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

Re: Dropping slurpd, manage/Relax control

Kurt D. Zeilenga wrote:
At 03:42 AM 12/14/2006, Pierangelo Masarati wrote:
Howard Chu wrote:
One feature that's still needed in some cases (e.g., using
syncrepl to push updates to another slave thru back-ldap) is an
updatedn identity with the privilege to write to unmodifiable
operational attributes. I guess this isn't something the Relax
control is intended to allow. We can keep using the updatedn but
it feels like this is something that should be generalized. E.g.
one might want to have a cluster of servers sending updates to
each other, with a unique identity for each server, and all of
them with privilege to write to operational attributes. I think
the updatedn feature captures the idea ("this identity is a DSA")
but it just needs to be more flexibly configured.
Makes sense.  In principle, slapd should allow any "push"
replication mechanism that complies with its requirements in terms
of operational attributes to work.  Another mechanism, which would
require more "pusher"'s modification, would be to use a control
that behaves like "rekax", which explicitly indicates a DSA is
replicating data.

I had long intended to replace the updatedn mechanism with a control
mechanism... but with the deprecation of slurpd didn't see any good
reason to take the pain that would be caused by such a change.  But
with other good reason for such a control, I would be happy to say
"good riddance" to the updatedn mechanism.

The new mechanism wouldn't behave a whole lot differently... If we were to make this a control I would only define it for Bind operations, thus identifying an entire session to be a server-to-server interaction. It doesn't make sense to me to attach such a control on a per-operation basis. But once we've established that the peer is a server, we can simplify a lot of other checks. (E.g., set a boolean in the Connection structure, instead of having to compare DNs all the time.) This would give us the freedom to do a bit more along the lines of X.500 DSP without the ambiguity that the current chaining/proxyauth/etc. approaches suffer from.

  -- Howard Chu
  Chief Architect, Symas Corp.  http://www.symas.com
  Director, Highland Sun        http://highlandsun.com/hyc
  OpenLDAP Core Team            http://www.openldap.org/project/