Hi, 

We are trying to access HDFS data in Spark local mode for D-Stream testing.
The testing code is obtained by simply replacing the input source of
examples/src/main/java/org/apache/spark/streaming/examples/JavaNetworkWordCo
unt.java 
to an HDFS path.

After building the package with maven, we used the following command to
execute the package:

./run-example org.apache.spark.streaming.examples.JavaHdfsWordCount local[2]
hdfs://master:54310/path-to-stream-data

But we get the following error:
Exception in thread "Thread-30" java.io.IOException: Failed on local
exception: java.io.EOFException; Host Details : local host is:
"zyz-1/127.0.0.1"; destination host is: ""master":54310;

The error message looks straightforward for the reason behind, but we
goolged it with no fortune. Is this an obvious wrong way to run the code?

The consistency of hadoop-client version is confirmed(both the version of
hadoop cluster and client are 0.20.2-cdh3u6). The package is run in a node
with hadoop client(0.20.2-cdh3u6) installed and the access to HDFS by
`hadoop dfs ­ls' is OK.

Any explanations and suggestions are appreciated.

Thanks in advance, 
Jianmin


Reply via email to