[
https://issues.apache.org/jira/browse/HDFS-17401?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17839188#comment-17839188
]
ASF GitHub Bot commented on HDFS-17401:
---------------------------------------
haiyang1987 commented on code in PR #6597:
URL: https://github.com/apache/hadoop/pull/6597#discussion_r1522764907
##########
hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/TestReconstructStripedBlocks.java:
##########
@@ -575,5 +576,80 @@ public void testReconstructionWithStorageTypeNotEnough()
throws Exception {
cluster.shutdown();
}
}
+ @Test
+ public void testDeleteOverReplicatedStripedBlock() throws Exception {
+ final HdfsConfiguration conf = new HdfsConfiguration();
+ conf.setInt(DFSConfigKeys.DFS_NAMENODE_REDUNDANCY_INTERVAL_SECONDS_KEY, 1);
+ conf.setBoolean(DFSConfigKeys.DFS_NAMENODE_REDUNDANCY_CONSIDERLOAD_KEY,
+ false);
+ StorageType[][] st = new StorageType[groupSize + 2][1];
+ for (int i = 0; i < st.length-1; i++){
+ st[i] = new StorageType[]{StorageType.SSD};
+ }
+ st[st.length -1] = new StorageType[]{StorageType.DISK};
+
+ cluster = new MiniDFSCluster.Builder(conf).numDataNodes(groupSize + 2)
+ .storagesPerDatanode(1)
+ .storageTypes(st)
+ .build();
+ cluster.waitActive();
+ DistributedFileSystem fs = cluster.getFileSystem();
+ fs.enableErasureCodingPolicy(
+ StripedFileTestUtil.getDefaultECPolicy().getName());
+ try {
+ fs.mkdirs(dirPath);
+ fs.setErasureCodingPolicy(dirPath,
+ StripedFileTestUtil.getDefaultECPolicy().getName());
+ fs.setStoragePolicy(dirPath, HdfsConstants.ALLSSD_STORAGE_POLICY_NAME);
+ DFSTestUtil.createFile(fs, filePath,
+ cellSize * dataBlocks * 2, (short) 1, 0L);
+ // Stop a dn
+ LocatedBlocks blks =
fs.getClient().getLocatedBlocks(filePath.toString(), 0);
+ LocatedStripedBlock block = (LocatedStripedBlock)
blks.getLastLocatedBlock();
+ DatanodeInfo dnToStop = block.getLocations()[0];
+
+ MiniDFSCluster.DataNodeProperties dnProp =
+ cluster.stopDataNode(dnToStop.getXferAddr());
+ cluster.stopDataNode(dnToStop.getXferAddr());
+ cluster.setDataNodeDead(dnToStop);
+
+ // Wait for reconstruction to happen
+ DFSTestUtil.waitForReplication(fs, filePath, groupSize, 15 * 1000);
+
+ DatanodeInfo dnToStop2 = block.getLocations()[1];
+ cluster.stopDataNode(dnToStop2.getXferAddr());
+ cluster.setDataNodeDead(dnToStop2);
+ DFSTestUtil.waitForReplication(fs, filePath, groupSize, 15 * 1000);
+
+ // Bring the dn back: 10 internal blocks now
+ cluster.restartDataNode(dnProp);
+ cluster.waitActive();
+ DFSTestUtil.verifyClientStats(conf, cluster);
+
+ // Currently namenode is able to track the missing block. And restart NN
Review 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?
> EC: Excess internal block may not be able to be deleted correctly when it's
> stored in fallback storage
> ------------------------------------------------------------------------------------------------------
>
> Key: HDFS-17401
> URL: https://issues.apache.org/jira/browse/HDFS-17401
> Project: Hadoop HDFS
> Issue Type: Bug
> Affects Versions: 3.3.6
> Reporter: Ruinan Gu
> Assignee: Ruinan Gu
> Priority: Major
> Labels: pull-request-available
>
> 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)]
> The are two index 0 internal block 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.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
