HDFS-14758. Make lease hard limit configurable and reduce the default.

Contributed by hemanthboyina.
This commit is contained in:
Kihwal Lee 2020-02-11 12:40:00 -06:00
parent e637797211
commit 9b8a78d97b
12 changed files with 50 additions and 23 deletions

View File

@ -577,10 +577,10 @@ public class DFSClient implements java.io.Closeable, RemotePeerFactory,
} catch (IOException e) {
// Abort if the lease has already expired.
final long elapsed = Time.monotonicNow() - getLastLeaseRenewal();
if (elapsed > HdfsConstants.LEASE_HARDLIMIT_PERIOD) {
if (elapsed > dfsClientConf.getleaseHardLimitPeriod()) {
LOG.warn("Failed to renew lease for " + clientName + " for "
+ (elapsed/1000) + " seconds (>= hard-limit ="
+ (HdfsConstants.LEASE_HARDLIMIT_PERIOD / 1000) + " seconds.) "
+ (dfsClientConf.getleaseHardLimitPeriod() / 1000) + " seconds.) "
+ "Closing all files being written ...", e);
closeAllFilesBeingWritten(true);
} else {

View File

@ -242,6 +242,9 @@ public interface HdfsClientConfigKeys {
String DFS_PROVIDED_ALIASMAP_INMEMORY_RPC_ADDRESS =
"dfs.provided.aliasmap.inmemory.dnrpc-address";
String DFS_LEASE_HARDLIMIT_KEY = "dfs.namenode.lease-hard-limit-sec";
long DFS_LEASE_HARDLIMIT_DEFAULT = 20 * 60;
/**
* These are deprecated config keys to client code.
*/

View File

@ -151,6 +151,7 @@ public class DfsClientConf {
private final boolean dataTransferTcpNoDelay;
private final boolean deadNodeDetectionEnabled;
private final long leaseHardLimitPeriod;
public DfsClientConf(Configuration conf) {
// The hdfsTimeout is currently the same as the ipc timeout
@ -285,6 +286,10 @@ public class DfsClientConf {
HdfsClientConfigKeys.StripedRead.THREADPOOL_SIZE_KEY +
" must be greater than 0.");
replicaAccessorBuilderClasses = loadReplicaAccessorBuilderClasses(conf);
leaseHardLimitPeriod =
conf.getLong(HdfsClientConfigKeys.DFS_LEASE_HARDLIMIT_KEY,
HdfsClientConfigKeys.DFS_LEASE_HARDLIMIT_DEFAULT) * 1000;
}
@SuppressWarnings("unchecked")
@ -618,6 +623,13 @@ public class DfsClientConf {
return deadNodeDetectionEnabled;
}
/**
* @return the leaseHardLimitPeriod
*/
public long getleaseHardLimitPeriod() {
return leaseHardLimitPeriod;
}
/**
* @return the replicaAccessorBuilderClasses
*/

View File

@ -116,17 +116,6 @@ public final class HdfsConstants {
* lease, another client can preempt the lease.
*/
public static final long LEASE_SOFTLIMIT_PERIOD = 60 * 1000;
/**
* For a HDFS client to write to a file, a lease is granted; During the lease
* period, no other client can write to the file. The writing client can
* periodically renew the lease. When the file is closed, the lease is
* revoked. The lease duration is bound by a
* {@link HdfsConstants#LEASE_SOFTLIMIT_PERIOD soft limit} and this hard
* limit. If after the hard limit expires and the client has failed to renew
* the lease, HDFS assumes that the client has quit and will automatically
* close the file on behalf of the writer, and recover the lease.
*/
public static final long LEASE_HARDLIMIT_PERIOD = 60 * LEASE_SOFTLIMIT_PERIOD;
// SafeMode actions
public enum SafeModeAction {

View File

@ -1741,4 +1741,10 @@ public class DFSConfigKeys extends CommonConfigurationKeys {
@Deprecated
public static final long DFS_CLIENT_KEY_PROVIDER_CACHE_EXPIRY_DEFAULT =
HdfsClientConfigKeys.DFS_CLIENT_KEY_PROVIDER_CACHE_EXPIRY_DEFAULT;
public static final String DFS_LEASE_HARDLIMIT_KEY =
HdfsClientConfigKeys.DFS_LEASE_HARDLIMIT_KEY;
public static final long DFS_LEASE_HARDLIMIT_DEFAULT =
HdfsClientConfigKeys.DFS_LEASE_HARDLIMIT_DEFAULT;
}

