On Thu, Jan 6, 2011 at 6:26 PM, Emmanuel Lecharny <elecharny@gmail.com> wrote:
Hi,

today, while tracking some issue in the server, I jumped into some code in the ExceptionInterceptor that I find questionable. Let me explain something we already discussed on IRC.


You know I thought about this to a degree after our conversations. I think the Cursor is setup to be automatically positioned on the first element if one exists. This is why we are checking if an element is available. So calling available() might be the right thing to do and makes sense.

So even inside a block of code inside an interceptor after Cursor creation it might check if the underlying candidate has elements available before wrapping it or even adding its own filter.
 
When we do a search, we go through the chain of interceptors, down to the backend. In the backend, we process the request, and construct a Cursor based on the search filter and the scope.

This cursor has not yet fetch any entry.


Cursors returned are in fact designed to prefetch. Under the hood they know if they have anything available so they can position themselves right before the first element they will return.
 
Then we go back the chain up to the SearchHandler. In the process, we construct a list of filters to apply to each entry we will fetch. Those filters simply select good candidates, and reject bad ones. They also eliminate not requested attributes or values.


Exactly.
 
When we go up the chain, and process the ExceptionInterceptor, the following code is executed :
   public EntryFilteringCursor search( NextInterceptor nextInterceptor, SearchOperationContext searchContext )
   {
       EntryFilteringCursor cursor = nextInterceptor.search( searchContext );

       // Check that if the cursor is empty, it's not because the DN is invalid.
       if ( !cursor.available() && !searchContext.getDn().isEmpty() && !subschemSubentryDn.equals( searchContext.getDn()) )
       {
           // We just check that the entry exists only if we didn't found any entry
           assertHasEntry( searchContext, "Attempt to search under non-existant entry:", searchContext.getDn() );
       }

       return cursor;
   }

The assertHasEntry method is :

   private void assertHasEntry( OperationContext opContext, String msg, DN dn ) throws LdapException
   {
       if ( !opContext.hasEntry( dn, ByPassConstants.HAS_ENTRY_BYPASS ) )
       {
           throw new LdapNoSuchObjectException( msg + dn.getName() );
       }
   }

And now, the problem : the cursor.available() call *always* return false, thus we most of the case do a lookup, which is costly. Every single search we do will always fetch N+1 entry : The N entry to return, plus the first entry a second time.


Hmmm why does it return the first entry a second time?
 
This is useless.


Yes it should not do this.
 
The reason why this check is being done is that we need to send a LDAP_NO_SUCH_OBJECT result if there is no entry returned.

We discussed some options we have. Currently, we have 2 :
- make the cursor return something different from false if there is an entry, even if we don't prefetch the entry.
- postpone this test and handle the case in the SearchHandler

Option #1 implies a huge refactoring of the cursor API. Not sure we want to do this
Option #2 is probably better. We just have to check that it's valid to check for the absence of an entry if the candidate have been discarded by some upper filter (for instance, if we rejected some entry because the ACIs mandated it). In this case, should we send back a LDAP_NO_SUCH_OBJECT?

I will experiment with Option #2, re-read the RFC, and update you.

What worries me is that this worked just fine for years. It should not be failing us now: there has to be something wrong here. I can take a look at the code and maybe we can discuss this more together on IRC.

--
Alex Karasulu
My Blog :: http://www.jroller.com/akarasulu/
Apache Directory Server :: http://directory.apache.org
Apache MINA :: http://mina.apache.org
To set up a meeting with me: http://tungle.me/AlexKarasulu