shahrs87 commented on code in PR #6566:
URL: https://github.com/apache/hadoop/pull/6566#discussion_r1508229688
##########
hadoop-hdfs-project/hadoop-hdfs-client/src/main/java/org/apache/hadoop/hdfs/DataStreamer.java:
##########
@@ -1618,33 +1625,47 @@ private void setupPipelineForAppendOrRecovery() throws
IOException {
LOG.warn(msg);
lastException.set(new IOException(msg));
streamerClosed = true;
- return;
+ return false;
}
- setupPipelineInternal(nodes, storageTypes, storageIDs);
+ return setupPipelineInternal(nodes, storageTypes, storageIDs);
}
- protected void setupPipelineInternal(DatanodeInfo[] datanodes,
+ protected boolean setupPipelineInternal(DatanodeInfo[] datanodes,
StorageType[] nodeStorageTypes, String[] nodeStorageIDs)
throws IOException {
boolean success = false;
long newGS = 0L;
+ boolean isCreateStage = BlockConstructionStage.PIPELINE_SETUP_CREATE ==
stage;
while (!success && !streamerClosed && dfsClient.clientRunning) {
if (!handleRestartingDatanode()) {
- return;
+ return false;
}
- final boolean isRecovery = errorState.hasInternalError();
+ final boolean isRecovery = errorState.hasInternalError() &&
!isCreateStage;
+
+
if (!handleBadDatanode()) {
- return;
+ return false;
}
handleDatanodeReplacement();
+ // During create stage, if we remove a node (nodes.length - 1)
Review Comment:
I think this comment needs to be updated.
##########
hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestDistributedFileSystem.java:
##########
@@ -2651,5 +2653,154 @@ public void
testNameNodeCreateSnapshotTrashRootOnStartup()
}
}
+ @Test
+ public void testSingleRackFailureDuringPipelineSetupMinReplicationPossible()
throws Exception {
+ Configuration conf = getTestConfiguration();
+ conf.setClass(
+ DFSConfigKeys.DFS_BLOCK_REPLICATOR_CLASSNAME_KEY,
+ BlockPlacementPolicyRackFaultTolerant.class,
+ BlockPlacementPolicy.class);
+ conf.setBoolean(
+ HdfsClientConfigKeys.BlockWrite.ReplaceDatanodeOnFailure.ENABLE_KEY,
+ false);
+ conf.setInt(HdfsClientConfigKeys.BlockWrite.ReplaceDatanodeOnFailure.
+ MIN_REPLICATION, 2);
+ // 3 racks & 3 nodes. 1 per rack
+ try (MiniDFSCluster cluster = new
MiniDFSCluster.Builder(conf).numDataNodes(3)
+ .racks(new String[] {"/rack1", "/rack2", "/rack3"}).build()) {
+ cluster.waitClusterUp();
+ DistributedFileSystem fs = cluster.getFileSystem();
+ // kill one DN, so only 2 racks stays with active DN
+ cluster.stopDataNode(0);
+ // create a file with replication 3, for rack fault tolerant BPP,
+ // it should allocate nodes in all 3 racks.
+ DFSTestUtil.createFile(fs, new Path("/testFile"), 1024L, (short) 3,
1024L);
+ cluster.shutdown(true);
+ }
+ }
+
+ @Test
+ public void
testSingleRackFailureDuringPipelineSetupMinReplicationImpossible() throws
Exception {
+ Configuration conf = getTestConfiguration();
+ conf.setClass(
+ DFSConfigKeys.DFS_BLOCK_REPLICATOR_CLASSNAME_KEY,
+ BlockPlacementPolicyRackFaultTolerant.class,
+ BlockPlacementPolicy.class);
+ conf.setBoolean(
+ HdfsClientConfigKeys.BlockWrite.ReplaceDatanodeOnFailure.ENABLE_KEY,
+ false);
+ conf.setInt(HdfsClientConfigKeys.BlockWrite.ReplaceDatanodeOnFailure.
