thinkaurelius / faunus Goto Github PK

g.V.groupCount(Vertex,"{it.age}")

Add configuration options for username/password.

We might also be able to accomodate that through Rexster.

...I like Faunus semantics as its a "hop" not a "drop." Think about which model to resolve to.

Keep faunus from getting banged up by line wrapping on consoles with lower width. The faunus tag line tends to break him up...maybe move it above or below the art.

...so that data can be loaded with filters.

gremlin.bat is basically identical to other gremlin.bat files save we now have -i parameter.

If the v.estimate is set to -1 then do a g.V.count() to get the vertex count for a proper split.

Separate the configuration from the job propagation?

As it stands, there are three representations of the same information in memory at every record parse:

1. the String of the record (up to \n)
2. the JSONObject representation of the String
3. the FaunusVertex representation of the JSONObject

If we could go from String -> FaunsVertex, that would be good. However, the best would be InputStreamReader -> FaunusVertex.

...also, make use of Titan edge compression techniques.

v[0] 10
v[1] 1
v[2] 4

....with ability to control the toString().

public void initialize(final InputSplit genericSplit, final TaskAttemptContext context) throws IOException {
final FileSplit split = (FileSplit) genericSplit;
final Configuration job = context.getConfiguration();
//this.maxLineLength = job.getInt("mapred.linerecordreader.maxlength", Integer.MAX_VALUE);

char direction.

so hadoop properties percolate?

Each Map is parsing some file of data and inserting it into a different machine in Titan.

http://chasebradford.wordpress.com/category/hadoop/

Make the Writable and in-memory representation more efficient.

In the JSONRecordReader/Writer use GraphSONFactory for graph Element serialization/deserialization:

https://github.com/tinkerpop/blueprints/blob/master/blueprints-core/src/main/java/com/tinkerpop/blueprints/util/io/graphson/GraphSONFactory.java#L41

http://vangjee.wordpress.com/2012/02/29/computing-pearson-correlation-using-hadoops-mapreduce-mr-paradigm/

...then submit all those to global JobConf.

(author -> paper <- author) -> coauthor

...can lead to multiple coauthor relationships between two people. As such, allow traverse to have a "weight" or "count" argument so that same edges are grouped and property weighted.

traverse(OUT,"author",IN,"author","coauthor",DROP,"weight")

Still faster than serializing to disk?

solve the super node problem for a vertex (can't always assume we can hold in memory)

FaunusVertexRelationLoader loader = new FaunusVertexRelationLoader(IDHandler.getKeyID(row.left.duplicate()));

public FaunusVertexRelationLoader(final long id) {
    Preconditions.checkArgument(id > 0);
    vertex = new FaunusVertex(id);
}

https://svn.apache.org/repos/asf/whirr/tags/release-0.7.0/services/hadoop/src/main/resources/whirr-hadoop-default.properties

http://chasebradford.wordpress.com/category/hadoop/

Would be serial, on the initial load, but once in Hadoop, all operations are parallel.

As Map+Reduce?Map* is the general solution for in-memory mapping.

try/catch logger in GraphSONRecordReader.