On 6/15/11 7:11 PM, Stefan Seelmann wrote:
On Wed, Jun 15, 2011 at 7:01 PM, Emmanuel Lecharny<[email protected]>  wrote:
On 6/15/11 6:17 PM, Pierre-Arnaud Marcelot wrote:
On 15 juin 2011, at 18:01, Emmanuel Lecharny wrote:

On 6/15/11 5:19 PM, Emmanuel Lecharny wrote:
Hi,

sadly, things are not as rosy as I exepected...

I have written a few tests to check if we correctly detect cycles while
processing a search, and at some point, I found that cycle are not an issue
at all. But wait, it's not a good news :/

Let's say you have the following entries :

cn=test
  cn=foo,cn=test
    cn=barAlias,cn=foo,cn=test ->    cn=bar,cn=test
  cn=bar,cn=test
    cn=dohAlias,cn=foo,cn=test ->    cn=doh,cn=test
  cn=doh,cn=test

Logically, doing a SUBTREE search on cn=foo,cn=test, you should get the
following entries :
cn=foo,cn=test
cn=bar,cn=test
cn=doh,cn=test

This is *not* what we get. Currently, you'll have :
cn=foo,cn=test
cn=bar,cn=test

The second alias dereferencing is never done.

Obviously, it helps when it come sto avoid cycle, but this is certainly
not the expected behavior.

Now, if we want to do alias chasing on the server, we will have to
modify the way we handle alias : each one of them will issue a new search,
wth a new cursor.

Hopefully, stacking the aliases to be processed works well with the fact
that we have to stack the aliases for cycle detection. One possible solution
would be to process this stack alias after alias, and if we get back an
alias, we add it in the stack if it's not already present (otherwise, that
means we have a cycle).

In our sample, we will have the following stack :
() : empty stack
(cn=barAlias,cn=foo,cn=test) : first alias met
->    here, we have processed all the entries for the initial search
  [cn=foo,cn=test]
  [cn=bar,cn=test] (the dereferenced alias)
->    now, we get the leftmost element in the stack, and launch a new
search
(<cn=barAlias,cn=foo,cn=test>) : this alias is being processed, thus
the<>
(<cn=barAlias,cn=foo,cn=test>, cn=dohAlias,cn=foo,cn=test) : second
alias met
->    again, all the entries have been processed, we take the next alias
in the stack
  [cn=doh,cn=test]
(<cn=barAlias,cn=foo,cn=test>,<cn=dohAlias,cn=foo,cn=test>) : second
alias met
->    there are no further entries. We are done

Seems to work... Did I miss something ?
Yes, I missed something (thanks Pierre-Arnaud for pointing this out) :

if the alias get dereferenced to an entry below one of the DN already
stored in the stack, or to the descendant of one of the stored DNs, then we
don't need to proceed the search for this alias, as the entries have already
been provided. That will avoid sending duplicate entries to the user.
Yeah, a basic example would be to execute the same search as above but one
level higher, from "cn=test".

In that case, the first search will already give us the three resulting
entries:
- cn=foo,cn=test
- cn=bar,cn=test
- cn=doh,cn=test

And then, in the other searches, triggered when following aliases, some of
these entries would match again. For instance:
- cn=bar,cn=test
- cn=doh,cn=test

We have to make sure that we don't send them multiple times to the user.
Maybe by storing a list of all the DNs we already returned to the user...
But that might be a little bit overkill and suck too much memory.
By storing the root DN of each consecutive search, we will avoid such
problem.

In this case, the initial search is done with cn=foo,cn=test, and this is
the stored DN.
The first search will just returns the cn=foo,cn=test entry and the
cn=barAlias,cn=foo,cn=test entry, which will be dereferenced.

This dereferemced operation consists on doing a search on the aliasedObject
stored in the alias. As it's a new search, we store the base DN
(cn=bar,cn=test), and we launch the new search. Etc...

Should work pretty well without having to store all the entries' DNs
Yes, that stored DNs help for both, (1) avoiding duplicates and (2)
avoiding cycles. But remember, for millions of aliases you have to
store millions of DNs ;-)
Too bad for the millions of aliases :)

Those using too many aliases may :
- either have a very badly designed database
- or managing the CIA employees...

--
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com

Reply via email to