Mailing-List: contact user-help@couchdb.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@couchdb.apache.org
Received-SPF: neutral (athena.apache.org: local policy)
Message-ID: <4CF4E802.8060906@zedeler.dk>
Date: Tue, 30 Nov 2010 13:03:14 +0100
From: Michael Zedeler <michael@zedeler.dk>
User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US;
 rv:1.9.2.12pre) Gecko/20101026 Shredder/3.1.6pre
MIME-Version: 1.0
To: user@couchdb.apache.org
Subject: A possible approach for constructing trees of documents optimized
 for fast querying
Content-Type: multipart/alternative;
 boundary="------------090402070600000807080203"

--------------090402070600000807080203
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Hi everybody.

I have come up with a way to create a tree where each node is a document 
in CouchDB. The tree is ordered.

Maybe someone on the list tried something similar and could comment on 
the approach. For others, this could inspire a new way of using CouchDB.

The performance features are:

    * Fast reads.
    * Slow writes (updates, deletions, moving subtrees).

The basic notion is that every document gets a path attribute that 
specifies where in the tree the document is located.

The root path looks like this:

[0]

(JSON notation.)

A child of the root looks like this:

[0, 0]

Adding another child to the root provides this path:

[0, 32768]

(32768 equals 2**15 and it is somewhat arbitraryly chosen.)

The resulting documents could look like this:

{ '_id': 'root', 'path': [0] }
{ '_id': 'child_1', 'path': [0, 0] }
{ '_id': 'child_2', 'path': [0, 32768] }

Adding a child to child_2 yields the path [0, 32768, 0].

Now querying the whole tree is done using

startkey: [-2**16, 2**16]
endkey: [-2**16, 2**16, {}]

Querying the subtree rooted in child_1:

startkey: [0, 0]
endkey: [0, 0, {}]

I have based this on the collation rules found here:

http://wiki.apache.org/couchdb/View_collation#Complex_keys

The constants -2**16 and 2**16 are just endpoints where you sequentially 
subdivide the interval into smaller segments to make room for more 
siblings. Thus inserting a node between child_1 and child_2 will return 
in the path [0, 32768/2] == [0, 16384]. As far as I understand it, 
JavaScript uses 64 bits to store numbers (they are always floats), so 
there should be ample room for many siblings under each node. At some 
point, it may be needed to re-balance the tree, which will be an 
operation requiring an update of most nodes in the subtree being balanced.

Any comments?

Regards,

Michael.

P. s. demonstration written in perl available on request.


--------------090402070600000807080203--