) {
> targetIndices[m++] = (short)i;
> hasValidTargets = true;
> }
> }
> }
> }
> {code}
> targetIndices[0]=6, and targetIndices[1] is aways 0 from initial value.
> The StripedReader is aways create reader from first 6 index block, and is
> [0,1,2,3,4,5]
> Use the index [0,1,2,3,4,5] to build target index[6,0] will trigger the isal
> bug. the block index6's data is corruption(all data is zero).
> I write a unit test can stabilize repreduce.
> {code:java}
> // code placeholder
> private int replicationStreamsHardLimit =
> DFSConfigKeys.DFS_NAMENODE_REPLICATION_STREAMS_HARD_LIMIT_DEFAULT;
> numDNs = dataBlocks + parityBlocks + 10;
> @Test(timeout = 240000)
> public void testFileDecommission() throws Exception {
> LOG.info("Starting test testFileDecommission");
> final Path ecFile = new Path(ecDir, "testFileDecommission");
> int writeBytes = cellSize * dataBlocks;
> writeStripedFile(dfs, ecFile, writeBytes);
> Assert.assertEquals(0, bm.numOfUnderReplicatedBlocks());
> FileChecksum fileChecksum1 = dfs.getFileChecksum(ecFile, writeBytes);
> final INodeFile fileNode = cluster.getNamesystem().getFSDirectory()
> .getINode4Write(ecFile.toString()).asFile();
> LocatedBlocks locatedBlocks =
> StripedFileTestUtil.getLocatedBlocks(ecFile, dfs);
> LocatedBlock lb = dfs.getClient().getLocatedBlocks(ecFile.toString(), 0)
> .get(0);
> DatanodeInfo[] dnLocs = lb.getLocations();
> LocatedStripedBlock lastBlock =
> (LocatedStripedBlock)locatedBlocks.getLastLocatedBlock();
> DatanodeInfo[] storageInfos = lastBlock.getLocations();
> //
> DatanodeDescriptor datanodeDescriptor =
> cluster.getNameNode().getNamesystem()
>
> .getBlockManager().getDatanodeManager().getDatanode(storageInfos[6].getDatanodeUuid());
> BlockInfo firstBlock = fileNode.getBlocks()[0];
> DatanodeStorageInfo[] dStorageInfos = bm.getStorages(firstBlock);
> // the first heartbeat will consume 3 replica tasks
> for (int i = 0; i <= replicationStreamsHardLimit + 3; i++) {
> BlockManagerTestUtil.addBlockToBeReplicated(datanodeDescriptor, new
> Block(i),
> new DatanodeStorageInfo[]{dStorageInfos[0]});
> }
> assertEquals(dataBlocks + parityBlocks, dnLocs.length);
> int[] decommNodeIndex = {3, 4};
> final List<DatanodeInfo> decommisionNodes = new ArrayList<DatanodeInfo>();
> // add the node which will be decommissioning
> decommisionNodes.add(dnLocs[decommNodeIndex[0]]);
> decommisionNodes.add(dnLocs[decommNodeIndex[1]]);
> decommissionNode(0, decommisionNodes, AdminStates.DECOMMISSIONED);
> assertEquals(decommisionNodes.size(), fsn.getNumDecomLiveDataNodes());
> bm.getDatanodeManager().removeDatanode(datanodeDescriptor);
> //assertNull(checkFile(dfs, ecFile, 9, decommisionNodes, numDNs));
> // Ensure decommissioned datanode is not automatically shutdown
> DFSClient client = getDfsClient(cluster.getNameNode(0), conf);
> assertEquals("All datanodes must be alive", numDNs,
> client.datanodeReport(DatanodeReportType.LIVE).length);
> FileChecksum fileChecksum2 = dfs.getFileChecksum(ecFile, writeBytes);
> Assert.assertTrue("Checksum mismatches!",
> fileChecksum1.equals(fileChecksum2));
> StripedFileTestUtil.checkData(dfs, ecFile, writeBytes, decommisionNodes,
> null, blockGroupSize);
> }
> {code}
>
--
This message was sent by Atlassian Jira
(v8.3.2#803003)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