View File

@ -41,7 +41,9 @@ import com.google.common.collect.Lists;
import org.apache.commons.lang3.StringUtils;
import org.apache.hadoop.classification.InterfaceAudience;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.BatchedRemoteIterator.BatchedListEntries;
import org.apache.hadoop.hdfs.DFSConfigKeys;
import org.apache.hadoop.hdfs.DFSUtil;
import org.apache.hadoop.hdfs.protocol.HdfsConstants;
import org.apache.hadoop.hdfs.protocol.OpenFileEntry;
@ -84,7 +86,7 @@ public class LeaseManager {
.getName());
private final FSNamesystem fsnamesystem;
private long softLimit = HdfsConstants.LEASE_SOFTLIMIT_PERIOD;
private long hardLimit = HdfsConstants.LEASE_HARDLIMIT_PERIOD;
private long hardLimit;
static final int INODE_FILTER_WORKER_COUNT_MAX = 4;
static final int INODE_FILTER_WORKER_TASK_MIN = 512;
private long lastHolderUpdateTime;
@ -112,7 +114,10 @@ public class LeaseManager {
private volatile boolean shouldRunMonitor;
LeaseManager(FSNamesystem fsnamesystem) {
Configuration conf = new Configuration();
this.fsnamesystem = fsnamesystem;
this.hardLimit = conf.getLong(DFSConfigKeys.DFS_LEASE_HARDLIMIT_KEY,
DFSConfigKeys.DFS_LEASE_HARDLIMIT_DEFAULT) * 1000;
updateInternalLeaseHolder();
}

View File

@ -5745,4 +5745,12 @@
octal or symbolic.
</description>
</property>
<property>
<name>dfs.namenode.lease-hard-limit-sec</name>
<value>1200</value>
<description>
Determines the namenode automatic lease recovery interval in seconds.
</description>
</property>
</configuration>

View File

@ -40,7 +40,6 @@ import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.hdfs.client.HdfsClientConfigKeys;
import org.apache.hadoop.hdfs.protocol.DatanodeInfo;
import org.apache.hadoop.hdfs.protocol.HdfsConstants;
import org.apache.hadoop.hdfs.protocol.LocatedBlocks;
import org.apache.hadoop.hdfs.server.datanode.DataNode;
import org.apache.hadoop.hdfs.server.namenode.FSDirectory;
@ -107,7 +106,9 @@ public class TestFileAppend4 {
// set the soft limit to be 1 second so that the
// namenode triggers lease recovery upon append request
cluster.setLeasePeriod(1000, HdfsConstants.LEASE_HARDLIMIT_PERIOD);
cluster.setLeasePeriod(1,
conf.getLong(DFSConfigKeys.DFS_LEASE_HARDLIMIT_KEY,
DFSConfigKeys.DFS_LEASE_HARDLIMIT_DEFAULT));
// Trying recovery
int tries = 60;

View File

@ -111,9 +111,10 @@ public class TestLease {
Assert.fail("Write failed.");
}
long hardlimit = conf.getLong(DFSConfigKeys.DFS_LEASE_HARDLIMIT_KEY,
DFSConfigKeys.DFS_LEASE_HARDLIMIT_DEFAULT) * 1000;
// make it look like the hard limit has been exceeded.
dfs.lastLeaseRenewal = Time.monotonicNow()
- HdfsConstants.LEASE_HARDLIMIT_PERIOD - 1000;
dfs.lastLeaseRenewal = Time.monotonicNow() - hardlimit - 1000;
dfs.renewLease();
// this should not work.

View File

@ -407,10 +407,10 @@ public class TestLeaseRecovery2 {
Map<String, String []> u2g_map = new HashMap<String, String []>(1);
u2g_map.put(fakeUsername, new String[] {fakeGroup});
DFSTestUtil.updateConfWithFakeGroupMapping(conf, u2g_map);
long hardlimit = conf.getLong(DFSConfigKeys.DFS_LEASE_HARDLIMIT_KEY,
DFSConfigKeys.DFS_LEASE_HARDLIMIT_DEFAULT) * 1000;
// Reset default lease periods
cluster.setLeasePeriod(HdfsConstants.LEASE_SOFTLIMIT_PERIOD,
HdfsConstants.LEASE_HARDLIMIT_PERIOD);
cluster.setLeasePeriod(HdfsConstants.LEASE_SOFTLIMIT_PERIOD, hardlimit);
//create a file
// create a random file name
String filestr = "/foo" + AppendTestUtil.nextInt();

View File

@ -659,7 +659,8 @@ public class TestFileTruncate {
NameNodeAdapter.getLeaseManager(cluster.getNamesystem())
.setLeasePeriod(HdfsConstants.LEASE_SOFTLIMIT_PERIOD,
HdfsConstants.LEASE_HARDLIMIT_PERIOD);
conf.getLong(DFSConfigKeys.DFS_LEASE_HARDLIMIT_KEY,
DFSConfigKeys.DFS_LEASE_HARDLIMIT_DEFAULT) * 1000);
checkFullFile(p, newLength, contents);
fs.delete(p, false);

View File

@ -304,7 +304,8 @@ public class TestINodeFileUnderConstructionWithSnapshot {
}
} finally {
NameNodeAdapter.setLeasePeriod(fsn, HdfsConstants.LEASE_SOFTLIMIT_PERIOD,
HdfsConstants.LEASE_HARDLIMIT_PERIOD);
conf.getLong(DFSConfigKeys.DFS_LEASE_HARDLIMIT_KEY,
DFSConfigKeys.DFS_LEASE_HARDLIMIT_DEFAULT) * 1000);
}
}
}