+1
The dn2subentry cache sounds like a valid optimization and making it a
top level accesible cache also sounds right.
Additionally i checked out depa on DnNode and it seems out of place in
shared. Had the thought that it might be best close to where it is
being used. WDYT?
Sent from my iPhone
On Jan 2, 2011, at 11:36 AM, Emmanuel Lecharny <[email protected]>
wrote:
Hi,
we currently use a UUID -> Subentry cache, used for the most common
operations like retrieving a subentry when processing an entry (as
we store the subentry's UUID in an entry selected by the subentry
SubtreeSpecification, to avoid some costly operations on disk
following a ModDN operation).
That's ok, and enough, except that when we add an entry, we have to
check that it's not added under a subentry (no entries are allowed
under a subentry). How do we do that ? If we don't have a DN ->
Subentry cache, that means we have to either read the added entry's
parent, and check if it's not a subentry. Costly, and not
convenient. There are also many other operations for which we would
like to retrieve a subentry directly from its DN, not its UUID, like
when we delete a subentry (we just get its DN, so we have to do
either a lookup to get it back in order to retrieve its reference in
the UUID -> Subentry cache, or fetch the AP and look into all the
associated Subentries to get back its UUID. Painful).
So I will create a second cache, a DN -> Subentry that will help to
manipulate the subentries.
I have created two JIRAs (https://issues.apache.org/jira/browse/DIRSHARED-72 and https://issues.apache.org/jira/browse/DIRSHARED-73
) which are closely related to the DnNode data structure in order to
be able to easily handle a ModDN operation, as such an operation is
not supported by the data structure (and as a side effect, if we do
a ModDN on a referral or on its parents, the referral won't be
cached anymore). The other JIRA is to make this structure thread safe.
One more requirement : at some point, we may want to make the
subentryCache a unique data structure handling the two types of
cache internally. That will probably be the next step.
--
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com
