HDFS-8056. Decommissioned dead nodes should continue to be counted as dead after NN restart. (mingma)

(cherry picked from commit 1c4951a7a0)
This commit is contained in:
Ming Ma 2015-11-19 10:04:01 -08:00
parent c74e42b4a2
commit 5a3db21563
4 changed files with 43 additions and 2 deletions

View File

@ -798,6 +798,9 @@ Release 2.8.0 - UNRELEASED
HDFS-9252. Change TestFileTruncate to use FsDatasetTestUtils to get block
file size and genstamp. (Lei (Eddy) Xu via cmccabe)
HDFS-8056. Decommissioned dead nodes should continue to be counted as dead
after NN restart. (mingma)
OPTIMIZATIONS
HDFS-8026. Trace FSOutputSummer#writeChecksumChunks rather than

View File

@ -1267,7 +1267,7 @@ public class DatanodeManager {
if (listDeadNodes) {
for (InetSocketAddress addr : includedNodes) {
if (foundNodes.matchedBy(addr) || excludedNodes.match(addr)) {
if (foundNodes.matchedBy(addr)) {
continue;
}
// The remaining nodes are ones that are referenced by the hosts
@ -1284,6 +1284,9 @@ public class DatanodeManager {
addr.getPort() == 0 ? defaultXferPort : addr.getPort(),
defaultInfoPort, defaultInfoSecurePort, defaultIpcPort));
setDatanodeDead(dn);
if (excludedNodes.match(addr)) {
dn.setDecommissioned();
}
nodes.add(dn);
}
}

View File

@ -924,6 +924,41 @@ public class TestDecommission {
cluster.shutdown();
}
/**
* Tests dead node count after restart of namenode
**/
@Test(timeout=360000)
public void testDeadNodeCountAfterNamenodeRestart()throws Exception {
LOG.info("Starting test testDeadNodeCountAfterNamenodeRestart");
int numNamenodes = 1;
int numDatanodes = 2;
startCluster(numNamenodes, numDatanodes, conf);
DFSClient client = getDfsClient(cluster.getNameNode(), conf);
DatanodeInfo[] info = client.datanodeReport(DatanodeReportType.LIVE);
DatanodeInfo excludedDatanode = info[0];
String excludedDatanodeName = info[0].getXferAddr();
writeConfigFile(hostsFile, new ArrayList<String>(Arrays.asList(
excludedDatanodeName, info[1].getXferAddr())));
decommissionNode(0, excludedDatanode.getDatanodeUuid(), null,
AdminStates.DECOMMISSIONED);
cluster.stopDataNode(excludedDatanodeName);
DFSTestUtil.waitForDatanodeState(
cluster, excludedDatanode.getDatanodeUuid(), false, 20000);
//Restart the namenode
cluster.restartNameNode();
assertEquals("There should be one node alive", 1,
client.datanodeReport(DatanodeReportType.LIVE).length);
assertEquals("There should be one node dead", 1,
client.datanodeReport(DatanodeReportType.DEAD).length);
cluster.shutdown();
}
/**
* Test using a "registration name" in a host include file.
*

View File

@ -151,7 +151,7 @@ public class TestHostFileManager {
Assert.assertEquals(1, dm.getDatanodeListForReport(HdfsConstants
.DatanodeReportType.DEAD).size());
excludedNodes.add(entry("127.0.0.3"));
Assert.assertEquals(0, dm.getDatanodeListForReport(HdfsConstants
Assert.assertEquals(1, dm.getDatanodeListForReport(HdfsConstants
.DatanodeReportType.DEAD).size());
}
}