+ MIN_REPLICATION, 3);
+ // 3 racks & 3 nodes. 1 per rack
+ try (MiniDFSCluster cluster = new
MiniDFSCluster.Builder(conf).numDataNodes(3)
+ .racks(new String[] {"/rack1", "/rack2", "/rack3"}).build()) {
+ cluster.waitClusterUp();
+ DistributedFileSystem fs = cluster.getFileSystem();
+ // kill one DN, so only 2 racks stays with active DN
+ cluster.stopDataNode(0);
+ boolean threw = false;
+ try {
+ DFSTestUtil.createFile(fs, new Path("/testFile"), 1024L, (short) 3,
1024L);
+ } catch (IOException e) {
+ // success
+ threw = true;
+ } finally {
+ cluster.shutdown(true);
+ }
+ assertTrue("Failed to throw IOE when creating a file with less "
+ + "DNs than required for min replication", threw);
+ }
+ }
+
+ @Test
+ public void
testMultipleRackFailureDuringPipelineSetupMinReplicationPossible() throws
Exception {
+ Configuration conf = getTestConfiguration();
+ conf.setClass(
+ DFSConfigKeys.DFS_BLOCK_REPLICATOR_CLASSNAME_KEY,
+ BlockPlacementPolicyRackFaultTolerant.class,
+ BlockPlacementPolicy.class);
+ conf.setBoolean(
+ HdfsClientConfigKeys.BlockWrite.ReplaceDatanodeOnFailure.ENABLE_KEY,
+ false);
+ conf.setInt(HdfsClientConfigKeys.BlockWrite.ReplaceDatanodeOnFailure.
+ MIN_REPLICATION, 1);
+ // 3 racks & 3 nodes. 1 per rack
+ try (MiniDFSCluster cluster = new
MiniDFSCluster.Builder(conf).numDataNodes(3)
+ .racks(new String[] {"/rack1", "/rack2", "/rack3"}).build()) {
+ cluster.waitClusterUp();
+ DistributedFileSystem fs = cluster.getFileSystem();
+ // kill 2 DN, so only 1 racks stays with active DN
+ cluster.stopDataNode(0);
+ cluster.stopDataNode(1);
+ // create a file with replication 3, for rack fault tolerant BPP,
+ // it should allocate nodes in all 3 racks.
+ DFSTestUtil.createFile(fs, new Path("/testFile"), 1024L, (short) 3,
1024L);
+ cluster.shutdown(true);
Review Comment:
Do we want to move `cluster.shutdown(true);` to finally block?
##########
hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestDistributedFileSystem.java:
##########
@@ -2651,5 +2653,154 @@ public void
testNameNodeCreateSnapshotTrashRootOnStartup()
}
}
+ @Test
+ public void testSingleRackFailureDuringPipelineSetupMinReplicationPossible()
throws Exception {
+ Configuration conf = getTestConfiguration();
+ conf.setClass(
+ DFSConfigKeys.DFS_BLOCK_REPLICATOR_CLASSNAME_KEY,
+ BlockPlacementPolicyRackFaultTolerant.class,
+ BlockPlacementPolicy.class);
+ conf.setBoolean(
+ HdfsClientConfigKeys.BlockWrite.ReplaceDatanodeOnFailure.ENABLE_KEY,
+ false);
+ conf.setInt(HdfsClientConfigKeys.BlockWrite.ReplaceDatanodeOnFailure.
+ MIN_REPLICATION, 2);
+ // 3 racks & 3 nodes. 1 per rack
+ try (MiniDFSCluster cluster = new
MiniDFSCluster.Builder(conf).numDataNodes(3)
+ .racks(new String[] {"/rack1", "/rack2", "/rack3"}).build()) {
+ cluster.waitClusterUp();
+ DistributedFileSystem fs = cluster.getFileSystem();
+ // kill one DN, so only 2 racks stays with active DN
+ cluster.stopDataNode(0);
+ // create a file with replication 3, for rack fault tolerant BPP,
+ // it should allocate nodes in all 3 racks.
+ DFSTestUtil.createFile(fs, new Path("/testFile"), 1024L, (short) 3,
1024L);
+ cluster.shutdown(true);
Review Comment:
Do we want to move `cluster.shutdown(true);` to finally block?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
