Orzo.js

Orzo.js is a simple multi-threaded, in-memory ETL tool making use of Map-Reduce to distribute processing across multiple CPUs/cores. Tasks are programmed in JavaScript.

Orzo.js comes with a bunch of predefined functions you can use to perform your tasks and analyze your data. You can also create your own or use 3rd party CommonJS modules.

Orzo.js can be run in a server mode which provides:

• REST API
• web interface
• AMQP integration

To get more information plese visit www.orzojs.org.

How to start

Let Orzo.js generate a simple task template:

orzojs -t myscript.js

The command will create a template of your script along with TypeScript d.ts. file for comfortable editing in JavaScript IDE (WebStorm, Visual Studio Code, ...).

How to process data

You do not have to worry about the whole calculation workflow - it is taken care of by the application. All you have to do is to register several obligatory functions:

• dataChunks,
• applyItems,
• map,
• reduce,
• finish.

During each calculation phase, Orzo.js calls your function passing it an appropriate data.

Example

Let's say we have 10 log files (server.0.log, server.1.log, ..., server.9.log). Each line of the log file represents a valid JSON-encoded data.

Splitting data into chunks

We have to decide how we define a chunk. In our case it may be a single file or a list of files. We could define this manually but orzo offers a convenient function for this: orzo.directoryReader. This function will automatically walk through provided directory (or directories), apply optional filter and create a predefined number of path groups.

dataChunks(10, function (idx) {
    return orzo.directoryReader('./data', idx);
});

The code have told Orzo.js that:

1. there will be 10 chunks (i.e. 10 worker threads)
2. once it requires a chunk with ID = idx, it should use a directory reader which is expected to return its idx-th file group.

Now we are going to specify how the map function will be applied on a chunk. Our directoryReader returns an Iterator which we will use in the following code. To be able to process a concrete file, we have to read it somehow. Orzo.js offers a simple line-by-line reader orzo.fileReader:

processChunk(function (fileList, map) {
    var fr; 
    while (fileList.hasNext()) {
        fr = orzo.fileReader(fileList.next()); // next file path in a group
        while (fr.hasNext()) {   
            map(fr.next()); // next line in a file
        }
    }
});

Map-Reduce

Now we can define map and reduce functions. Let's say our JSON log lines contain an errorType attribute (with possible values WARNING, RECOVERABLE, FATAL) and we want to calculate their total occurrences.

map(function (logLine) {
    var parsed = JSON.parse(logLine);
    emit(parsed.errorType, 1); // for each type we emit 1
});

In our case the reduce function will just calculate a sum of all emitted 1s for all the different keys (WARNING, RECOVERABLE, FATAL). Orzo.js reduce function has actually two signatures

• there is an optional first (numeric) parameter which can specify how many worker threads should Orzo.js create. If omitted then the same number as in case of dataChunks is used.

Let's say 2 threads are enought here:

reduce(2, function (key, values) {
    emit(D(values).sum());
});

(Here we have used a special factory function datalib.D which wraps our data can apply some basic statistical functions on it).

Finally, we want to print the result:

finish(function (results) {
    results.sorted.each(function (key, values) {
        // our reduce emits just a single value but Orzo.js always
        // creates an array which means we have to handle 'values'
        // as an array
        orzo.printf('%s -> %d occurrences\n', key, values[0]);
    });    
});

Download

Please visit Orzo.js download page.

You can also build Orzo.js from sources. If you have JDK 8+ and a working Maven installation just type the following commands in Orzo.js root directory:

mvn clean dependency:copy-dependencies package

Further documentation

Please refer to the API page on Orzo's website.