[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Transactions in LDAP



At 11:47 PM 9/2/97 -0400, Christopher W Apple wrote:
>>The consensus was to keep it simple (no requirement for 2-phase commit).  
>>The proposal on the table is to add an ldap extension (could be an ldap
>>control or ldap protocol extension or something else) for begin/end/abort 
>>transaction.  Three scenarios were identified and these will be used to
>>evaluate the solution:
>>  client-server:  treat several ldap operations as a transaction (single 
>>			server)
>>  client-server: above case extended to multi-server for transactional 
>>			support in the ldap in support of intelligent networks 
>>			and routers
>
>Since there is already talk of having multi-master replication policies,
>does it make sense to treat this case as a particular configuration of
>using client-server LDAP transactions against multiple-masters of information
>in conjunction with such a replication policy?

yes...

>
>>  server-server: replication cases
>
>Is the idea here to have one server act as a client and the other as
>the true server within the context of one transaction? And that the roles
>might be reversed on the next transaction?

Yes.  The case is the following:  A set of entries/attributes are updated
in the directory server and that modification was handled as a transaction.
If that server has replicas, then those updates need to be handled as a
transaction during the replication process.  Of course, this can get
very complicated quickly when the transaction spanned multiple servers, for
example.  We'll just need to work through these cases and probably make
some simplifying assumptions where reasonable.

>
>>
>>Cisco volunteered to co-author a draft with IBM in this ldap transactions
>>area. 
>
>Is this going to be a requirements document, or a extension/control
>specification straight out of the BOF?

Thanks for asking - I knew someone would :-).  Since this topic can get
very complex quickly, I believe there should be a set of requirements
written so what's written in the specification will have context and
justification - it could even be part of the extension/control specification.

>
>Chris Apple
>Internet Directory Group
>AT&T Laboratories
>capple@master.control.att.com
>+1 908 582 2409
>