Indexing Framework

An indexing framework for iRODS based middleware

Quickstart

1. Install Fabric
   sudo apt-get install fabric
2. Download install script wget https://raw.githubusercontent.com/DICE-UNC/indexing-irods/master/fabfile.py
3. Install it fab install
4. Start using fab start_all_services

Overview

The indexing framework comes in an OSGI bundle that provides two services

databook.listener.service.MessagingService

and

databook.listener.service.IndexingService

The former is used to interface with messaging services such as AMQP 0.9, AMQP 1.0, etc. This can be done by defining a Camel route.

The latter is used to interface with indexers. IndexingService provides methods for an indexer to register and unregister. When an indexer is registered to the IndexingService, the IndexingService will relay all messages that it receives to the indexer in the form of POJO. The indexer also acquires a reference to a scheduler automatically which allows the indexer to submit time-consuming tasks to be scheduled. Currently, the supported tasks are those which require access to the underlying data grid. Each task is composed of an object indicating which data object it needs to access and a continuation object which will be called by the scheduler when the data object and system resources become available. A simple scheduler based on a thread pool model is included. More complex schedulers can be implemented by implementing the

databook.listener.Scheduler

interface.

Installation

• Install Apache ServiceMix 5.0.0

• Checkout indexing framework source from github

• Run

    mvn install
  

  This will generate an OSGI bundle in the maven project target directory

• Deploy generated target to ServiceMix 5.0.0

Message Format

The message format is defined in JSON Schema. Look for the schema directory in the source tree for the schemas. The jsonschema2pojo plugin is used in the build process to generate Java classes corresponding to the schemas. Jackson is used to translate between JSON messages and Java objects. Indexers do not need to parse JSON, the messages are passed into indexers as Java objects, and fields of JSON can be accessed from fields of the Java objects.

All messages have the follow format:

{
    messages: [ ... ]
}

And it is mapped to objects of type Messages in Java. The Indexer interface provides a "messages" method which take in a Messages object. This is the method called on each indexer by the indexing framework for each incoming set of messages.

Each message has the following format

{
    operation: ...,
    hasPart: [ ... ]
}

The operation property may have the one of the following:

create, delete, modify, union, diff, retrieve

The hasPart property may have any number of objects. The first five operation are used by the incoming messages from the indexing framework. The last one is used by indexer for sending messages to the Scheduler. Their semantics are as follows:

• create(o): create an object.
• delete(o): delete an object.
• modify(o0, o1): modify an object identified by o0.uri from o0 by o1. For each property p,

if o0.p == null and o1.p == null then do nothing;

if o0.p == null and o1.p != null then modify whatever value p had to o1.p;

if o0.p != null and o1.p == null then delete whatever value o0.p had;

if o0.p != null and o1.p != null then modify whatever value o0.p had to o1.p.

The intuition is that o0 provides uri for the object to be modified and act as an indicator object for deletion of properties.

• union(o0, o1): similar to modify, o0 provides uri. For each non array property p, if o1.p != null and the value p had was empty then set value of p to o1.p. For each array property p, set the value of p to the value it had union o1.p.
• diff(o0, o1): similar to modify, o0 provides uri. For each non array property p, if o1.p != null and the value p had equals o1.p then set value of p to empty. For each array property p, set the value of p to the value it had diff o1.p.
• retrieve: retrieve the objects into a cache resource for indexing.

Configuring ServiceMix with Message Routing

The indexing framework does not require a specific messaging protocol. It can be configured using a Apache Camel route.

AMQP 1.0

Configuring iRODS

Enable the databook.re rule base. In iRODS, edit server/config/server.config.

Add "databook" the list of rule sets.

To connect the rules with PEPs, there are two options:

(1) If there are no user defined rules in core.re, the default PEP rules can be used. To use the default PEP rules, simplely add "databook_pep" before "core". This works for default installations where the default rules are empty. If there are user defined PEP rules in "core", then they will be overriden by the databook PEP rules. In this case the databook.re rules need to be added to user defined PEP rules.

(2) The rule in databook.re to call from user defined PEPs have the following naming convention: the "ac" prefix is omitted. For example, from

acPostProcForPut

to

postProcForPut

Add call to the databook rules from user defined PEP rules

For example, if you have a rule

acPostProcForPut {
    ...
}

then add call to data book rules

acPostProcForPut {
    ...
    postProcForPut
}

Indexers

ElasticSearch

The ElasticSearch indexer comes in a seperate OSGI bundle.

VIVO

The VIVO indexer comes in a seperate OSGI bundle.

Advanced Topics

Composing JSON Messages Using the Java DSL

The Java DSL for composing JSON Message is generated from an EriLex specification. This DSL allows very flexible composition of JSON Messages against the schemas. Some examples:

.object()
    .key("key1").value("value1")
    .key("key2").value("value2")
    .key("key3").value("value3")
.end()

and

.object()
    .key("key1").array()
        .elem("elem1")
        .elem("elem2")
        .elem("elem3")
    .end()
.end()

and

.object()
    .key("key1").array()
        .object("elem1")
            .key("key1").value("key1")
        .end()
        .object("elem2")
            .key("key2").value("key2")
        .end()
        .object("elem3")
            .key("key3").value("key3")
        .end()
    .end()
    .key("key2").value("value2")
    .key("key3").value("value3")
.end()

Objects and arrays can be nested arbitrarily.

Rule-based Mapping of Java Objects to Indexes

Some indexers, such as triple store or graph database based indexers, have to serialize Java objects into textual representation in a specific format. This is supported by Rule-base mapping in the indexing framework.

For example, suppose we have a JSON object

{
    "type": "DataObject"
    "label": "DataObject1"
    "title": "DataObject2"
}

But we want to index the "label" field and the "title" field differently, then we can define two rules. The rules match fields in a similar way Java resolves ad hoc polymorphism. This is possible because in the JSON schemas subtype hierachies are defined.

The rules are defined by implementing the ObjectPropertyRule interface. There are five methods to implement:

create, delete, modify, union, diff

Their semantics are as follows: For non-array properties:

• create(v): Set the value of a previous undefined property
• delete(v): Reset the value of a property, making it undefined
• modify(v): Change the value of a property, if the new value is null, then the property becomes undefined
• union(v): If the property is undefined, then set it to v
• diff(v): If the property is defined and equal to v, then reset the property, making it undefined

For array properties:

We do not distinguish between undefined and empty array.

• create(v): Set the value of a previous undefined property
• delete(v): Reset the value of a property, making it undefined
• modify(v): Changes the value of a property, if the new value is null, then the property becomes undefined
• union(v): Set the property to the union of its current value and v
• diff(v): Set the property to the difference of its current value and v

The defined rules can be registered to a RuleSet object.