chearch

Chearch is a simple search engine written in Cray's Chapel language.

This application demonstrates how to use various important features of Chapel, such as locales, and how to minimize RPC traffic through features such as local. It also shows how to build a simple, efficient, inverted index using only integer represesentions.

Project link to this page: http://chearch.pw (church pew)

Features of the search engine

lock-free, using atomic operations for all appropriate operations
string-free, the entire engine is integer-based. This minimizes memory footprint while improves processing speed
boolean queries, using an integer-based (no strings, remember?) query language called CHASM (Chearch Assembly)
document-based hash partitioning
distributed loads of indexes from storage (in-progress)
- async queue indexer
- batch load indexer
online query and indexing support via libev-backed TCP connection (in-progress)
support for in-memory and on-disk (future) index segments
parallel scatter-gather across partitions using native Chapel forall support

Performance

Of course, one of the main motivations for all of this is performance. While tuning and profiling are ongoing, initial results indicate:

single locale queries takes 10s of microseconds
cross-locale queries (scatter-gather) can be milliseconds

Storage

The current implementation is configured to store Tweets: Each segment can hold 16M documents and the max document length is 256 characters. One full memory segment should be about 1GB RAM.

Sample

A simple example (from test/helloworld.chpl) which indexes a document (id 10) with two terms: 2 and 3

use SearchIndex;

proc main() {

    writeln("initialize search index");

    initPartitions();

    writeln("add document id 10 with terms 2 and 3");

    {
        var terms: [0..1] IndexTerm;
        terms[0].term = 2;
        terms[0].textLocation = 6;
        terms[1].term = 3;
        terms[1].textLocation = 15;
        addDocument(terms, 10);
    }

    {
        var terms: [0..1] IndexTerm;
        terms[0].term = 2;
        terms[0].textLocation = 6;
        addDocument(terms, 15);
    }

    // create CHASM instruction buffer
    var buffer = new InstructionBuffer(1024);
    var writer = new InstructionWriter(buffer);

    // write the CHASM code to implement the query
    writeln("querying for term IDs 2");
    writer.write_push_term(2);
    forall result in query(new Query(buffer)) {
        writeln(result);
    }

    writeln("querying for term IDs 3");
    buffer.clear();
    writer.write_push_term(3);
    forall result in query(new Query(buffer)) {
        writeln(result);
    }

    writeln("querying for term IDs 2 OR 3");
    buffer.clear();
    writer.write_push_term(2);
    writer.write_push_term(3);
    writer.write_or();
    forall result in query(new Query(buffer)) {
        writeln(result);
    }

    writeln("querying for term IDs 2 AND 3");
    buffer.clear();
    writer.write_push_term(2);
    writer.write_push_term(3);
    writer.write_and();
    forall result in query(new Query(buffer)) {
        writeln(result);
    }

    delete buffer;
}

Output

initialize search index
add document id 10 with terms 2 and 3
querying for term IDs 2
(term = 2, textLocation = 6, externalDocId = 15)
(term = 2, textLocation = 6, externalDocId = 10)
querying for term IDs 3
(term = 3, textLocation = 15, externalDocId = 10)
querying for term IDs 2 OR 3
(term = 2, textLocation = 6, externalDocId = 15)
(term = 3, textLocation = 15, externalDocId = 10)
(term = 2, textLocation = 6, externalDocId = 10)
querying for term IDs 2 AND 3
(term = 3, textLocation = 15, externalDocId = 10)
(term = 2, textLocation = 6, externalDocId = 10)

SETUP

General setup is expecting an OSX brew installation:

brew install libev

Chapel (http://chapel.cray.com/download.html)

brew install chapel

COMPILING

make

RUN

./bin/chearch

TEST

make chearch_test
./bin/chearch_test

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting