Bug 55706 – openldap: default slapd.conf creates a directory that is nowhere near crash-safe

• Product: Gentoo Linux
• Status: RESOLVED FIXED

[Description Jeremy Kitchen 2004-06-30 12:09:06 UTC]

the default slapd.conf uses bdb but defines no checkpoints, therefore, if your system crashes, all of your data in your directory (at least since the last shutdown) is gone.

Reproducible: Always
Steps to Reproduce:
1. start slapd with deafault slapd.conf
2. add a ton of stuff to it
3. power off your machine
4. power your machine back on
5. cry.

Actual Results:  
all of the data since the last 'clean' shutdown is gone.  This is a bad thing.

Expected Results:  
if nothing else, it should have had *SOME* of the updates you made.

I would suggest adding the following line (or at least commenting it and
describing its function) to the default slapd.conf:
checkpoint      20      30

that will make slapd have a 'checkpoint' every 20 kilobytes written to the
database, or every 30 minutes, whichever comes first.

You might also add a couple of other commented entries for larger systems that
might actually be doing 20 kilobytes all the time, or 30 minutes is too fast.

http://sapiens.wustl.edu/~sysmain/info/openldap/openldap_configure_bdb.html has
some more description about some sane default settings.
http://www.openldap.org/faq/data/cache/893.html is some openldap.org
documentation about bdb

[Comment 1 Jeremy Kitchen 2004-07-21 08:35:57 UTC]

this was reported almost 2 months ago.. I figured it would be a "oh yea, oops" and be in.. but I see nothing :)

Shall I provide a patch to the slapd.conf file?

[Comment 2 Robin Johnson 2004-07-21 08:59:05 UTC]

patches greatly accepted.
I'm basically the only person handling openldap presently, and I've got a lot of other things on my plate too.

For the most part, at the moment that means unless it breaks at work (where I've got gentoo deployed to ~20 servers), or I need something new at work/home, It's minor version bumps and the such only.

[Comment 3 Jeremy Kitchen 2004-07-21 13:00:36 UTC]

Created attachment 35911 [details, diff]
patch to default slapd.conf to make the database more reliable.

you must not ever have power outages or unexpected shutdowns then, as it does
break then :)

unified diff enclosed.	Not sure if I would revbump or whatever, but I would
think that you'd want this in the config file.

I also added that bind_v2 be allowed, since I believe that's what squirrelmail
uses with php and ldap.

Enjoy.

[Comment 4 Robin Johnson 2004-07-21 13:20:00 UTC]

my LDAP server has a redudnant power supply, the hardware is good and solid (after my own burn-in testing), makes use of 3ware RAID1, and i've got a big friggin UPS for the server room :-).

i'll see about putting it in sometime tonight maybe.

[Comment 5 Robin Johnson 2004-08-22 13:24:43 UTC]

the checkpoint statement is now used in cvs.
I'm NOT putting the 'allow bind_v2' in place. If you have an old application that needs the backwards support of v2 binding, then you should enable it only for your configuration that needs it.