RE: back-bdb DB_RECOVER and soft restart

["Howard Chu" <hyc@highlandsun.com>]

> -----Original Message-----
> From: owner-openldap-devel@OpenLDAP.org
> [mailto:owner-openldap-devel@OpenLDAP.org]On Behalf Of Lars Uffmann

> On Wed, 2003-09-17 at 18:39, Michael Ströder wrote:
> > Luke Howard wrote:
> >  >>That's what back-bdb used to do, in its infancy. It
> turned out to cause
> >  >>slapd startup to be very slow.
> >  >
> >  > Hmm, I thought Sun ONE did this. Does anyone know what
> they do? They also
> >  > use Sleepycat AFAIK...
> >
> > ..and IMHO Sun ONE DS *is* slow on startup...
>
> The time for recovery depends on how often you checkpoint
> your database
> during normal operation. Often you will find hundreds od megabytes of
> BDB logfiles, specially with netscape/iPlanet. Never seen a SunONE
> server, though.

This is really a side issue. The serious problem is that you still need an
interlock scheme to prevent multiple slap executables from recovering the
environment at the same time. In the old back-bdb, bdb_init always did a
recovery, so if you ran slapcat while slapd was running you would destroy the
database. Since there is a great advantage to being able to use slapcat and
slapd simultaneously, we still need an interlock scheme. We could special
case slapcat and say that it never does DB_RECOVER, but it would still be
useful for slapadd/slapindex to invoke recovery if needed, assuming it would
not interfere with a running slapd.

All of the reasons were already covered in this email thread last year.

  -- Howard Chu
  Chief Architect, Symas Corp.       Director, Highland Sun
  http://www.symas.com               http://highlandsun.com/hyc
  Symas: Premier OpenSource Development and Support