I have a treelike data structure whose leaves are documentversions mapped to a table 'tbl_documentversion' that contains a given versions metadata like 'name', 'type', 'date_of_creation' etc. and the content proper. In order to let the user easily navigate this tree and quickly find the documentversion he/she is interested in I would like to prefetch all data except for the content which is typically 20 - 50 Megs for a single documentversion. I think the performance implications are obvious.

So what I would like to do is tell Hibernate to postpone fetching the content of any given documentversion until my code calls documentVersion.getContent(). This is what I have thought of so far:

1. Using Hibernate's 'lazy' attribute somehow. Unfortunately, the <property> element does not support this.

2. Transfering the content to a separate table 'tbl_content' linked to 'tbl_documentversion' via a one-to-one association. Unfortunately, <one-to-one> does not support the 'lazy' attribute either.

3. Writing a custom proxy class for DocumentVersion. I have no experience in this area and remain in doubt if doing so would allow me to achieve what I am trying to do. Any hints?

4. Writing a lightweight DocumentVersion, say DocumentVersionLight, that does not have the 'content' property, and making DocumentVersion extend DocumentVersionLight. It seems however that Hibernate does not support this since instances of both classes would be mapped to the same row in the same table. Moreover, there wouldn't be any discriminator-column since essentially the lightweight class would rather be an excerpt of the full class than its ancestor.

Currently I am investigating into the possibilities the latest Hibernate 2.1 betas might offer, especially into native SQL queries. But what makes the current solution so attractive is that right now I can get away with something like

Tree data = (Tree) session.load(Tree.class, id);

and Hibernate will walk the tree for me and fetch all nodes. Now if only I could tell Hibernate to stop when it reaches the 'content' property of the 'documentversion' node.

Any help would be greatly appreciated. Thanks,

Olaf

(a) one-to-one may certainly be lazy (use the proxy attribute)
(b) If supported by your driver You should use a JDBC Blob or Clob to represent the 50 meg of data - these are streamed lazily at the driver level
(c) check out the "lightweight class" pattern on the wiki

Thanks for your quick reply.



Thanks for the tip. I will surely do so.