HDFS-5274 added support for tracing requests through HDFS, using the open source tracing library, Apache HTrace. Setting up tracing is quite simple, however it requires some very minor changes to your client code.
The tracing system works by collecting information in structs called ‘Spans’. It is up to you to choose how you want to receive this information by using implementation of SpanReceiver interface bundled with HTrace or implementing it by yourself.
HTrace provides options such as
In order to set up SpanReceivers for HDFS servers, configure what SpanReceivers you’d like to use by putting a comma separated list of the fully-qualified class name of classes implementing SpanReceiver in hdfs-site.xml property: dfs.htrace.spanreceiver.classes.
<property>
<name>dfs.htrace.spanreceiver.classes</name>
<value>org.apache.htrace.impl.LocalFileSpanReceiver</value>
</property>
<property>
<name>dfs.htrace.local-file-span-receiver.path</name>
<value>/var/log/hadoop/htrace.out</value>
</property>
You can omit package name prefix if you use span receiver bundled with HTrace.
<property>
<name>dfs.htrace.spanreceiver.classes</name>
<value>LocalFileSpanReceiver</value>
</property>
You also need to add the jar bundling SpanReceiver to the classpath of Hadoop on each node. (LocalFileSpanReceiver in the example above is included in the jar of htrace-core which is bundled with Hadoop.)
$ cp htrace-htraced/target/htrace-htraced-3.2.0-incubating.jar $HADOOP_HOME/share/hadoop/common/lib/
You can use hadoop trace command to see and update the tracing configuration of each servers. You must specify IPC server address of namenode or datanode by -host option. You need to run the command against all servers if you want to update the configuration of all servers.
hadoop trace -list shows list of loaded span receivers associated with the id.
$ hadoop trace -list -host 192.168.56.2:9000 ID CLASS 1 org.apache.htrace.impl.LocalFileSpanReceiver $ hadoop trace -list -host 192.168.56.2:50020 ID CLASS 1 org.apache.htrace.impl.LocalFileSpanReceiver
hadoop trace -remove removes span receiver from server. -remove options takes id of span receiver as argument.
$ hadoop trace -remove 1 -host 192.168.56.2:9000 Removed trace span receiver 1
hadoop trace -add adds span receiver to server. You need to specify the class name of span receiver as argument of -class option. You can specify the configuration associated with span receiver by -Ckey=value options.
$ hadoop trace -add -class LocalFileSpanReceiver -Cdfs.htrace.local-file-span-receiver.path=/tmp/htrace.out -host 192.168.56.2:9000 Added trace span receiver 2 with configuration dfs.htrace.local-file-span-receiver.path = /tmp/htrace.out $ hadoop trace -list -host 192.168.56.2:9000 ID CLASS 2 org.apache.htrace.impl.LocalFileSpanReceiver
In order to trace, you will need to wrap the traced logic with tracing span as shown below. When there is running tracing spans, the tracing information is propagated to servers along with RPC requests.
In addition, you need to initialize SpanReceiverHost once per process.
import org.apache.hadoop.hdfs.HdfsConfiguration;
import org.apache.hadoop.tracing.SpanReceiverHost;
import org.apache.htrace.Sampler;
import org.apache.htrace.Trace;
import org.apache.htrace.TraceScope;
...
SpanReceiverHost.getInstance(new HdfsConfiguration());
...
TraceScope ts = Trace.startSpan("Gets", Sampler.ALWAYS);
try {
... // traced logic
} finally {
if (ts != null) ts.close();
}
The TracingFsShell.java shown below is the wrapper of FsShell which start tracing span before invoking HDFS shell command.
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FsShell;
import org.apache.hadoop.hdfs.DFSConfigKeys;
import org.apache.hadoop.hdfs.HdfsConfiguration;
import org.apache.hadoop.tracing.SpanReceiverHost;
import org.apache.hadoop.util.ToolRunner;
import org.apache.htrace.Sampler;
import org.apache.htrace.Trace;
import org.apache.htrace.TraceScope;
public class TracingFsShell {
public static void main(String argv[]) throws Exception {
Configuration conf = new HdfsConfiguration();
FsShell shell = new FsShell();
conf.setQuietMode(false);
shell.setConf(conf);
SpanReceiverHost.get(conf, DFSConfigKeys.DFS_SERVER_HTRACE_PREFIX);
int res = 0;
try (TraceScope ts = Trace.startSpan("FsShell", Sampler.ALWAYS)) {
res = ToolRunner.run(shell, argv);
} finally {
shell.close();
}
System.exit(res);
}
}
You can compile and execute this code as shown below.
$ javac -cp `hadoop classpath` TracingFsShell.java $ java -cp .:`hadoop classpath` TracingFsShell -ls /
The DFSClient can enable tracing internally. This allows you to use HTrace with your client without modifying the client source code.
Configure the span receivers and samplers in hdfs-site.xml by properties dfs.client.htrace.sampler and dfs.client.htrace.sampler. The value of dfs.client.htrace.sampler can be NeverSampler, AlwaysSampler or ProbabilitySampler.
You do not need to enable this if your client program has been modified to use HTrace.
<property>
<name>dfs.client.htrace.spanreceiver.classes</name>
<value>LocalFileSpanReceiver</value>
</property>
<property>
<name>dfs.client.htrace.sampler</name>
<value>ProbabilitySampler</value>
</property>
<property>
<name>dfs.client.htrace.sampler.fraction</name>
<value>0.5</value>
</property>