[prev in list] [next in list] [prev in thread] [next in thread] 

List:       ldap
Subject:    Dismaying experiences with LDAP and Berkeley DB
From:       Randy Kunkee <kunkee () pluto ! ops ! NeoSoft ! com>
Date:       1998-06-30 17:37:54
[Download RAW message or body]

I've been running a master and 3 replicas using Umich ldap-3.3 + a
bunch of patches.  The master is on a Solaris 2.5.1 while the replicas
are running on various versions of FreeBSD.  This seems to work pretty
well, but I have noticed a couple of things perhaps worthy of note
and someone else's comments.

First, I have noticed some loss of data if a system crashes (we had
a power failure recently unfortunately the systems involved were not
on a UPS).  My speculation is that Berkeley DB (I'm using 2.3.16 in
compatability mode everywhere) has some of the data cached in memory
and unless slapd is actually shut down, this data doesn't get flushed
in any timely manner.  Has anyone else noticed this?  Suddenly I'll
get missing or changed objects, even ones missing that were added
long ago (and surely synced to disk).

I recently had another problem where I was doing
a bunch of adds, but unfortunately I had forgotten to set the 
'objectclass' attribute in the objects I was adding.  I have schemacheck
turned on, but the master did not catch it.  This killed the replicas
(though this conclusion is somewhat speculative, the coincidence is too
high to ignore).  Yup, suddenly, all of my replicas' slapd processes
died at the same time.  Meanwhile slurpd continue to suck up data
from replog and zero this file, even though the replicas were down.

Which brings me to my last point:  How tolerant is slurpd to
down replicas?  I haven't done much testing of this, so obviously I
need to do more, but I was wondering if anyone had more confidence
in slurpd than I do?

Thanks in advance,

Randy

[prev in list] [next in list] [prev in thread] [next in thread]