[Date Prev][Date Next]
Re: caching magic
Igor Brezac wrote:
On Thu, 17 Jun 2004, Quanah Gibson-Mount wrote:
--On Thursday, June 17, 2004 4:09 PM -0400 Igor Brezac <email@example.com>
This can go in DB_CONFIG:
True, but I rather not see that documented via OpenLDAP if there's a
mechanism provided via slapd.conf for doing it, so that we don't end up
with confused users.
Actually False, as I already pointed out. The only reason there is a
slapd.conf mechanism for shm_key is because it cannot be accomplished
any other way. If there was an existing DB_CONFIG mechanism that was
adequate by itself, then we would not duplicate it in slapd. The main
exception to this policy is "dbnosync" which was intentionally preserved
for backward compatibility with back-ldbm. ("lockdetect" is another
exception which turned into an obvious mistake - the slapd.conf keyword
duplicates the functionality of DB_CONFIG, but it doesn't accept all the
values that are legal for DB_CONFIG. Nor is it particularly useful to
even change this setting. This keyword needs to be removed from
back-bdb...) Otherwise, if you see it in the slapd-bdb manpage, that
means it needed to be there because there's no other way to do it.
I agree that we should not confuse users. As things are right now, you
can specify shm_key in slapd.conf, but you cannot specify how much shared
memory to allocate. This may be confusing to some.
I think it would be cleaner if everything goes to DB_CONFIG or slapd.conf.
It might look cleaner yes, but for reasons I already spelled out in the
FAQ-o-Matic, it is not feasible. It is impossible to do all the
configuration solely via DB_CONFIG, and it is unreliable to do it all
via slapd.conf. The one feature that would make it feasible would be the
ability to call BDB's config parsing routines from slapd, so that we
could place arbitrary BDB directives into slapd.conf, but the BDB
library doesn't export that interface.
-- Howard Chu
Chief Architect, Symas Corp. Director, Highland Sun
Symas: Premier OpenSource Development and Support