[hibernate-dev] [Search] proposing an alternative to depth in @IndexedEmbedded

Zach Kurey pushedbytime at gmail.com
Thu Aug 25 11:47:30 EDT 2011


On Aug 25, 2011, at 2:28 AM, Hardy Ferentschik wrote:

> Or just 'include' and 'exclude'.
> I feel we are becoming overly verbose in the API design (which is besides  
> this
> particular issue)

That doesn't seem clear to me.  If its just include/exclude, what exactly is being included/excluded?  I think 'paths' has to be part of it, which is basically where we started.  But there is a desire to have exclude vs. include, so that became includePaths, excludePaths.  But then it seemed ambiguous if the user needed to specify the attribute name itself that the annotation is on(in the examples 'see'), so 'sub' paths was added to make it clearer.  But includePaths/excludePaths seems fine as well, since again I think its clear the property the annotation is declared on doesn't need to be repeated in the path.

@IndexEmbedded(includePaths={"see.a.b.c"}) // redundant, prefer includePaths={"a.b.c"}
private SomeType see;

> My experience that is is not a good idea to change implicitly
> ignore default values. In this case I would rather see an additional
> enum parameter which allows the user to explicitly select the embedding
> mode (BY_DEPTH, BY_PATH) or maybe introduce a new annotation @IndexSubPath

The mode option is confusing if the exclude option is also in the mix, because 'exclude' applies when when the default depth approach is used.  With separate annotations how do you envision supporting the exclude option?  Would it be like this set of options:

Two new annotations:
@IndexEmbedded(depth = N)
@ExcludePaths(paths={...})
private SomeType someProperty;

or
@IndexEmbedded(depth = N)
private SomeType someProperty;

or
@IndexSubPath(paths={...})
private SomeType someProperty

Or would it be one new 'exclude' attribute in IndexEmbedded and one new annotation for IndexSubPaths:

@IndexEmbedded(depth = N, excludePaths={...})
private SomeType someProperty;

or
@IndexEmbedded(depth = N)
private SomeType someProperty;

or
@IndexSubPaths(paths={...})
private SomeType someProperty

If 'includeSubPaths' is specified as an attribute of IndexEmbedded, it seems really clear to me that depth no longer applies since you are specifying specific paths that clearly have a depth.  I also think its cleaner, than the combination of attributes + annotations above.  But regardless, any of the approaches above would achieve the same end.  Maybe splitting off into another annotation gives you more configuration flexibility in the future for dealing with specific paths that I can't think of at the moment.

> One thing worth mentioning is that generally the index size is not a big  
> problem
> 

It's not the size of the resulting index that has been problematic.  It's the expense to re-write a large document by having to recursively traverse down 'depth' at every IndexEmbedded property.  It adds to DB load.  It adds CPU overhead on whatever system is doing the indexing(which can be very high when there are a lot of unneeded fields getting put into the documents), and it can significantly slow down large transactions(even in async mode since that consumes CPU and other resources).  If the CPU load gets significant enough then the user apps have to offload onto other servers to do the indexing and copying slave copies of the indexes around a cluster, which adds to operational complexity.  I could go on.  = )

I'd just prefer that Hibernate Search be as light weight as possible, and thats hard to do with depth + a complex object model.

> and we used to say that from a query point of view you have more  
> flexibility
> if all properties are indexed (compared to limiting yourself already at  
> index
> time to what you can search).

For more dynamic search apps, I get that perspective.  But in our case we know at app dev time what fields we need to be indexed to satisfy our searches.  Requirements on what we need to search on don't change post a particular release, and certainly not at runtime.

Zach


More information about the hibernate-dev mailing list