HDFS-10326. Disable setting tcp socket send/receive buffers for write pipelines. Contributed by Daryn Sharp.

This commit is contained in:
Haohui Mai 2017-08-08 14:58:11 -07:00
parent e0c24145d2
commit 71b8dda4f6
3 changed files with 20 additions and 13 deletions

View File

@ -48,8 +48,8 @@ public final class HdfsConstants {
public static final byte COLD_STORAGE_POLICY_ID = 2;
public static final String COLD_STORAGE_POLICY_NAME = "COLD";
// TODO should be conf injected?
public static final int DEFAULT_DATA_SOCKET_SIZE = 128 * 1024;
public static final int DEFAULT_DATA_SOCKET_SIZE = 0;
/**
* A special path component contained in the path for a snapshot file/dir
*/

View File

@ -2545,13 +2545,14 @@
<property>
<name>dfs.client.socket.send.buffer.size</name>
<value>131072</value>
<value>0</value>
<description>
Socket send buffer size for a write pipeline in DFSClient side.
This may affect TCP connection throughput.
If it is set to zero or negative value,
no buffer size will be set explicitly,
thus enable tcp auto-tuning on some system.
The default value is 0.
</description>
</property>
@ -3025,23 +3026,25 @@
<property>
<name>dfs.datanode.transfer.socket.send.buffer.size</name>
<value>131072</value>
<value>0</value>
<description>
Socket send buffer size for DataXceiver (mirroring packets to downstream
in pipeline). This may affect TCP connection throughput.
If it is set to zero or negative value, no buffer size will be set
explicitly, thus enable tcp auto-tuning on some system.
The default value is 0.
</description>
</property>
<property>
<name>dfs.datanode.transfer.socket.recv.buffer.size</name>
<value>131072</value>
<value>0</value>
<description>
Socket receive buffer size for DataXceiver (receiving packets from client
during block writing). This may affect TCP connection throughput.
If it is set to zero or negative value, no buffer size will be set
explicitly, thus enable tcp auto-tuning on some system.
The default value is 0.
</description>
</property>

View File

@ -30,7 +30,6 @@
import java.io.IOException;
import java.net.Socket;
import static org.apache.hadoop.hdfs.client.HdfsClientConfigKeys.DFS_CLIENT_SOCKET_SEND_BUFFER_SIZE_DEFAULT;
import static org.apache.hadoop.hdfs.client.HdfsClientConfigKeys.DFS_CLIENT_SOCKET_SEND_BUFFER_SIZE_KEY;
import static org.junit.Assert.assertTrue;
@ -42,15 +41,16 @@ public class TestDFSClientSocketSize {
}
/**
* The setting of socket send buffer size in
* {@link java.net.Socket#setSendBufferSize(int)} is only a hint. Actual
* value may differ. We just sanity check that it is somewhere close.
* Test that the send buffer size default value is 0, in which case the socket
* will use a TCP auto-tuned value.
*/
@Test
public void testDefaultSendBufferSize() throws IOException {
assertTrue("Send buffer size should be somewhere near default.",
getSendBufferSize(new Configuration()) >=
DFS_CLIENT_SOCKET_SEND_BUFFER_SIZE_DEFAULT / 2);
final int sendBufferSize = getSendBufferSize(new Configuration());
LOG.info("If not specified, the auto tuned send buffer size is: {}",
sendBufferSize);
assertTrue("Send buffer size should be non-negative value which is " +
"determined by system (kernel).", sendBufferSize > 0);
}
/**
@ -73,6 +73,10 @@ public void testSpecifiedSendBufferSize() throws IOException {
sendBufferSize1 > sendBufferSize2);
}
/**
* Test that if the send buffer size is 0, the socket will use a TCP
* auto-tuned value.
*/
@Test
public void testAutoTuningSendBufferSize() throws IOException {
final Configuration conf = new Configuration();
@ -80,7 +84,7 @@ public void testAutoTuningSendBufferSize() throws IOException {
final int sendBufferSize = getSendBufferSize(conf);
LOG.info("The auto tuned send buffer size is: {}", sendBufferSize);
assertTrue("Send buffer size should be non-negative value which is " +
"determined by system (kernel).", sendBufferSize > 0);
"determined by system (kernel).", sendBufferSize > 0);
}
private int getSendBufferSize(Configuration conf) throws IOException {