hadoop mapper static initialisation

Question

Hi,

I have a code fragment in which I am using a static code block to initialize a variable.

public static class JoinMap extends 
            Mapper<IntWritable, MbrWritable, LongWritable, IntWritable> {
        .......
        public static RTree rt = null;
        static {
            String rtreeFileName = "R.rtree";
            rt = new RTree(rtreeFileName);
        }
        public void map(IntWritable key, MbrWritable mbr,Context context)
                throws IOException, InterruptedException {
                            .........               
                List elements = rt.overlaps(mbr.getRect());
                .......

        }

    }

My problem is that the variable rt in the above code fragment is not getting initialised. Can anybody suggest a fix or an alternate way to initialise the variable. I don't want to initialise it inside my map function since that slows down the entire process.

Answer 1

That really isn't possible, unless Java itself is broken. The static initializers always fire at class load time.

Maybe whatever you're observing has an alternate explanation, like, something is setting rt back to null. Or, are you observing rt in other statically-initialized expression above? it would see null until rt's initialization finishes.

But, nevertheless, I'd say it's tidier to override the setup() method (configure() in the old API) and do initialization there. It'll happen once.

Answer 2

Is there some reason you don't want to initialize rt like this?:

public static RTree rt = new RTree("R.rtree");

Answer 3

I'm no Java expert but it look like you have two "static" assignments for the variable rt: You have:

public static RTree rt = null;

AND

rt = new RTree(rtreeFileName);

In which order are these assigments done?

Try this instead and see if it helps

    public static final RTree rt = new RTree("R.rtree");

As far as my Java knowledge goes this assures you'll have only single assignment.