Re: dropping back-ldbm (was: commit: ldap/servers/slapd connection.c)

[Hallvard B Furuseth <h.b.furuseth@usit.uio.no>]

Quanah Gibson-Mount writes:
><h.b.furuseth@usit.uio.no> wrote:
>>Pierangelo Masarati writes:
>>>Quanah Gibson-Mount writes:
>>>
>>>> Since LDBM is kind of deprecated in favor of BDB, etc, at what point
>>>> does OL simply drop LDBM?  Would it in particular simplify things?
>>>
>>> I wouldn't consider this an option yet.
>>
>> I wouldn't consider it an option until BDB is as easy to configure as
>> LDBM is, if that ever happens.
>>
>> Maybe it would be possible to provide a BDB configuration which is easy
>> to tweak because it disables features which makes BDB superior to LDBM
>> but need to be configured in some way?
>
> I think this is really bad reasoning.  If everyone followed that line of
> thought, most software would be even more bloated than it is now.

Huh?  Improving some software doesn't usually mean that one must make it
much harder to configure in order to get as good performance as before.
I was suggesting a way to _avoid_ the bloat you refer to.

> The fact that you've had issues getting BDB to work in your
> environment is unfortunate, but it is well documented at this point as
> to both how to build the BDB libraries, and how to configure the BDB
> environment.

Yet how to configure BDB remains a FAQ.

And our "issues" were not related to that, but to BDB bugs or
limitations.

> (...)
>> Maybe we've just have had bad timing, but so far BDB has had issues
>> every time we have had time to look at switching to BDB at our site.
>> E.g. one time I think we needed to apply some patch to Sleepycat in
>> order to turn off transaction logging in slapadd.
>>
>> Anyway, LDBM works for us - a read-only database which is built
>> regularly from scratch - so upgrading is a low priority.
>
> And people in that situation can do what they've done for years --
> Stick to Windows 95, Windows 98, OpenLDAP 2.0, etc, etc.

...and keep all the other problems that have been fixed since those
releases, and maybe downgrade a bunch of older libraries that the old
release expects.  No thanks.

> If they can't find the incentive to upgrade, then don't.

We've found the incentive to upgrade several times.

If LDBM had been removed, we might also have found the incentive to work
around the BDB bugs we encountered, or to live with the ways BDB was a
poorer solution than LDBM at those times.

In any case, I guess my main point is that BDB still seems rather more
fragile than LDBM.

-- 
Hallvard