Hi,

Can someone suggest the best way to intercept Hibernate search indexing process. Consider a scenario like this:
I have a Hibernate/JPA entity which can be indexed i.e. it is annotated with @Indexed. I have a business requirement that whenever this entity is created/updated/deleted, a set of business rules will be invoked, which will decide whether that entity should be indexed or not. Kindly suggest how to tackle this situation?

Thanks,
Deepa.

there are some solutions:

1) index a boolean flag so you can filter results on it, hiding results marked as "not public". This boolean flag in the entity could contain some code. The drawback is that you get bigger indexes, have some additional cpu consumption indexing stuff you don't want, and possibly as a design having this boolean getter in your model is not very clean.
The good thing is that you can later change your mind and have a special search which can include the hidden results.

2) replace hibernate search's event listener with your own

3) propose a better alternative we could integrate in Search; patches welcome :-) I am thinking of a strategy-parameter to be added to the main annotation

@Indexed(name="myindex",exclusionSelector=myStrategy.class)

Implementing a validator-like interface, and the default strategy would be something returning true at alla times.

Sanne,

Thanks for the reply. I thought of trying out the second solution that you mentioned ie writing a custom index event listener instead of using [code]FullTextIndexEventListener[/code]. I have added all event properties to persistence.xml and pointed it to a custom listener class as follows:
<property name="hibernate.ejb.event.post-update"
value="search.event.MyIndexEventListener"/>
[code]processWork()[/code] seems to be the starting point for initiating indexing. For testing purpose I commented out this piece of code, but still indexing is happening. I would appreciate if you could help me to solve this issue?


Thanks,
Deepa

I made the following changes in the configuration file and code:

1. Sethibernate.search.autoregister_listeners = false in persistence.xml
2. Configured events and mapped them to a custom listener class. Eg:
<property name="hibernate.ejb.event.post-update"
value="lucene.search.MyIndexEventListener"/>
3. Written a new custom listener class by extending FullTextIndexEventListener

public class MyIndexEventListener extends FullTextIndexEventListener {

@Override
public void onPostInsert(PostInsertEvent postInsertEvent) {
if(xxxx) return;//add appropriate logic here
super.onPostInsert(postInsertEvent);
}
}

This way I am able to write necessary logic in the event handling methods and control auto indexing. Please suggest if this is the right approach.

Thanks,
Deepa

Hi,
yes I think that should to the trick.
The problem is that in future versions the listener could be declared "final", so I think we should integrate a fetaure like this in the main code.

Are you going to take special care about removing Documents from the index for which the "business rule" doesn't apply anymore?

I have a similar case for this, essentially I have some entities that would provide no value if indexed so would like to be able to provide a class that would determine if the Entity should be indexed or not. Is this something that is still under consideration?

It's not currently on the roadmap. IMHO it should be a good feature, but this could raise some discussions on the design of your model, and Emmanuel is striving to keep the stuff clean and simple.

Usually a filter is good enough, and people making queries don't get confused.

If you have a brilliant use case for it, that would be a good reason to implement it. Could you explain the case?

Sure, basically our system receives various messages and is responsible for "normalizing" those messages to a standard format. All of the normalized messages make sense to index and some of the "unnormalized" messages make sense to index as well, but there are some instances where the information that is within the message has no value added when indexed (i.e. the message is some non human readable message consisting of arrays of numbers and literally no text).

I will take a look at Filters as I have not yet come across these, is there a documentation page that you could point me at?

How do you plan to separate the valuable information from the numbers and codes? If you're going for text processing, you should IMHO define your own Analyzer to process your messages. That's were Analyzers were designed for, and is an efficient and flexible solution. So in that case you actually don't want to avoid indexing.

the reference is speaking about the filters:
http://www.hibernate.org/hib_docs/search/reference/en/html_single/#d0e2950
Also the book as a full chapter about them and the best practices about it.

Thanks I will definitely take a look at writing a custom analyzer, I have not yet had a chance to do any research into this but I am sure there is a bit of documentation.