Debugging “Wrong FS expected: file:///” exception from HDFS

inShare

The Performance Zone is supported by New Relic and AppDynamics. Both are leaders in the APM space with high-profile customers and massive cost reductions for those users.

I just spent some time putting together some basic Java code to read some data from HDFS. Pretty basic stuff. No map reduce involved. Pretty boilerplate code like the stuff from this popular tutorial on the topic.

No matter what, I kept hitting my head on this error:

Exception in thread “main” java.lang.IllegalArgumentException: Wrong FS: hdfs://localhost:9000/user/hadoop/DOUG_SVD/out.txt, expected: file:///

If you checkout the tutorial above, what’s supposed to be happening is that an instance of Hadoop’s Configuration should encounter a fs.default.name property, in one of the config files its given. The Configuration should realize that this property has a value of hdfs://localhost:9000. When you use the Configuration to create a Hadoop FileSystem instance, it should happily read this property from Configuration and process paths from HDFS. That’s a long way of saying these three lines of Java code:

view source print ?

1. // pickup config files off classpath

2. Configuration conf = new Configuration()

3. // explicitely add other config files

4. conf.addResource("/home/hadoop/conf/core-site.xml");

5. // create a FileSystem object needed to load file resources

6. FileSystem fs = FileSystem.get(conf);

7. // load files and stuff below!

Well… My Hadoop config files (core-site.xml) appear setup correctly. It appears to be in my CLASSPATH. I’m even trying to explicitly add the resource. Basically I’ve followed all the troubleshooting tips you’re supposed to follow when you encounter this exception. But I’m STILL getting this exception. Head meet wall. This has to be something stupid.

Troubleshooting Hadoop’s Configuration & FileSystem Objects

Well before I reveal my dumb mistake in the above code, it turns out there’s some helpful functions to help debug these kind of problems:

As Configuration is just a bunch of key/value pairs from a set of resources, its useful to know what resources it thinks it loaded and what properties it thinks it loaded from those files.

• getRaw() — return the raw value for a configuration item (like conf.getRaw("fs.default.name"))
• toString() — Configuration‘s toString shows the resources loaded

You can similarly checkout FileSystem‘s helpful toString method. It nicely lays out where it thinks its pointing (native vs HDFS vs S3 etc).

So if you similarly are looking for a stupid mistake like I was, pepper your code with printouts of these bits of info. They will at least point you in a new direction to search for your dumb mistake.

Drumroll Please

Turns out I missed the crucial step of passing a Path object not a String to addResource. They appear to do slightly different things. Adding a String adds a resource relative to the classpath. Adding a Path is used to add a resource at an absolute location and does not consider the classpath. So to explicitly load the correct config file, the code above gets turned into (drumroll please):

view source print ?

1. // pickup config files off classpath

2. Configuration conf = new Configuration()

3. // explicitely add other config files

4. // PASS A PATH NOT A STRING!

5. conf.addResource(new Path("/home/hadoop/conf/core-site.xml"));

6. FileSystem fs = FileSystem.get(conf);

7. // load files and stuff below!

Then Tada! everything magically works! Hopefully these tips can save you the next time you encounter these kinds of problems.