directory-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Emmanuel Lécharny <>
Subject Re: [ApacheDS] how can a partition return more than one result ?
Date Mon, 11 Jun 2012 15:04:57 GMT
Le 6/11/12 3:21 PM, Garbage a écrit :
> Von meinem iPad gesendet
>>> But I am only able to return ONE entry, I didn't find or understand the concept
how MULTIPLE entries can be returned. Can someone show me the right direction ?
>> The idea is to use a Cursor that maps around the partition and fetch the entries
one by one.
>> The way the server works is that based on your filter, you select the right index
to use to fetch the entries. There are may possibilities here :
>> - first, you may have to do a full scan (the filter is not selective enough, for
instance). In this case, you don't use any index, you just use the MasterTable to get the
entries. Now, for each entry you fetch, you'll have to filter them to see if it's a valid
entry - or not.
>> - or you can select an index. You will fetch the index elements, and for each of
them, fetch the associated entry.  Once done, you can check against the filter if the entry
is valid - or not
>> In any case, the cursor is your friend here : it maps the next() operation on top
of your index.
>> Now, if your Partition is a Btree, it's easier, as the AbstractBTreePartition class
already handles everyting for you. If you don't inherit from this Abstract class, then it's
way more complicated. I'll suggest you have a look at the AbstractBTreePartition to get a
clue about how we process a search over a BTree based partition.
> If I got you right I am to create my own cursor object which retrieves and stores the
array with the items I want to deliver. The cursor then implements the next and get method
in order to send the array one by one. But where is the cursor being called ?

It's called at the top level, as we write the entries one by one, 
waiting for the socket to be ready for a new write. The cursor does not 
have to do anything but to fetch the next entry when requested.
> I tried to base my own partition on the abstract btree partition but what am I supposed
to return from getRootId and getDefaultId and even worse from convertAndInit ?
The RootId is pretty simple : it's 0L.
The DefaultId is always 1L too.

The convertAndInit methid is pretty stupid (and I do think we will get 
rid of it soon). It basically transform an index from a "generic" form 
to a usable form. The rational is that when we initialize the server, we 
don't yet know the kind of backend we will use for a partition, so we 
created a dumb index containing the configuration needed to initialize 
the index in all the case. Then we initialize the backend, and we now 
can create the real index (I mean, the JdbmIndex if we have a 
JdbmPartition, an AvlIndex for AvlPartition, and an AcmeIndex for your 

Now, we have pretty much simplified the process, because we always know 
what kind of index we will have to create, so we create them 
immediately. Sadly, there is still one special index that requires some 
extra effort, this is the RdnIndex.

So, for instance, the JdbmPartition uses this kind of method :

     protected Index<?> convertAndInit( Index<?> index ) throws Exception
         JdbmIndex<?> jdbmIndex;

// first, deal with the RdnIndex which is a bit specific. We will create 
a new one using the configuration put into the generic JdbmIndex

         if ( index.getAttributeId().equals( 
ApacheSchemaConstants.APACHE_RDN_AT_OID ) )
             jdbmIndex = new JdbmRdnIndex();
ApacheSchemaConstants.APACHE_RDN_AT_OID );
             jdbmIndex.setCacheSize( index.getCacheSize() );
             jdbmIndex.setNumDupLimit( JdbmIndex.DEFAULT_DUPLICATE_LIMIT );
             jdbmIndex.setWkDirPath( index.getWkDirPath() );
         else if ( index instanceof JdbmIndex<?> )

// Then for each other index, as they are alreadt plain JdbmIndex, just 
use them and inject the workpath into it

             jdbmIndex = ( JdbmIndex<?> ) index;

             if ( jdbmIndex.getWkDirPath() == null )
                 jdbmIndex.setWkDirPath( partitionPath );

// Should never occur...

             LOG.debug( "Supplied index {} is not a JdbmIndex.  "
                 + "Will create new JdbmIndex using copied configuration 
parameters.", index );
             jdbmIndex = new JdbmIndex( index.getAttributeId(), true );
             jdbmIndex.setCacheSize( index.getCacheSize() );
             jdbmIndex.setNumDupLimit( JdbmIndex.DEFAULT_DUPLICATE_LIMIT );
             jdbmIndex.setWkDirPath( index.getWkDirPath() );

// last thing, we just initialize the index.

         jdbmIndex.init( schemaManager, 
schemaManager.lookupAttributeTypeRegistry( index.getAttributeId() ) );

         return jdbmIndex;

You should basically do the exact same thing.

> Sorry for the beginner level questions but ApacheDS is quite complex and new to me.
meh... It's also complex to me, except that I'm working on it for now 7 
years :/

Btw, we are likely to change many things to the partition in the next 
few months, as we have some problem with concurrency, so keep tuned.

Emmanuel Lécharny

View raw message